Ah, but NetBurst was from 2000 or so, and it was retired in 2006. I would guess that had Intel kept pushing with NetBurst and transitioned to 45nm and now 32nm, they too could be hitting ~8.5GHz while sucking down a ton of power. Still not sure what AMD was thinking with this long pipeline for Bulldozer. :-\
Any numbers to back this up ? I don't think AMD has spilled the beans on particular details of Bulldozer architecture yet (unlike Intel does and AMD did for their previous architectures).
Two cores in marketing terms. No different than saying an Intel SB chip with 4 physical cores + hyper-threading = 8 cores. Of course, Intel doesn't do that.
What is it exactly all these experiments are supposed to tell me ? That, yes, the standard performance on bulldozer is significantly slower than on i7, but I can (and am supposed to) easily overclock it to 4GHz or more, thus making it competitive ? Or that I have to treat the extra cores as spare tires in case I get a flat ?
I simply fail to see the real world significance. Please enlighten me.
Well, I don't think there is any real world significance. Only one Bulldozer module (i.e. two "cores") was enabled, so this was just aiming for as high frequency as possible. With all cores enabled, you will be looking at much lower frequencies.
Basically, this is one kind of a sport. We don't benefit from it but a world record is a world record. You don't benefit if someone breaks the 100m sprint record, do you? ;-)
There is no real-work significance. All we learn from this is that Bulldozer can be clocked really high in theory (much like NetBurst). It still sucks in normal usage, even overclocked under water cooling against Sandy Bridge.
I agree about these experiments, but it has been shown that the performance issues are due to how Windows is currently assigning threads to the cores. It's a scheduler problem as much as it is a design issue.
Basically, if Windows(or any other OS) is set to assign threads to one core per module as the way to assign threads, you get higher performance than if you load up core 1, 2(module 1), then 3 and 4(module 2), leaving the remaining two modules unused. The whole "module" approach is a big part of the performance problem. Some have even wondered if you force the Core i7 scheduler to be used instead of the normal "default" scheduler if that would work better with Bulldozer at this point.
Jeez, what's with all the negativity? While the max overclock isn't 24/7 stable, it IS an indicator of how well the architecture overclocks in general, at all levels of cooling. It could help someone avoid a really terrible line of