I typed it with tongue firmly in cheek. TFLOPS ultimately doesn't mean everything, but it's probably a safe bet that there will be instances of Vega FE beating out Pascal in some tasks. Unfortunately for consumers (depending upon what sort of consumer you are), this will probably be priced closer to Titan XP/Xp than the 1080Ti if they are targeting compute folks over gamers.
Or maybe they'll pull a Ryzen and undercut NVIDIA by 50%. haha
They won't, it's stated all over the slides that it's aimed at professionals and it won't come cheap. The consumer focused vega card will come later. The PRO card are already priced around the $999 USD so why would they sell a more powerful card for a lower price? It doesn't make any sense.
Yes, and we don't yet know how Vega FLOPS compare. 11 Pascal FLOPS is probably usually faster than 13 Fury FLOPS would be. AMD is promising better architectural efficiency but that remains to be seen.
You are wrong.. the 1060 its 5 tflops aproximately. You're nota taking un consideration the auto oc feature of Pascal cards, the 1060 auto overclocks nearly 2000mhz si its not 3.9 tflops
Where do you get this from? In what games is the GTX 1060 averaging 2000 MHz without overclocking?
Regardless, it doesn't matter. You are agreeing that AMD's listed TFLOPs are worth less than NVIDIA's listed TFLOPs, which was the point of the thread. Whether it's true because NVIDIA's cards are able to consistently clock above their listed frequencies or because NVIDIA's architecture spends less time waiting around doesn't matter. (As long as NVIDIA's cards stay within their thermal envelopes, which they seem to do better than AMD's, in fact.)
Gaming performance isn't the best measure when talking about professional graphics cards. It's silly to say "X TFLOPS on card A equals Y TFLOPS on card B because of gaming benchmarks" when discussing professional cards that are used for different software.
Don't know who you're responding to, I compared RX 480 and 1060 GT which neither being part of pro segment and I specifically mentioned we need a better picture of Vega's architecture.
"Gaming performance isn't the best measure when talking about professional graphics cards. It's silly to say "X TFLOPS on card A equals Y TFLOPS on card B because of gaming benchmarks" when discussing professional cards that are used for different software."
I think vladx's use of gaming performance to demonstrate that TFLOPs cannot be used to directly compare performance between two different architectures is perfectly fine. The person he replied to performed just such a comparison. And although vladx did say that Vega has to improve for AMD to catch up to NVIDIA, and perhaps that's not true in professional applications (I don't really know), the original post should not be assumed to be about professional performance. The OP simply said "I really want Vega to succeed", not "I really want this Vega Frontier Edition to succeed". I'm guessing he meant Vega in all its incarnations.
When it comes to actual FLOPs ... FLOPS are FLOPS, they are the same. When it comes to how FLOPS translate to gaming performance .. that's another discussion, yes.
No, and no. A TFLOPS means nothing at all. Because and operation is not like another. We are used to FMA OPS to calculate theoretical peak TFLOPS capacities, but those do not really mean nothing. How many FMA do you use in an algorithm? How long can you sustain many FMA one after the other before bandwidth do fed them becomes a bottleneck? A TFLOPS is not a precise unit of measure. It's just a "standard" way to classify computing devices. It may be that my device can't do FMA in a single cycle, so having half of your teorethica computational peak, but I may go faster than you. So again a TFLOPS means nothing. What it really counts is the efficiency at which the architecture execute complex algorithms (and that is before you tale into account other resources like ROPS or TMUs or geometric management for a 3D pipeline). By what we have seen today, nvidia is way more efficient than AMD in both area/perfomance and w/performance.
Nvidia is using a bigger die to get similar numbers. Not sure where the area/performance you speak of stems from. AMD appears to be cheaper in a $/performance and area/performance comparison with Nvidia's latest and greatest design. I'm sure someone will correct me if I'm wrong, but the new Nvidia die is at the TSMC reticle limit.
If you're talking about Volta, you're right, it is a much bigger die. What it is NOT though, is a competitor for this card. This Vega boasts 13TFlops, Volta boasts 120TFlops per chip for DNN (FP16 matrix multiply/add). It will also sell for a much higher price (Volta mezzanine cards are around $16k each). This is like comparing a Formula1 car and a Lexus. Not the same usage or market. This is supposed to compete with Titan XP. In that comparison, AMD uses much more expensive RAM (HBM2 vs GDDR5X) and die size is so far undisclosed, but past chips size and power consumption have not been in AMD's favor.
Yup. The leaks have an 8GB HB2, full die Vega gaming card at $599msrp, but I don't think they can stick to that number and be successful. Unfortunately for AMD, the 1080 Ti / GP102 will be a year old at that point, on a more mature process with assumedly better yields making pricecuts less painful for Nvidia then they might otherwise be.
At least we have some competition to talk about on the high end again!
I'm constantly surprised that the gaming market is still into glass panels and Max Power LEDs on their machines. Even as a teenager I'd have thought they were horrific. The fact that people care what the internals of their computer looks like baffles me.
Unfortunately, it's increasingly difficult to get components with desired specs that come without the LED circus, you actually have to actively search for it when you head in high end desktop HW.
I don’t know a single person who likes those. It’s like a whole industry is running off a coked-up marketer’s drug-induced "This is what kids like!" idea and everyone has to suffer for it.
I just bought a solid case with no silly see-through panels on my most recent build, which was my first in a few years. I'm still weirded out by the trend towards showing off the guts of your build. I'm still used to putting the components into the case and never seeing them again until upgrade. I can't imagine choosing the better looking GPU over the most powerful GPU in a given price range.
Not to mention highend gaming accesories. Thankfully there seems to be a trend towards releasing more aesthetically pleasing designs without too many silly LED's and no childish buzzwords or demon/dragon eye atrocities.
Have we found one? The mythical consumer that actually WANTS the 'gam8r' aesthetic with blinged out pointy shiny plastic bits and RGB lighting out the wazoo?
Did you mean "gam3r"? Anyway, my gfx board has RGB LEDs, but that's coincidental; an Asus Strix happened to be the cheapest thing available, and the software can make the lighting reflect temps so it's actually kinda useful.
Never saw the point of good looking internal components. Once I put a graphics card in my computer, I don't see it again except for when I need to change something inside the case. It sits there out of the view under my desk.
If the 1080 Ti is more expensive and performs worse I'd buy it over this Gold Edition thing because that's just how much of a turn off a water cooled GPU is for me at this point. It was a disastrous move to make the Fury X water-cooled only. I hope they see sense and open this up to AIBs this time.
Because of the implication that the blue edition is a cut down, less lower performing variant. People made this argument with Fiji, why not get a Fury if you don't like watercooling? Because you could have both with the 980 Ti. In fact there are Nvidia SKUs that offer watercooling as an option. Having it as an AIB option caters to two different audiences, this "watercooling or nothing" approach does not.
It has exactly the same end result. The only thing that is different is that AMD are offering a water-cooled version themselves.
The water-cooled Nvidia cards will be able to perform better than their non-water cooled ones. Does that for some reason put you off them? Apparently, not, so why not the same for AMD?
You've made up a message in your own mind that was not made by AMD.
Didn't the Fury X HAVE to be water-cooled because of its TDP? If it had been air-cooled it would either have been very noisy, or more likely constantly throttling because that's just too much heat to blow away from a box that size.
Or it had to run at a lower clock just like the Fury without X. But that card had not the bars slightly longer than the 980Ti in 2 tests of 20. So they could not sell it that way, and they decide to go for the water cooling solution had have a little extra OC just to make the failure GPU shine a bit. They know it was a fail. But they made all they could to make it appear as good ads the concurrent solution. My fear is that they are doing the same with Vega. It would mean they have nothing really good to offer.
No. Horsepoop. AMD could have made an aircooled Fury X... it would have run hotter and been louder. There are aircooled overclocked 390X cards. An aircooled Fury X actually would have been fairly similar to the non-X Fury in terms of thermals and noise, they even carry the same TDP. They could have kept noise under control with large fans but it still would have run hotter and the watercooler included at the price tag made it a decent deal.
I would tend to agree that AMD's choice to go workstation first is most likely being led by supply issues with HBM2 I would disagree that it is anything to do with performance per dollar, in fact I would presume the opposite as the pro market is already dominated by nVidia and AMD would be competing directly with them to gain market share, they can't really accomplish this with poor performing cards in comparison with nVidia unless they plan to add a value sector to the pro market which is unlikely due to this being a very profitable sector and if there is one thing AMD needs more than anything right now it's profits and lots of them.
I imagine desktop VEGA to compete in performance with the 1070, 1080 and 1080Ti while costing a bit less but the issue I have there is that of RAM or Cache size on the VEGA parts. I imagine that VEGA will launch in 4GB and 8GB varieties with HBM2 and while 8GB is plenty of RAM, comparing it to the 11GB in the 1080Ti will affect people's decision on which card to get, I know far too many people who would buy a worse GPU because it had more RAM, 'cause more is better right? LMAO!
I imagine this is part of the reason why AMD is showing how VEGA is able to run well with 2GB RAM with the High Speed Cache Controller enabled as it will help alleviate fears that 8GB isn't enough RAM or Cache. I also get why they call it cache, It reminds me of the old Pentium 2 Single Edge Contact Cartridge or SECC design which was inserted into Slot-1 Motherboards. The reason the CPU had the Slot form factor was because the cache chips were on the daughter board around the CPU rather than in later designs built right into the CPU itself. This is a very similar design strategy and should offer comparable speed benefits. I just look towards the next evolution of this process which should be the cache being baked into the GPU itself, there are issues with that as HBM memory is multi-layered and I imagine this would present significant design issues to incorporate this into the core of a GPU but hey, I'm a dreamer :-)
Actually the reason they went with slot-1 was to prevent AMD from sharing a socket with them. Super socket 7 was very good for AMD , external cache was the way it was done up to that point. The K6-2 and K6-3 were clocking at 500mhz plus on socket 7. The k6-3 was a beast with that extra cache but most people only saw it wasn't clocking as high and didnt realize the performance it offered.
It'll be over a year since the P100 is released and only a month after the VEGA FE that the V100 is released. It should be compared against the more recent one.
"Nobody outside the DOE will be buying V100s this year."
According to NVIDIA's presentation that isn't true. They say that the DGX-1V is supposed to start shipping in Q3 and that third party servers will ship in Q4. Hyperscalers will probably start to get them at the same time DGX-1V shipments start or even earlier.
And they also forgot to mention that nVidia just let Volta out of the bag for, yup, you guessed it, the Pro market. Volta with it's 16GB of HBM2 memory with 896Gbps bandwidth, 5120 CUDA cores and 640 TENSOR Cores makes even VEGA look old tech in comparison (For the Pro Machine Learning market). I just hope for AMD's sake that either VEGA can compete with Volta, Volta is targeting a different sector of the market than VEGA or that AMD is willing to take a bigger hit to the margins with VEGA to stay competitive.
Volta on the desktop I imagine will...
Not have the TENSOR cores as they seem to be optimized for machine learning tasks (that's my understanding of them anyways)
Not come with 16GB of HBM2, more likely 8GB or 12GB+ of GDDR5X or 6. Premium models may offer 16GB HBM2 like the Titan Xv but the 2080 Ti will most likely come with 12GB+ GDDR5X or 6 depending on when it launches.
Not offer 896Gbps bandwidth due to only having 8GB HBM2 it will offer 480~Gbps like the VEGA counterpart.
This makes desktop VEGA and VOLTA potentially much closer in performance than the Pro versions meaning that AMD should be able to compete for some of the desktop market.
Currently Volta costs $18000 USD for a single board, this gives AMD a fair amount of wiggle room on competing price :-) lol
However I must say I am impressed with what nVidia has shown with Volta, even though I would never need such a product, I am impressed with the advancements they have made and the sheer size of the GPU is incredible, however it could be telling that nVidia NEEDED all that die space to get the performance boost, could Volta be less efficient than hoped? I don't know, it's still impressive.
As a rule of thumb, we don't publish competitive performance slides. While I doubt AMD is lying, it's not something we can verify at this time, and in these cases vendors usually cherry-pick at a minimum.
You seem to be jumping to the conclusion that this is the first Vega to launch and consumer comes later but don't quite think that AMD has stated that. Yesterday's event wasn't the right place to launch consumer products but that doesn't mean there won't be consumer Vega before this one. The 16GB of HBM makes it even less likely for this one to be the first to ship.
You also seem to miss that this is not Vega Instinct , it's a Pro card but seems that it's a kind of high volume sampling SKU so AMD is either not quite ready on the software side or can't make all that many yet , for high volume deployment in data centers.
BTW AMD lists 25TFLOPS FP16 not 26, so base clocks and they likely have a reason to do so.
As for all the preaching about prices and higher margins, that's false and hugely misleading. Higher margins and ASPs sure but much lower volumes and substantial investments in both hardware and software. The high end gaming is plenty profitable, if the product is good enough to sell because sites like this one are promoting ridiculously priced GPUs instead of urging users to never buy such poor value products. It's Trumpian that hardware sites are not making fun of such terrible value products instead of trying really hard to sell them. The press today serves the corporations not the consumer- wasn't like this just a few years ago. You are actively working hard to make people dumber and waste their money every time you don't just laugh at a 500$ GPU or a 700$ phone. Anyway, this is not 10 years ago when the Pro cards were so much more profitable and it's because the press is utterly corrupted and made the consumer much much stupider. Today high end gaming funds the pro and server cards and only in the next few years ,when the deep learning market grows to a sufficient size, it will be able to fully fund the product development, but only for the market leaders.
There was no reason for AMD to not launch a consumer card yesterday. It wasn't a developer's conference. It was an investor's day event. Investors don't care if a product is a consumer product or a professional product. They just want profitable products, and they'd like to hear about as much as possible.
I don't remember the exact quote, but from what I remember Raja Koduri said things like "There's been a lot of rumors. Everybody's asking where's Vega? When will it appear? Well today we can say the first Vega will be released at the end of June, called Vega Frontier." His quote did make it sound like no consumer Vega would be coming before Vega Frontier. Whether a consumer Vega will be appearing alongside Vega Frontier is less clear, but the situation suggests to me that it will not.
"if gaming is your primary reason for buying a GPU, I’d suggest waiting just a little while longer for the lower-priced, gaming-optimized Radeon RX Vega graphics card."
They outright stated consumer Vega is not coming first. And then put it in print as well.
Noone pointed out how they are comparing FP32 deepbench with P100 when P100 has 1:2 FP64 performance (5 TFLOPS of FP64) and this Vega card only has 1:16 (800 GFLOPS of FP64)?
I doubt DeepBench is concerned with FP64 data types at all.
The thing with those DeepBench numbers is AMD just put up some run times. They didn't specify exactly what the parameters were for the runs, or how they got one single "DeepBench" run time number when from what I see DeepBench consists of four separate types of tests.
I really hope Vega has a higher gaming performance/flops ratio. I'm tired of nvidia's prices. (Unless AMD follows suit, which they might. They are not in a position to start a price war)
Any Vega card that's expecting to make a profit will be priced higher than the 1080Ti. These things are BOM beasts, that HBM memory and interposer ain't cheap. There's a reason this thing is aimed at Pros, because consumers aren't gonna drop $1K on a Vega, it'd be DOA.
Sure AMD might sell a very limited batch at $600 to compete with 1080Ti, and claim they've got the perf/$ crown, but every card they sell will be $100s lost from their bottom line.
That's why they are trying to sell less Vega in the consumer market as possible. Just few to show in benchmark and show they are still alive (using a beefy large solution against a smaller cheaper one) and then trail until 7nm where maybe they hope to have some advantage over nvidia which is simply bulldozering them under all points of view.
They claimed Vega will be out in H1 2017. They are not ready with a product that can compete with nvidia at the same price level. This chip probably is like a Fiji against GM200: it costs too much and returns nothing. By the way, they have to maintain the promise to the investor for the GPU release, so they came out with this ridiculous plan to show (you certainly won't find the card at end of June on the market, not probably at end of July) the card at end of June and delay the presentation of the failing consumer card as much as possible.
You may in fact have the correct conclusion, but based on what premise. The 1080 and 1080Ti are gaming GPUs and have been evaluated in gaming applications. The VEGA FE presented here is a professional card that should be evaluated in professional applications (which are not comparable to gaming applications). There is no gaming evaluation presented in the article. There isn't even a professional evaluation. You just pronounced yourself correct without presenting any supporting evidence. Unfortunately, even if your conclusion ends up being correct, you statement is still a fallacy.
Obviously, I was extrapolating from this pro card to obtain the consumer version's performance which I'm betting will have a few units disabled compared to this one.
I suppose it's not as obvious to me as to you. The article doesn't present any benchmarks, much less gaming benchmarks. So I ask again, based on what premise?
Theoretical (Max) single / double / half precision TFLOPS perhaps. Historically these numbers haven't been very useful for comparing gaming performance between vendors.
Memory bandwidth - suffers the same problem as above.
Pixel fill rate - only a small part of the story that may be more or less useful depending on application.
Outside benchmarks - please share. I'd love to see some actual gaming (or gaming related) performance number. The validity of leaked or cherry picked benchmarks is perhaps questionable, but by necessity incorrect. Don't let that keep you from sharing.
I apologize if I sound obstinate to you. The truth is, I actually like your conclusion and I am inclined to agree with it based on a combination of specifications, theoretical performance numbers, historical tendencies, market forces, and "gut feeling". However, I have yet to see anywhere near enough evidence to conclude more than a very rough wag on this one. Known changes to the architecture make historical tendencies a fuzzy approximation at best. We don't yet know how much (or little) they will affect the gaming performance.
The software stack slide is telling; CUDA owns the acceleration space and OpenCL isn't very popular. AMD doesn't have an answer here and it's too late anyway because CUDA is established and it works. Nvidia are being bastards and not supporting OpenCL 2 either, locking the market in.
It's not too late, but it's going to take a lot of hard work, determination, and resources. Koduri's claim that 10 or 20 engineers working for a few weeks on each framework is all that's necessary is not auspicious. Developers need assurance that AMD are going to actively support the ecosystem, something that they haven't been doing up to this point. Those number of engineers for that amount of time probably is what it took them to be able to run one chosen benchmark well that matched up particularly well with their chosen architecture (my guess is that the high bandwidth cache is a prime candidate for an advantage to focus on). As far as I know, for general usage, the BLAS libraries in GPUOpen are significantly slower than NVIDIA's cuBLAS.
There's a lot more to support in GPU computing than just machine learning, as well. If they only focus on machine learning they will lose a lot of opportunities from companies that want to do simulations and analytics along with machine learning, which is probably the majority of them. Each application has its own issues, and the people in those market segments are mostly not machine learning experts. AMD has 3,000 employees in its Radeon Technologies Group. NVIDIA has 10,000 employees, and they don't have thousands of them sitting around doing nothing.
As far as OpenCL, even when NVIDIA's OpenCL support was more current, CUDA had the advantage because NVIDIA actively supported the space with high performance libraries. If NVIDIA controls both the architecture and the programming model they are able to bring features to market much faster and more efficiently, which is pretty important with the pace of innovation that's going on right now. My guess is that opening CUDA would probably be a more beneficial action for the community than supporting OpenCL at the moment, unless opening CUDA meant losing control of it.
Yes, it would be more beneficial to the community to open CUDA up to other vendors. However, I think it is about as likely to happen as opening up PhysX or G-Sync. nVidia doesn't exactly have a reputation for opening up proprietary tech.
Well, they are open sourcing their upcoming deep learning accelerator ASIC. They recently open sourced a lot of GameWorks, I think. They will open source things when it's beneficial to them. That's the same thing that can be said for AMD or most tech companies. Considering their output with these initiatives and market position, AMD open sourcing GPUOpen and FreeSync was beneficial to them.
NVIDIA has a history of not waiting around for committees and forging on their own the things they need/want, such as with CUDA, NVLink, and G-Sync. They are trying to build high-value platforms. They spend the money and take the risks in doing that and their motivation is profitability.
I also don't expect them to open up CUDA at the moment.
I'd forgotten NVLink. I'm not seeing any answer to that at all from AMD and NVLink 2 is fundamental to accelerated HPCs which are vaguely programmable.
Opencapi (phenomenal latency). Given the huge number of lanes that Naples has, it would be a good target for opencapi. There's also the in-progress genz fabric, but that's not close to being standardized, and is more of an interconnect between nodes, I suppose.
OpenCAPI is managed by the host processor (CPU). From the way I've seen it described, it is not meant for accelerators to use to link to each other without going through a host processor that supports OpenCAPI. AMD presumably might support OpenCAPI on their CPUs, but they can't count on their GPUs being used with their CPUs. Perhaps CCIX plans to allow accelerator to accelerator interconnect, I'm not sure.
Regardless, both OpenCAPI and CCIX are not available yet, and certainly not enabled on the upcoming Vega products.
Capi is definitely managed by the processor. Opencapi is supposed to be a "ground up" rethink of a high-speed, low latency interconnect for various on-node resources. Ibm appears to be calling theirs "Blue Link" (or something similar), and that's supposed to be released this year. I would be astonished if amd hasn't done some work towards supporting one of these standards on vega, even if not on this exact chip. Regardless, these are the options for amd, and since they are "standards" the hope is that others will begin buying into the spec as well. If opencapi remains a star network then amd can offer a competitive, even if not identical, solution instead of ceding the market to Nvidia.
@Yojimbo: "Well, they are open sourcing their upcoming deep learning accelerator ASIC."
Good Point.
@Yojimbo: "They recently open sourced a lot of GameWorks, I think."
I didn't know about that, but I'm not sure how much this one matters as gameworks products already run on competitor's GPUs and we already know that gameworkst is optimized for nVidia GPUs (Why wouldn't it be?).
@Yojimbo: "They will open source things when it's beneficial to them. That's the same thing that can be said for AMD or most tech companies."
I generally agree, but I would be remiss if I did not point out HyperTransport as an example of an AMD open standard initiative that predated it's proprietary Intel counterpart (QPI). It was also introduced in 2001 with their Athlon64 architecture, a time when AMD was doing well due to the recent success of the Athlon processors. It doesn't happen as often as I'd like, but sometimes tech companies will prioritize long term interoperability and viability over short term gain.
@Yojimbo: "Considering their output with these initiatives and market position, AMD open sourcing GPUOpen and FreeSync was beneficial to them."
I don't disagree. Historically, AMD has found themselves in this position far more often than nVidia.
@Yojimbo: "NVIDIA has a history of not waiting around for committees and forging on their own the things they need/want, such as with CUDA, NVLink, and G-Sync.
They don't need to wait on a committee to open up a standard. AMD forged ahead with mantle until such a point as the industry was ready to work the issue. They also don't necessarily need to give up control of their standards to allow other vendors to use them. It would even be reasonable to demand a licensing fee for access to some of their technologies.
@Yojimbo: "They are trying to build high-value platforms. They spend the money and take the risks in doing that and their motivation is profitability."
Keeping proprietary tech proprietary is their right. When they are are dominating a one or two player market, the risk is low, the returns are high, and the decision makes sense. If there were several major players or they weren't the dominant player, this would be far more risky and interoperability would be a bigger concern. Given the current market, I expect they'll keep their proprietary tech in house.
"But nVidia also has more markets to serve, their have their own CPUs, mobile, automotive..."
It's AMD that have their own CPUs, not NVIDIA. But I didn't include AMD's CPU business in their employee count. NVIDIA is rather focused. Virtually all their products are based on their GPU architecture. They do develop their own CPU core based on the ARM-v8 instruction set, but it is geared entirely for one purpose, to be the CPU for their self-driving car SoC. Mobile and automotive are close to the same market for NVIDIA. They probably put 100 employees or something on the Switch. NVIDIA have gaming GPUs, professional GPUs, datacenter, and Tegra (mobile/automotive). AMD are trying to go after all that same stuff other than automotive. Whatever number of employees NVIDIA use on the Switch, AMD probably use more on the XboX/PS4 projects.
NVIDIA have 3 times the numbers of employees as RTG. NVIDIA don't have 2/3 of their employees working on mobile and automotive. Most likely they have well under 1/3. Let's be conservative and place it at 1/3. That means NVIDIA have roughly twice the number of employees working on the other segments, about 3,000 people. No, they aren't all engineers, but AMD's 3,000 aren't all engineers, either. NVIDIA don't have 3,000 extra support employees that AMD don't need. NVIDIA's margins show that their business is not inefficient. A good number of those employees are working on basic software libraries (CUDA libraries), integration of their products into deep learning frameworks, and vertical integration for various market segments: gaming, healthcare, financial services, manufacturing, etc. These efforts aren't just fluff. They are critical to leveraging NVIDIA's hardware into these various segments.
I think I remember reading that Nvidia has more software engineers than hardware people. Knowing how much effort they put into their software stack, makes a lot of sense. When AMD released their first GCN, I commented that they really need to put down some serious effort driving the software support. Let's see what they do this time...
And CUDA is what it comes down to for me. I'm in that semi-pro subset for which $1-2k for a GPU is not unreasonable if it reduces the runtime of my algorithms by 30-40%. But having to rewrite and validate all my CUDA/Matlab code to use OpenCL is a deal breaker, even if AMD halves the price/perf ratio. Converting people in my situation is the hole that AMD needs to climb out of to succeed at compute, not just make faster/cheaper cards.
Opencl's with is getting better. Yesterday opencl 2.2 was released. That includes c++ as a first class citizen support, an update for their standard ir format (c++ support and runtime optimization, the will be very helpful for one of the big problems with opencl), and the official opencl compliance suite and specification had also been made freely available. A number of the big ml frameworks (tf, caffe, touch) either have or are working on openCL implementations.
The high bandwidth cache looks interesting. I wonder if it will make it more difficult to program for, however. The VRAM is pretty slow.
I went to Baidu's DeepBench site and didn't find the info related to Vega. My stream spazzed out and I missed some of what Koduri said when he was showing the benchmark. I wonder if the benchmark chosen makes good use of the high bandwidth cache. His graphic just says "time to complete DeepBench", but DeepBench consists of four different tests and each one can have vastly different configurations of the underlying matrices, networks, etc. (whatever is being used for that particular test). From what I see on the DeepBench site, his graphic doesn't mean much on its own. AMD always seems to pull that shit and it's getting annoying.
Koduri said that Vega was first coming in the terms of this Frontier GPU, which was geared towards professionals and machine intelligence (although it's not the already announced Radeon Instinct MI25 even though it seems to fit its specs). Does that mean that consumer Vega will follow later, or will consumer Vega be announced at a later time and still arrive in June?
both companies have done so. AMD isnt the only one.
At least AMD hasnt cheated in benchmarks like intel has done in the past (by disabling optimizations and multicore on AMD cpus for their tests for example) or reduced texture quality to shit levels to be on par (old Nvidia cards)
So this all good and all but are we to assume the consumer vega will have less of everything. I mean if this thing is on par or slightly faster or slower than Titan Xp what do we expect from the consumer version then.
It depends on yields. It depends if this Frontier product is a show product or a viable production product with reasonable volume. It's possible the high-end consumer Vega will have the same number of cores and the same clock frequency, but with 8 GB of VRAM instead of 16.
I have two questions based on the article: 1. "NVIDIA has been very successful in the machine learning market over the last year, and if AMD can replicate NVIDIA’s success, not only will they make the machine learning market far more competitive for everyone"
How big is the market compared to pro and consumer graphic card markets?
2. Can you compare the specs not just to previous AMD cards but also to Nvidia, both Pascal and Volta?
Regarding Google, they seem to have announced a bit of a coup with their tpu2. Not only is it faster, and more scalable but it can also now be used to train your network. I'll be very curious as to how serious Google becomes with this hardware. The combination of their massive r&d budget, personnel, infrastructure, and interests place then in an enviable position.
I never said you could, and none of these products (including the recently unbelief volta card) are going to be selling high volumes. That's why i think calling them competitors with Nvidia (as someone in this section did when mentioning Nvidia's volta tensor units) doesn't make sense. The tpus are an advantage for Google's services in the same way their infrastructure "ip" is. For the companies that would be interested in tpus, the price isn't as important as the performance. That's why Nvidia can charge 16k for a card. Google isn't going to sell these, but someone will because you just don't need all the isa overhead that gpgpus have. It's the classic asic advantage that isn't going to go away for a while (their disadvantage is the one you've what mentioned).
It's unclear how much more powerful or power efficient the TPU2 is than Volta. What's clear is that it's far less flexible. Unlike a GPU, a TPU only works with TensorFlow networks, from what I understand. It doesn't seem to give the option of quantization of networks, unlike Volta.
Most importantly, I think most machine learning will be used in conjunction with analytics, simulation, or visualization. Software is being developed that will allow GPUs to process all those applications (machine learning training, machine learning inferencing, analytics, simulations, and visualization) from common data structures. In such cases, by using GPUs, data will be able to be processed with one architecture and without needing to spend the time and power to move them around to do it.
If the goal is to apply machine learning and only machine learning to a problem, the TPU may be preferable. But if the data should also be processed in other ways, it will probably make more sense to use a GPU.
What point of mine are you arguing about? If volta was a better fit for them they wouldn't have needed to create their own asic. That it is an asic is part of the reason why it is more efficient (ops/W, ops/s, ops/mm², or it's simply more scalable as their nodes seem to indicate). Also, these new tpus appear to be vastly different from the first gen given that it is end to end sp. Btw, I'm not saying that Google is going to put Nvidia out of business. As I've said, they aren't even competitors right now. So, you don't need to defend the honor of Nvidia to me:)
This claim: "Regarding Google, they seem to have announced a bit of a coup with their tpu2."
I think calling it a "bit of a coup" is an overestimation of its likely potential.
"If volta was a better fit for them they wouldn't have needed to create their own asic."
The TPU2 has been in development for 2 years, probably. Without seeing more information about this TPU, how it's actually being used within Google and their cloud, and real-world performance comparisons between the TPU and Volta, you can't assume that the TPU is better. This TPU is an active area of research for Google. It's something they seem to have committed manpower and energy to, and I'm sure it's something they have more plans for in the future. I wouldn't expect them to suddenly find out a few months back "Oh, Volta can do more than we expected, let's just pull the plug on this." So you need more information to make a determination as to what the real world advantages of these TPUs really are.
More importantly you aren't arguing against my point here, since you said "If Volta was a better fit FOR THEM..." (emphasis is mine). My post was focused on the overall machine learning market potential. In fact I said "If the goal is to apply machine learning and only machine learning to a problem, the TPU may be preferable. But if the data should also be processed in other ways, it will probably make more sense to use a GPU." Google is certainly a company that has massive machine learning workloads that consist entirely of machine learning training and inferencing. However, I don't think such workloads will be the majority of machine learning workloads going forwards. That's why, even if this TPU does have a significant price/performance and w/performance advantage over GPUs (and again, what advantage it does have is not clear at this point), it still doesn't qualify as a "coup".
"Also, these new tpus appear to be vastly different from the first gen given that it is end to end sp."
I think it's half-precision not single precision. But in any case, that's actually a disadvantage for the TPU, as I pointed out. They don't give the option for network quantization. The advantage of the end to end half-precision is that they can do both training and inferencing without moving the data around, something that GPUs can also do. GPUs, however, can extend this same advantage to mixed-workload machine learning.
"As I've said, they aren't even competitors right now. So, you don't need to defend the honor of Nvidia to me:)"
This is a counter-productive thing to say. I'm not defending the honor of NVIDIA, I'm analyzing the market.
The TPU looks interesting. Particularly because it's not just an ASIC. It is capable of scaling. So they've done a lot of work on networking and partitioning workloads, etc. As far as NVIDIA is concerned, it is a threat, if only because it means Google will be buying less GPUs from them. It also obviously poses a question for the future of the machine learning compute market: Will ASIC architectures force GPUs out of the majority of machine learning workloads? Calling it a "coup" is an answer to that question implying that it is a significant threat to disrupt the machine learning compute market. I think such a judgment at this point is hyperbolic and unwarranted.
The renderings are the images officially released to the press, so they're more likely to be the correct ones. But we shall see. This wouldn't have been the first time someone has held up a dummy card or prototype.
On Raja's AMA on reddit, he just said: "I grabbed an engineering board from the lab on the way to the Sunnyvale auditorium, and that boards works well with a 6 and an 8 pin. We decided to put two 8 pin connectors in the production boards to give our Frontier users extra headroom"
No consumer cards until the 2nd half? That is a fail, fail, fail.
The RX5xx series is already a rehash that no one cares about. I'm really hoping Nvidia decides to cease upon this opportunity and do a 1.. 2.. punch with launch of a consumer Volta in the form of a GTX 2080 series card.
At this point it has just been too long and I'm sick and tired of the ridiculous hype train.
HEAR THIS AMD! - I, for one am deciding not to 'hold out' for a little longer, Nvidia is getting my money this year. It's been too long.
Well idk how this is posible but my r9 390 has 125 fps on 2k paired with 7700k stock not hitting 125 but sitting on it. So as far im concerned I dont need vega or volta, but do I want them? Absolutely yes. Besides learning from the history of this two give a month or two after release and u will be buying Vega for 250$
Why this card should not be a consumer card? The so called "compute community" can use them too if it wants too. Nothing is stopping them. The price of 4096 stream processors is about $500 and HBM2 is $100. So, consumers should be able to have it at about $650 to $700. But AMD is withholding this GPU from the consumer market to supposedly derive profits from it, whereas it could have doubled its profit margin if the card was offered to both the consumer market and enterprise markets at the same time. After all the main difference between Radeon and Firepro cards are only in their drivers. Why not here?
I think this card's exotic (and previously unheard-of) 16 GB HBC on a 2048-bit bus may have a large part in the BOM being larger than can be sustained for a consumer card.
The cost of all of that was factored in. NVidia makes a hell of a profit on those margins and it does pay for materials and R&D and distribution costs.
I think that was a bit tongue-in-cheek like asking if something can run Crysis. It can be interesting to see how they do in consumer use mostly for giggles though.
Even if the product will end up in workstations and professional use, it'll be the first Vega product that can be reviewed. As such, it will be a very good proxy for the later gaming cards - especially if the top RX Vega ends up having a fully enabled Vega GPU, too.
Yep, it's how Vega translates all those 'raw' TF into frames that's important. GCN so far hasn't been the best at that. But I figure RTG have likely worked out why, and done something about it.
Actually your all wrong. 3DFX just announced an updated graphics card for the EISA bus. The exact specs are not being released just yet, but rumors are it will be very good on power consumption. It won't even need a separate power connection Industry insiders say it should turn the market on its head. Be on the look out for in August, 1992.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
134 Comments
Back to Article
xTRICKYxx - Wednesday, May 17, 2017 - link
This is a smart move especially offering a watercooling option. I really want Vega to succeed!nathanddrews - Wednesday, May 17, 2017 - link
13 TFLOPS is 2 better than the 1080Ti/TitanXP (original XP), so success will depend upon the price to be sure.vladx - Wednesday, May 17, 2017 - link
Huh? Both 1080 Ti and original Titan Pascal are around 11 TFLOPs.bubblyboo - Wednesday, May 17, 2017 - link
11+2=13?vladx - Wednesday, May 17, 2017 - link
That 'better' confused me, 'more' would've been more appropriate in that statement.nathanddrews - Wednesday, May 17, 2017 - link
I typed it with tongue firmly in cheek. TFLOPS ultimately doesn't mean everything, but it's probably a safe bet that there will be instances of Vega FE beating out Pascal in some tasks. Unfortunately for consumers (depending upon what sort of consumer you are), this will probably be priced closer to Titan XP/Xp than the 1080Ti if they are targeting compute folks over gamers.Or maybe they'll pull a Ryzen and undercut NVIDIA by 50%. haha
cap87 - Wednesday, May 17, 2017 - link
They won't, it's stated all over the slides that it's aimed at professionals and it won't come cheap. The consumer focused vega card will come later. The PRO card are already priced around the $999 USD so why would they sell a more powerful card for a lower price? It doesn't make any sense.tamalero - Wednesday, May 17, 2017 - link
Procards are usually unlocked in other things and boost way way more memory.The consumer version will probably have 8Gb of vram than the 16 of this monster.
Meteor2 - Wednesday, May 17, 2017 - link
Same situation as Zen. Nvidia own GPU compute like Intel owned HEDT. If AMD doesn't go cheaper, why buy it?Yojimbo - Wednesday, May 17, 2017 - link
Yes, and we don't yet know how Vega FLOPS compare. 11 Pascal FLOPS is probably usually faster than 13 Fury FLOPS would be. AMD is promising better architectural efficiency but that remains to be seen.vladx - Wednesday, May 17, 2017 - link
Indeed, RX 480 with 5.8 TFLOPs is about equal to 1060 GT with 3.9 TFLOPs so we need to see how much improved Vega really is.juliovillaz1990 - Wednesday, May 17, 2017 - link
You are wrong.. the 1060 its 5 tflops aproximately. You're nota taking un consideration the auto oc feature of Pascal cards, the 1060 auto overclocks nearly 2000mhz si its not 3.9 tflopsYojimbo - Thursday, May 18, 2017 - link
Where do you get this from? In what games is the GTX 1060 averaging 2000 MHz without overclocking?Regardless, it doesn't matter. You are agreeing that AMD's listed TFLOPs are worth less than NVIDIA's listed TFLOPs, which was the point of the thread. Whether it's true because NVIDIA's cards are able to consistently clock above their listed frequencies or because NVIDIA's architecture spends less time waiting around doesn't matter. (As long as NVIDIA's cards stay within their thermal envelopes, which they seem to do better than AMD's, in fact.)
Alexvrb - Sunday, May 21, 2017 - link
Gaming performance isn't the best measure when talking about professional graphics cards. It's silly to say "X TFLOPS on card A equals Y TFLOPS on card B because of gaming benchmarks" when discussing professional cards that are used for different software.vladx - Monday, May 22, 2017 - link
Don't know who you're responding to, I compared RX 480 and 1060 GT which neither being part of pro segment and I specifically mentioned we need a better picture of Vega's architecture.Yojimbo - Monday, May 22, 2017 - link
"Gaming performance isn't the best measure when talking about professional graphics cards. It's silly to say "X TFLOPS on card A equals Y TFLOPS on card B because of gaming benchmarks" when discussing professional cards that are used for different software."I think vladx's use of gaming performance to demonstrate that TFLOPs cannot be used to directly compare performance between two different architectures is perfectly fine. The person he replied to performed just such a comparison. And although vladx did say that Vega has to improve for AMD to catch up to NVIDIA, and perhaps that's not true in professional applications (I don't really know), the original post should not be assumed to be about professional performance. The OP simply said "I really want Vega to succeed", not "I really want this Vega Frontier Edition to succeed". I'm guessing he meant Vega in all its incarnations.
extide - Thursday, May 18, 2017 - link
When it comes to actual FLOPs ... FLOPS are FLOPS, they are the same. When it comes to how FLOPS translate to gaming performance .. that's another discussion, yes.CiccioB - Thursday, May 18, 2017 - link
No, and no.A TFLOPS means nothing at all. Because and operation is not like another.
We are used to FMA OPS to calculate theoretical peak TFLOPS capacities, but those do not really mean nothing. How many FMA do you use in an algorithm? How long can you sustain many FMA one after the other before bandwidth do fed them becomes a bottleneck?
A TFLOPS is not a precise unit of measure. It's just a "standard" way to classify computing devices. It may be that my device can't do FMA in a single cycle, so having half of your teorethica computational peak, but I may go faster than you.
So again a TFLOPS means nothing. What it really counts is the efficiency at which the architecture execute complex algorithms (and that is before you tale into account other resources like ROPS or TMUs or geometric management for a 3D pipeline).
By what we have seen today, nvidia is way more efficient than AMD in both area/perfomance and w/performance.
fanofanand - Wednesday, May 24, 2017 - link
Nvidia is using a bigger die to get similar numbers. Not sure where the area/performance you speak of stems from. AMD appears to be cheaper in a $/performance and area/performance comparison with Nvidia's latest and greatest design. I'm sure someone will correct me if I'm wrong, but the new Nvidia die is at the TSMC reticle limit.frenchy_2001 - Wednesday, May 24, 2017 - link
If you're talking about Volta, you're right, it is a much bigger die. What it is NOT though, is a competitor for this card. This Vega boasts 13TFlops, Volta boasts 120TFlops per chip for DNN (FP16 matrix multiply/add). It will also sell for a much higher price (Volta mezzanine cards are around $16k each). This is like comparing a Formula1 car and a Lexus. Not the same usage or market.This is supposed to compete with Titan XP. In that comparison, AMD uses much more expensive RAM (HBM2 vs GDDR5X) and die size is so far undisclosed, but past chips size and power consumption have not been in AMD's favor.
Cygni - Wednesday, May 17, 2017 - link
Yup. The leaks have an 8GB HB2, full die Vega gaming card at $599msrp, but I don't think they can stick to that number and be successful. Unfortunately for AMD, the 1080 Ti / GP102 will be a year old at that point, on a more mature process with assumedly better yields making pricecuts less painful for Nvidia then they might otherwise be.At least we have some competition to talk about on the high end again!
Morawka - Monday, June 5, 2017 - link
AMD Tflops do not equal nvidia Tflops.. I know it sounds crazy and stupid, but it's true, at least when it comes to gaming performance.Carmen00 - Wednesday, May 17, 2017 - link
"So for toady’s AMD Financial Analyst Day"? I wonder if Toady will be joined by Duke Igthorn.SetiroN - Wednesday, May 17, 2017 - link
I'm afraid this is going to be just as mistimed as Fiji.HollyDOL - Wednesday, May 17, 2017 - link
Those cards are really ugly, especially the gold edition. Thank god their target audience couldn't care less about visuals.It will be interesting to see the benchmarks. Until then it's just a promiseware. Hopefully they gave Ryan one or two to play with.
nagi603 - Wednesday, May 17, 2017 - link
It's a friggin' workstation card. If you want visuals in your workstation, you should see a doctor.HollyDOL - Wednesday, May 17, 2017 - link
quote myself: "Thank god their target audience couldn't care less about visuals."What's so hard to understand on that sentence?
MattMe - Wednesday, May 17, 2017 - link
I'm constantly surprised that the gaming market is still into glass panels and Max Power LEDs on their machines. Even as a teenager I'd have thought they were horrific. The fact that people care what the internals of their computer looks like baffles me.HollyDOL - Wednesday, May 17, 2017 - link
Unfortunately, it's increasingly difficult to get components with desired specs that come without the LED circus, you actually have to actively search for it when you head in high end desktop HW.xype - Wednesday, May 17, 2017 - link
I don’t know a single person who likes those. It’s like a whole industry is running off a coked-up marketer’s drug-induced "This is what kids like!" idea and everyone has to suffer for it.PhilipJ - Wednesday, May 17, 2017 - link
I enjoy the RGB LEDs & glass, but I could easily live without it. Though how do you suffer from it?gerz1219 - Wednesday, May 17, 2017 - link
I just bought a solid case with no silly see-through panels on my most recent build, which was my first in a few years. I'm still weirded out by the trend towards showing off the guts of your build. I'm still used to putting the components into the case and never seeing them again until upgrade. I can't imagine choosing the better looking GPU over the most powerful GPU in a given price range.tuxRoller - Wednesday, May 17, 2017 - link
Mostly grand mal.extide - Thursday, May 18, 2017 - link
eh,, It's all complex partial for meboozed - Friday, May 19, 2017 - link
I'd rather not pay for fripperies.Writer's Block - Wednesday, May 17, 2017 - link
For sure.I've seen some impressive lookin' pieces of shit, and never understood it.
Kvaern1 - Wednesday, May 17, 2017 - link
Not to mention highend gaming accesories. Thankfully there seems to be a trend towards releasing more aesthetically pleasing designs without too many silly LED's and no childish buzzwords or demon/dragon eye atrocities.cocochanel - Wednesday, May 17, 2017 - link
+1!!!LoLedzieba - Wednesday, May 17, 2017 - link
Have we found one? The mythical consumer that actually WANTS the 'gam8r' aesthetic with blinged out pointy shiny plastic bits and RGB lighting out the wazoo?bananaforscale - Monday, May 22, 2017 - link
Did you mean "gam3r"? Anyway, my gfx board has RGB LEDs, but that's coincidental; an Asus Strix happened to be the cheapest thing available, and the software can make the lighting reflect temps so it's actually kinda useful.fanofanand - Wednesday, May 24, 2017 - link
Reddit is crawling with the weirdos......eddman - Wednesday, May 17, 2017 - link
Never saw the point of good looking internal components. Once I put a graphics card in my computer, I don't see it again except for when I need to change something inside the case. It sits there out of the view under my desk.Meteor2 - Wednesday, May 17, 2017 - link
Lol I love the design! It's different, and kinda retro.othertomperson - Wednesday, May 17, 2017 - link
If the 1080 Ti is more expensive and performs worse I'd buy it over this Gold Edition thing because that's just how much of a turn off a water cooled GPU is for me at this point. It was a disastrous move to make the Fury X water-cooled only. I hope they see sense and open this up to AIBs this time.ET - Wednesday, May 17, 2017 - link
So why not buy the Blue Edition? I don't get why you have a problem with AMD trying to cater to two audiences.othertomperson - Wednesday, May 17, 2017 - link
Because of the implication that the blue edition is a cut down, less lower performing variant. People made this argument with Fiji, why not get a Fury if you don't like watercooling? Because you could have both with the 980 Ti. In fact there are Nvidia SKUs that offer watercooling as an option. Having it as an AIB option caters to two different audiences, this "watercooling or nothing" approach does not.Tams80 - Thursday, May 18, 2017 - link
It has exactly the same end result. The only thing that is different is that AMD are offering a water-cooled version themselves.The water-cooled Nvidia cards will be able to perform better than their non-water cooled ones. Does that for some reason put you off them? Apparently, not, so why not the same for AMD?
You've made up a message in your own mind that was not made by AMD.
Meteor2 - Wednesday, May 17, 2017 - link
Didn't the Fury X HAVE to be water-cooled because of its TDP? If it had been air-cooled it would either have been very noisy, or more likely constantly throttling because that's just too much heat to blow away from a box that size.CiccioB - Thursday, May 18, 2017 - link
Or it had to run at a lower clock just like the Fury without X.But that card had not the bars slightly longer than the 980Ti in 2 tests of 20. So they could not sell it that way, and they decide to go for the water cooling solution had have a little extra OC just to make the failure GPU shine a bit. They know it was a fail. But they made all they could to make it appear as good ads the concurrent solution.
My fear is that they are doing the same with Vega. It would mean they have nothing really good to offer.
Alexvrb - Sunday, May 21, 2017 - link
No. Horsepoop. AMD could have made an aircooled Fury X... it would have run hotter and been louder. There are aircooled overclocked 390X cards. An aircooled Fury X actually would have been fairly similar to the non-X Fury in terms of thermals and noise, they even carry the same TDP. They could have kept noise under control with large fans but it still would have run hotter and the watercooler included at the price tag made it a decent deal.roc1 - Wednesday, May 17, 2017 - link
Going workstation first appears provoked by insufficient supply or inability to compete in terms of performance per dollar.I really wish AMD had more objective slides and comparisons in their slides.
xype - Wednesday, May 17, 2017 - link
Newsflash: Not everything is a conspiracy.Workstation = big margins. That’s what AMD needs, more income. As a business decision it’s the only thing that makes sense.
JKay6969AT - Wednesday, May 17, 2017 - link
I would tend to agree that AMD's choice to go workstation first is most likely being led by supply issues with HBM2 I would disagree that it is anything to do with performance per dollar, in fact I would presume the opposite as the pro market is already dominated by nVidia and AMD would be competing directly with them to gain market share, they can't really accomplish this with poor performing cards in comparison with nVidia unless they plan to add a value sector to the pro market which is unlikely due to this being a very profitable sector and if there is one thing AMD needs more than anything right now it's profits and lots of them.I imagine desktop VEGA to compete in performance with the 1070, 1080 and 1080Ti while costing a bit less but the issue I have there is that of RAM or Cache size on the VEGA parts. I imagine that VEGA will launch in 4GB and 8GB varieties with HBM2 and while 8GB is plenty of RAM, comparing it to the 11GB in the 1080Ti will affect people's decision on which card to get, I know far too many people who would buy a worse GPU because it had more RAM, 'cause more is better right? LMAO!
I imagine this is part of the reason why AMD is showing how VEGA is able to run well with 2GB RAM with the High Speed Cache Controller enabled as it will help alleviate fears that 8GB isn't enough RAM or Cache. I also get why they call it cache, It reminds me of the old Pentium 2 Single Edge Contact Cartridge or SECC design which was inserted into Slot-1 Motherboards. The reason the CPU had the Slot form factor was because the cache chips were on the daughter board around the CPU rather than in later designs built right into the CPU itself. This is a very similar design strategy and should offer comparable speed benefits. I just look towards the next evolution of this process which should be the cache being baked into the GPU itself, there are issues with that as HBM memory is multi-layered and I imagine this would present significant design issues to incorporate this into the core of a GPU but hey, I'm a dreamer :-)
atragorn - Tuesday, May 23, 2017 - link
Actually the reason they went with slot-1 was to prevent AMD from sharing a socket with them.Super socket 7 was very good for AMD , external cache was the way it was done up to that point.
The K6-2 and K6-3 were clocking at 500mhz plus on socket 7.
The k6-3 was a beast with that extra cache but most people only saw it wasn't clocking as high and didnt realize the performance it offered.
Kvaern1 - Wednesday, May 17, 2017 - link
Volta wants a word with you.lefty2 - Wednesday, May 17, 2017 - link
You forgot to mention that Vega was 50% faster than P100 in Deep Bench.nevcairiel - Wednesday, May 17, 2017 - link
If any comparison would be meaningful, that would be to a Tesla V100, beating the competitors last years product a year later should be a given.lefty2 - Wednesday, May 17, 2017 - link
except that V100 isn't available til Augustcap87 - Wednesday, May 17, 2017 - link
It'll be over a year since the P100 is released and only a month after the VEGA FE that the V100 is released. It should be compared against the more recent one.Meteor2 - Wednesday, May 17, 2017 - link
Nobody outside the DOE will be buying V100s this year.Yojimbo - Thursday, May 18, 2017 - link
"Nobody outside the DOE will be buying V100s this year."According to NVIDIA's presentation that isn't true. They say that the DGX-1V is supposed to start shipping in Q3 and that third party servers will ship in Q4. Hyperscalers will probably start to get them at the same time DGX-1V shipments start or even earlier.
JKay6969AT - Wednesday, May 17, 2017 - link
And they also forgot to mention that nVidia just let Volta out of the bag for, yup, you guessed it, the Pro market. Volta with it's 16GB of HBM2 memory with 896Gbps bandwidth, 5120 CUDA cores and 640 TENSOR Cores makes even VEGA look old tech in comparison (For the Pro Machine Learning market). I just hope for AMD's sake that either VEGA can compete with Volta, Volta is targeting a different sector of the market than VEGA or that AMD is willing to take a bigger hit to the margins with VEGA to stay competitive.Volta on the desktop I imagine will...
Not have the TENSOR cores as they seem to be optimized for machine learning tasks (that's my understanding of them anyways)
Not come with 16GB of HBM2, more likely 8GB or 12GB+ of GDDR5X or 6. Premium models may offer 16GB HBM2 like the Titan Xv but the 2080 Ti will most likely come with 12GB+ GDDR5X or 6 depending on when it launches.
Not offer 896Gbps bandwidth due to only having 8GB HBM2 it will offer 480~Gbps like the VEGA counterpart.
This makes desktop VEGA and VOLTA potentially much closer in performance than the Pro versions meaning that AMD should be able to compete for some of the desktop market.
Currently Volta costs $18000 USD for a single board, this gives AMD a fair amount of wiggle room on competing price :-) lol
However I must say I am impressed with what nVidia has shown with Volta, even though I would never need such a product, I am impressed with the advancements they have made and the sheer size of the GPU is incredible, however it could be telling that nVidia NEEDED all that die space to get the performance boost, could Volta be less efficient than hoped? I don't know, it's still impressive.
lmcd - Wednesday, May 17, 2017 - link
Any more buzzwords to capitalize here?rarson - Wednesday, May 17, 2017 - link
Tensor cores are there to compete with Google.tuxRoller - Wednesday, May 17, 2017 - link
That's an interesting use of the word "compete".Ryan Smith - Wednesday, May 17, 2017 - link
As a rule of thumb, we don't publish competitive performance slides. While I doubt AMD is lying, it's not something we can verify at this time, and in these cases vendors usually cherry-pick at a minimum.jjj - Wednesday, May 17, 2017 - link
You seem to be jumping to the conclusion that this is the first Vega to launch and consumer comes later but don't quite think that AMD has stated that. Yesterday's event wasn't the right place to launch consumer products but that doesn't mean there won't be consumer Vega before this one. The 16GB of HBM makes it even less likely for this one to be the first to ship.You also seem to miss that this is not Vega Instinct , it's a Pro card but seems that it's a kind of high volume sampling SKU so AMD is either not quite ready on the software side or can't make all that many yet , for high volume deployment in data centers.
BTW AMD lists 25TFLOPS FP16 not 26, so base clocks and they likely have a reason to do so.
As for all the preaching about prices and higher margins, that's false and hugely misleading. Higher margins and ASPs sure but much lower volumes and substantial investments in both hardware and software. The high end gaming is plenty profitable, if the product is good enough to sell because sites like this one are promoting ridiculously priced GPUs instead of urging users to never buy such poor value products. It's Trumpian that hardware sites are not making fun of such terrible value products instead of trying really hard to sell them. The press today serves the corporations not the consumer- wasn't like this just a few years ago. You are actively working hard to make people dumber and waste their money every time you don't just laugh at a 500$ GPU or a 700$ phone. Anyway, this is not 10 years ago when the Pro cards were so much more profitable and it's because the press is utterly corrupted and made the consumer much much stupider. Today high end gaming funds the pro and server cards and only in the next few years ,when the deep learning market grows to a sufficient size, it will be able to fully fund the product development, but only for the market leaders.
Yojimbo - Wednesday, May 17, 2017 - link
There was no reason for AMD to not launch a consumer card yesterday. It wasn't a developer's conference. It was an investor's day event. Investors don't care if a product is a consumer product or a professional product. They just want profitable products, and they'd like to hear about as much as possible.I don't remember the exact quote, but from what I remember Raja Koduri said things like "There's been a lot of rumors. Everybody's asking where's Vega? When will it appear? Well today we can say the first Vega will be released at the end of June, called Vega Frontier." His quote did make it sound like no consumer Vega would be coming before Vega Frontier. Whether a consumer Vega will be appearing alongside Vega Frontier is less clear, but the situation suggests to me that it will not.
Meteor2 - Wednesday, May 17, 2017 - link
Yes that's a bit worrying. I was expecting something consumer at Computex or a special event in June.rarson - Wednesday, May 17, 2017 - link
http://pro.radeon.com/en-us/vega-frontier-edition/"if gaming is your primary reason for buying a GPU, I’d suggest waiting just a little while longer for the lower-priced, gaming-optimized Radeon RX Vega graphics card."
They outright stated consumer Vega is not coming first. And then put it in print as well.
Yojimbo - Thursday, May 18, 2017 - link
Well, there you go, I guess. Thanks.DJ_DC - Wednesday, May 17, 2017 - link
Noone pointed out how they are comparing FP32 deepbench with P100 when P100 has 1:2 FP64 performance (5 TFLOPS of FP64) and this Vega card only has 1:16 (800 GFLOPS of FP64)?A bit disingenuous don't you think?
MrSpadge - Wednesday, May 17, 2017 - link
There are use cases for FP32 and there are cases for FP64 - both are important.Yojimbo - Wednesday, May 17, 2017 - link
I doubt DeepBench is concerned with FP64 data types at all.The thing with those DeepBench numbers is AMD just put up some run times. They didn't specify exactly what the parameters were for the runs, or how they got one single "DeepBench" run time number when from what I see DeepBench consists of four separate types of tests.
eddman - Wednesday, May 17, 2017 - link
I really hope Vega has a higher gaming performance/flops ratio. I'm tired of nvidia's prices. (Unless AMD follows suit, which they might. They are not in a position to start a price war)webdoctors - Wednesday, May 17, 2017 - link
Any Vega card that's expecting to make a profit will be priced higher than the 1080Ti. These things are BOM beasts, that HBM memory and interposer ain't cheap. There's a reason this thing is aimed at Pros, because consumers aren't gonna drop $1K on a Vega, it'd be DOA.Sure AMD might sell a very limited batch at $600 to compete with 1080Ti, and claim they've got the perf/$ crown, but every card they sell will be $100s lost from their bottom line.
CiccioB - Thursday, May 18, 2017 - link
That's why they are trying to sell less Vega in the consumer market as possible.Just few to show in benchmark and show they are still alive (using a beefy large solution against a smaller cheaper one) and then trail until 7nm where maybe they hope to have some advantage over nvidia which is simply bulldozering them under all points of view.
They claimed Vega will be out in H1 2017. They are not ready with a product that can compete with nvidia at the same price level. This chip probably is like a Fiji against GM200: it costs too much and returns nothing. By the way, they have to maintain the promise to the investor for the GPU release, so they came out with this ridiculous plan to show (you certainly won't find the card at end of June on the market, not probably at end of July) the card at end of June and delay the presentation of the failing consumer card as much as possible.
vladx - Wednesday, May 17, 2017 - link
Just like I figured, looks like performance is between a 1080 and 1080Ti.BurntMyBacon - Wednesday, May 17, 2017 - link
You may in fact have the correct conclusion, but based on what premise. The 1080 and 1080Ti are gaming GPUs and have been evaluated in gaming applications. The VEGA FE presented here is a professional card that should be evaluated in professional applications (which are not comparable to gaming applications). There is no gaming evaluation presented in the article. There isn't even a professional evaluation. You just pronounced yourself correct without presenting any supporting evidence. Unfortunately, even if your conclusion ends up being correct, you statement is still a fallacy.vladx - Wednesday, May 17, 2017 - link
Obviously, I was extrapolating from this pro card to obtain the consumer version's performance which I'm betting will have a few units disabled compared to this one.BurntMyBacon - Thursday, May 18, 2017 - link
I suppose it's not as obvious to me as to you. The article doesn't present any benchmarks, much less gaming benchmarks. So I ask again, based on what premise?Theoretical (Max) single / double / half precision TFLOPS perhaps. Historically these numbers haven't been very useful for comparing gaming performance between vendors.
Memory bandwidth - suffers the same problem as above.
Pixel fill rate - only a small part of the story that may be more or less useful depending on application.
Outside benchmarks - please share. I'd love to see some actual gaming (or gaming related) performance number. The validity of leaked or cherry picked benchmarks is perhaps questionable, but by necessity incorrect. Don't let that keep you from sharing.
I apologize if I sound obstinate to you. The truth is, I actually like your conclusion and I am inclined to agree with it based on a combination of specifications, theoretical performance numbers, historical tendencies, market forces, and "gut feeling". However, I have yet to see anywhere near enough evidence to conclude more than a very rough wag on this one. Known changes to the architecture make historical tendencies a fuzzy approximation at best. We don't yet know how much (or little) they will affect the gaming performance.
Meteor2 - Wednesday, May 17, 2017 - link
The software stack slide is telling; CUDA owns the acceleration space and OpenCL isn't very popular. AMD doesn't have an answer here and it's too late anyway because CUDA is established and it works. Nvidia are being bastards and not supporting OpenCL 2 either, locking the market in.Yojimbo - Wednesday, May 17, 2017 - link
It's not too late, but it's going to take a lot of hard work, determination, and resources. Koduri's claim that 10 or 20 engineers working for a few weeks on each framework is all that's necessary is not auspicious. Developers need assurance that AMD are going to actively support the ecosystem, something that they haven't been doing up to this point. Those number of engineers for that amount of time probably is what it took them to be able to run one chosen benchmark well that matched up particularly well with their chosen architecture (my guess is that the high bandwidth cache is a prime candidate for an advantage to focus on). As far as I know, for general usage, the BLAS libraries in GPUOpen are significantly slower than NVIDIA's cuBLAS.There's a lot more to support in GPU computing than just machine learning, as well. If they only focus on machine learning they will lose a lot of opportunities from companies that want to do simulations and analytics along with machine learning, which is probably the majority of them. Each application has its own issues, and the people in those market segments are mostly not machine learning experts. AMD has 3,000 employees in its Radeon Technologies Group. NVIDIA has 10,000 employees, and they don't have thousands of them sitting around doing nothing.
As far as OpenCL, even when NVIDIA's OpenCL support was more current, CUDA had the advantage because NVIDIA actively supported the space with high performance libraries. If NVIDIA controls both the architecture and the programming model they are able to bring features to market much faster and more efficiently, which is pretty important with the pace of innovation that's going on right now. My guess is that opening CUDA would probably be a more beneficial action for the community than supporting OpenCL at the moment, unless opening CUDA meant losing control of it.
BurntMyBacon - Wednesday, May 17, 2017 - link
Yes, it would be more beneficial to the community to open CUDA up to other vendors. However, I think it is about as likely to happen as opening up PhysX or G-Sync. nVidia doesn't exactly have a reputation for opening up proprietary tech.Yojimbo - Wednesday, May 17, 2017 - link
Well, they are open sourcing their upcoming deep learning accelerator ASIC. They recently open sourced a lot of GameWorks, I think. They will open source things when it's beneficial to them. That's the same thing that can be said for AMD or most tech companies. Considering their output with these initiatives and market position, AMD open sourcing GPUOpen and FreeSync was beneficial to them.NVIDIA has a history of not waiting around for committees and forging on their own the things they need/want, such as with CUDA, NVLink, and G-Sync. They are trying to build high-value platforms. They spend the money and take the risks in doing that and their motivation is profitability.
I also don't expect them to open up CUDA at the moment.
Meteor2 - Wednesday, May 17, 2017 - link
I'd forgotten NVLink. I'm not seeing any answer to that at all from AMD and NVLink 2 is fundamental to accelerated HPCs which are vaguely programmable.tuxRoller - Wednesday, May 17, 2017 - link
Opencapi (phenomenal latency). Given the huge number of lanes that Naples has, it would be a good target for opencapi.There's also the in-progress genz fabric, but that's not close to being standardized, and is more of an interconnect between nodes, I suppose.
Yojimbo - Wednesday, May 17, 2017 - link
OpenCAPI is managed by the host processor (CPU). From the way I've seen it described, it is not meant for accelerators to use to link to each other without going through a host processor that supports OpenCAPI. AMD presumably might support OpenCAPI on their CPUs, but they can't count on their GPUs being used with their CPUs. Perhaps CCIX plans to allow accelerator to accelerator interconnect, I'm not sure.Regardless, both OpenCAPI and CCIX are not available yet, and certainly not enabled on the upcoming Vega products.
tuxRoller - Friday, May 19, 2017 - link
Capi is definitely managed by the processor. Opencapi is supposed to be a "ground up" rethink of a high-speed, low latency interconnect for various on-node resources.Ibm appears to be calling theirs "Blue Link" (or something similar), and that's supposed to be released this year.
I would be astonished if amd hasn't done some work towards supporting one of these standards on vega, even if not on this exact chip.
Regardless, these are the options for amd, and since they are "standards" the hope is that others will begin buying into the spec as well.
If opencapi remains a star network then amd can offer a competitive, even if not identical, solution instead of ceding the market to Nvidia.
BurntMyBacon - Thursday, May 18, 2017 - link
@Yojimbo: "Well, they are open sourcing their upcoming deep learning accelerator ASIC."Good Point.
@Yojimbo: "They recently open sourced a lot of GameWorks, I think."
I didn't know about that, but I'm not sure how much this one matters as gameworks products already run on competitor's GPUs and we already know that gameworkst is optimized for nVidia GPUs (Why wouldn't it be?).
@Yojimbo: "They will open source things when it's beneficial to them. That's the same thing that can be said for AMD or most tech companies."
I generally agree, but I would be remiss if I did not point out HyperTransport as an example of an AMD open standard initiative that predated it's proprietary Intel counterpart (QPI). It was also introduced in 2001 with their Athlon64 architecture, a time when AMD was doing well due to the recent success of the Athlon processors. It doesn't happen as often as I'd like, but sometimes tech companies will prioritize long term interoperability and viability over short term gain.
@Yojimbo: "Considering their output with these initiatives and market position, AMD open sourcing GPUOpen and FreeSync was beneficial to them."
I don't disagree. Historically, AMD has found themselves in this position far more often than nVidia.
@Yojimbo: "NVIDIA has a history of not waiting around for committees and forging on their own the things they need/want, such as with CUDA, NVLink, and G-Sync.
They don't need to wait on a committee to open up a standard. AMD forged ahead with mantle until such a point as the industry was ready to work the issue. They also don't necessarily need to give up control of their standards to allow other vendors to use them. It would even be reasonable to demand a licensing fee for access to some of their technologies.
@Yojimbo: "They are trying to build high-value platforms. They spend the money and take the risks in doing that and their motivation is profitability."
Keeping proprietary tech proprietary is their right. When they are are dominating a one or two player market, the risk is low, the returns are high, and the decision makes sense. If there were several major players or they weren't the dominant player, this would be far more risky and interoperability would be a bigger concern. Given the current market, I expect they'll keep their proprietary tech in house.
peevee - Wednesday, May 17, 2017 - link
"NVIDIA has 10,000 employees, and they don't have thousands of them sitting around doing nothing."But nVidia also has more markets to serve, their have their own CPUs, mobile, automotive...
And I wonder what percentage of those employees are engineers.
Yojimbo - Thursday, May 18, 2017 - link
"But nVidia also has more markets to serve, their have their own CPUs, mobile, automotive..."It's AMD that have their own CPUs, not NVIDIA. But I didn't include AMD's CPU business in their employee count. NVIDIA is rather focused. Virtually all their products are based on their GPU architecture. They do develop their own CPU core based on the ARM-v8 instruction set, but it is geared entirely for one purpose, to be the CPU for their self-driving car SoC. Mobile and automotive are close to the same market for NVIDIA. They probably put 100 employees or something on the Switch. NVIDIA have gaming GPUs, professional GPUs, datacenter, and Tegra (mobile/automotive). AMD are trying to go after all that same stuff other than automotive. Whatever number of employees NVIDIA use on the Switch, AMD probably use more on the XboX/PS4 projects.
NVIDIA have 3 times the numbers of employees as RTG. NVIDIA don't have 2/3 of their employees working on mobile and automotive. Most likely they have well under 1/3. Let's be conservative and place it at 1/3. That means NVIDIA have roughly twice the number of employees working on the other segments, about 3,000 people. No, they aren't all engineers, but AMD's 3,000 aren't all engineers, either. NVIDIA don't have 3,000 extra support employees that AMD don't need. NVIDIA's margins show that their business is not inefficient. A good number of those employees are working on basic software libraries (CUDA libraries), integration of their products into deep learning frameworks, and vertical integration for various market segments: gaming, healthcare, financial services, manufacturing, etc. These efforts aren't just fluff. They are critical to leveraging NVIDIA's hardware into these various segments.
hammer256 - Thursday, May 18, 2017 - link
I think I remember reading that Nvidia has more software engineers than hardware people. Knowing how much effort they put into their software stack, makes a lot of sense.When AMD released their first GCN, I commented that they really need to put down some serious effort driving the software support. Let's see what they do this time...
1mpetuous - Wednesday, May 17, 2017 - link
And CUDA is what it comes down to for me. I'm in that semi-pro subset for which $1-2k for a GPU is not unreasonable if it reduces the runtime of my algorithms by 30-40%. But having to rewrite and validate all my CUDA/Matlab code to use OpenCL is a deal breaker, even if AMD halves the price/perf ratio. Converting people in my situation is the hole that AMD needs to climb out of to succeed at compute, not just make faster/cheaper cards.Haawser - Friday, May 19, 2017 - link
Why not use HIP ? https://github.com/GPUOpen-ProfessionalCompute-Too...tuxRoller - Wednesday, May 17, 2017 - link
Opencl's with is getting better. Yesterday opencl 2.2 was released. That includes c++ as a first class citizen support, an update for their standard ir format (c++ support and runtime optimization, the will be very helpful for one of the big problems with opencl), and the official opencl compliance suite and specification had also been made freely available.A number of the big ml frameworks (tf, caffe, touch) either have or are working on openCL implementations.
Yojimbo - Wednesday, May 17, 2017 - link
The high bandwidth cache looks interesting. I wonder if it will make it more difficult to program for, however. The VRAM is pretty slow.I went to Baidu's DeepBench site and didn't find the info related to Vega. My stream spazzed out and I missed some of what Koduri said when he was showing the benchmark. I wonder if the benchmark chosen makes good use of the high bandwidth cache. His graphic just says "time to complete DeepBench", but DeepBench consists of four different tests and each one can have vastly different configurations of the underlying matrices, networks, etc. (whatever is being used for that particular test). From what I see on the DeepBench site, his graphic doesn't mean much on its own. AMD always seems to pull that shit and it's getting annoying.
Koduri said that Vega was first coming in the terms of this Frontier GPU, which was geared towards professionals and machine intelligence (although it's not the already announced Radeon Instinct MI25 even though it seems to fit its specs). Does that mean that consumer Vega will follow later, or will consumer Vega be announced at a later time and still arrive in June?
vladx - Wednesday, May 17, 2017 - link
Unfortunately cherry-picking is what AMD knows to do best.tamalero - Wednesday, May 17, 2017 - link
both companies have done so. AMD isnt the only one.At least AMD hasnt cheated in benchmarks like intel has done in the past (by disabling optimizations and multicore on AMD cpus for their tests for example) or reduced texture quality to shit levels to be on par (old Nvidia cards)
vladx - Wednesday, May 17, 2017 - link
And yes, the article states the consumer Vega will almost surely launch sometime in the second half of the year.rocky12345 - Wednesday, May 17, 2017 - link
So this all good and all but are we to assume the consumer vega will have less of everything. I mean if this thing is on par or slightly faster or slower than Titan Xp what do we expect from the consumer version then.Yojimbo - Wednesday, May 17, 2017 - link
It depends on yields. It depends if this Frontier product is a show product or a viable production product with reasonable volume. It's possible the high-end consumer Vega will have the same number of cores and the same clock frequency, but with 8 GB of VRAM instead of 16.peevee - Wednesday, May 17, 2017 - link
I have two questions based on the article:1. "NVIDIA has been very successful in the machine learning market over the last year, and if AMD can replicate NVIDIA’s success, not only will they make the machine learning market far more competitive for everyone"
How big is the market compared to pro and consumer graphic card markets?
2. Can you compare the specs not just to previous AMD cards but also to Nvidia, both Pascal and Volta?
peevee - Wednesday, May 17, 2017 - link
3. What is the FP64 performance? What is INT8 performance?tuxRoller - Wednesday, May 17, 2017 - link
This is the prettiest card i can recall.It has a very Google aesthetic.
tuxRoller - Wednesday, May 17, 2017 - link
Regarding Google, they seem to have announced a bit of a coup with their tpu2. Not only is it faster, and more scalable but it can also now be used to train your network.I'll be very curious as to how serious Google becomes with this hardware. The combination of their massive r&d budget, personnel, infrastructure, and interests place then in an enviable position.
Meteor2 - Wednesday, May 17, 2017 - link
Yeah but you can't buy TPUs, only access to them (via Google Compute). I can't believe that will be very popular (or cheap).tuxRoller - Wednesday, May 17, 2017 - link
I never said you could, and none of these products (including the recently unbelief volta card) are going to be selling high volumes. That's why i think calling them competitors with Nvidia (as someone in this section did when mentioning Nvidia's volta tensor units) doesn't make sense. The tpus are an advantage for Google's services in the same way their infrastructure "ip" is.For the companies that would be interested in tpus, the price isn't as important as the performance. That's why Nvidia can charge 16k for a card.
Google isn't going to sell these, but someone will because you just don't need all the isa overhead that gpgpus have. It's the classic asic advantage that isn't going to go away for a while (their disadvantage is the one you've what mentioned).
Yojimbo - Thursday, May 18, 2017 - link
It's unclear how much more powerful or power efficient the TPU2 is than Volta. What's clear is that it's far less flexible. Unlike a GPU, a TPU only works with TensorFlow networks, from what I understand. It doesn't seem to give the option of quantization of networks, unlike Volta.Most importantly, I think most machine learning will be used in conjunction with analytics, simulation, or visualization. Software is being developed that will allow GPUs to process all those applications (machine learning training, machine learning inferencing, analytics, simulations, and visualization) from common data structures. In such cases, by using GPUs, data will be able to be processed with one architecture and without needing to spend the time and power to move them around to do it.
If the goal is to apply machine learning and only machine learning to a problem, the TPU may be preferable. But if the data should also be processed in other ways, it will probably make more sense to use a GPU.
tuxRoller - Thursday, May 18, 2017 - link
What point of mine are you arguing about?If volta was a better fit for them they wouldn't have needed to create their own asic. That it is an asic is part of the reason why it is more efficient (ops/W, ops/s, ops/mm², or it's simply more scalable as their nodes seem to indicate).
Also, these new tpus appear to be vastly different from the first gen given that it is end to end sp.
Btw, I'm not saying that Google is going to put Nvidia out of business. As I've said, they aren't even competitors right now. So, you don't need to defend the honor of Nvidia to me:)
Yojimbo - Friday, May 19, 2017 - link
This claim: "Regarding Google, they seem to have announced a bit of a coup with their tpu2."I think calling it a "bit of a coup" is an overestimation of its likely potential.
"If volta was a better fit for them they wouldn't have needed to create their own asic."
The TPU2 has been in development for 2 years, probably. Without seeing more information about this TPU, how it's actually being used within Google and their cloud, and real-world performance comparisons between the TPU and Volta, you can't assume that the TPU is better. This TPU is an active area of research for Google. It's something they seem to have committed manpower and energy to, and I'm sure it's something they have more plans for in the future. I wouldn't expect them to suddenly find out a few months back "Oh, Volta can do more than we expected, let's just pull the plug on this." So you need more information to make a determination as to what the real world advantages of these TPUs really are.
More importantly you aren't arguing against my point here, since you said "If Volta was a better fit FOR THEM..." (emphasis is mine). My post was focused on the overall machine learning market potential. In fact I said "If the goal is to apply machine learning and only machine learning to a problem, the TPU may be preferable. But if the data should also be processed in other ways, it will probably make more sense to use a GPU." Google is certainly a company that has massive machine learning workloads that consist entirely of machine learning training and inferencing. However, I don't think such workloads will be the majority of machine learning workloads going forwards. That's why, even if this TPU does have a significant price/performance and w/performance advantage over GPUs (and again, what advantage it does have is not clear at this point), it still doesn't qualify as a "coup".
"Also, these new tpus appear to be vastly different from the first gen given that it is end to end sp."
I think it's half-precision not single precision. But in any case, that's actually a disadvantage for the TPU, as I pointed out. They don't give the option for network quantization. The advantage of the end to end half-precision is that they can do both training and inferencing without moving the data around, something that GPUs can also do. GPUs, however, can extend this same advantage to mixed-workload machine learning.
"As I've said, they aren't even competitors right now. So, you don't need to defend the honor of Nvidia to me:)"
This is a counter-productive thing to say. I'm not defending the honor of NVIDIA, I'm analyzing the market.
The TPU looks interesting. Particularly because it's not just an ASIC. It is capable of scaling. So they've done a lot of work on networking and partitioning workloads, etc. As far as NVIDIA is concerned, it is a threat, if only because it means Google will be buying less GPUs from them. It also obviously poses a question for the future of the machine learning compute market: Will ASIC architectures force GPUs out of the majority of machine learning workloads? Calling it a "coup" is an answer to that question implying that it is a significant threat to disrupt the machine learning compute market. I think such a judgment at this point is hyperbolic and unwarranted.
VulkanMan - Wednesday, May 17, 2017 - link
Ryan, the card Raja showed was 8+6 for power. So, the renderings were not correct.https://s12.postimg.org/5ievshwu5/x2_GNw_Bxw2_X-b_...
Ryan Smith - Wednesday, May 17, 2017 - link
The renderings are the images officially released to the press, so they're more likely to be the correct ones. But we shall see. This wouldn't have been the first time someone has held up a dummy card or prototype.VulkanMan - Thursday, May 18, 2017 - link
On Raja's AMA on reddit, he just said: "I grabbed an engineering board from the lab on the way to the Sunnyvale auditorium, and that boards works well with a 6 and an 8 pin. We decided to put two 8 pin connectors in the production boards to give our Frontier users extra headroom"Cellar Door - Wednesday, May 17, 2017 - link
No consumer cards until the 2nd half? That is a fail, fail, fail.The RX5xx series is already a rehash that no one cares about. I'm really hoping Nvidia decides to cease upon this opportunity and do a 1.. 2.. punch with launch of a consumer Volta in the form of a GTX 2080 series card.
At this point it has just been too long and I'm sick and tired of the ridiculous hype train.
HEAR THIS AMD! - I, for one am deciding not to 'hold out' for a little longer, Nvidia is getting my money this year. It's been too long.
MsYunaAyashi - Thursday, May 18, 2017 - link
very well then. Good bye!Onasis - Wednesday, May 17, 2017 - link
Well idk how this is posible but my r9 390 has 125 fps on 2k paired with 7700k stock not hitting 125 but sitting on it. So as far im concerned I dont need vega or volta, but do I want them? Absolutely yes. Besides learning from the history of this two give a month or two after release and u will be buying Vega for 250$Chaser - Wednesday, May 17, 2017 - link
AMD GPUs are becoming 2nd hand junk or vaporware. Only their fanboys care anymore.supdawgwtfd - Wednesday, May 17, 2017 - link
Bahahahaahhaahahhahah!!!!Wow... biased much?
fanofanand - Wednesday, May 24, 2017 - link
I think you accidentally typed in anandtech, when you meant to put in wccftechversesuvius - Thursday, May 18, 2017 - link
Why this card should not be a consumer card? The so called "compute community" can use them too if it wants too. Nothing is stopping them. The price of 4096 stream processors is about $500 and HBM2 is $100. So, consumers should be able to have it at about $650 to $700. But AMD is withholding this GPU from the consumer market to supposedly derive profits from it, whereas it could have doubled its profit margin if the card was offered to both the consumer market and enterprise markets at the same time. After all the main difference between Radeon and Firepro cards are only in their drivers. Why not here?Hul8 - Thursday, May 18, 2017 - link
I think this card's exotic (and previously unheard-of) 16 GB HBC on a 2048-bit bus may have a large part in the BOM being larger than can be sustained for a consumer card.vladx - Thursday, May 18, 2017 - link
Cost of materials is just a part, there's also R&D, marketing and even distribution costs to take into account.versesuvius - Friday, May 19, 2017 - link
The cost of all of that was factored in. NVidia makes a hell of a profit on those margins and it does pay for materials and R&D and distribution costs.bananaforscale - Monday, May 22, 2017 - link
$100 for 16GB of HBM2? [citation needed]ET - Thursday, May 18, 2017 - link
I will never buy this, but I do hope that hardware sites will get Vega Frontier for review and benchmark lots of games on it.MsYunaAyashi - Thursday, May 18, 2017 - link
this aint for gaming lmao. this is for workstation data centersIcehawk - Thursday, May 18, 2017 - link
I think that was a bit tongue-in-cheek like asking if something can run Crysis. It can be interesting to see how they do in consumer use mostly for giggles though.Hul8 - Saturday, May 20, 2017 - link
Even if the product will end up in workstations and professional use, it'll be the first Vega product that can be reviewed. As such, it will be a very good proxy for the later gaming cards - especially if the top RX Vega ends up having a fully enabled Vega GPU, too.Hul8 - Saturday, May 20, 2017 - link
Pro drivers (if it uses them) may hinder this, of course.toyotabedzrock - Thursday, May 18, 2017 - link
The MHz per SP teraflop is almost identical between Fiji and Vega.Haawser - Friday, May 19, 2017 - link
Yep, it's how Vega translates all those 'raw' TF into frames that's important. GCN so far hasn't been the best at that. But I figure RTG have likely worked out why, and done something about it.Threska - Sunday, May 21, 2017 - link
It would be nice to see what these Pro cards bring to the virtualization market? Also high-margin.mdw9604 - Tuesday, May 23, 2017 - link
Actually your all wrong. 3DFX just announced an updated graphics card for the EISA bus. The exact specs are not being released just yet, but rumors are it will be very good on power consumption. It won't even need a separate power connection Industry insiders say it should turn the market on its head. Be on the look out for in August, 1992.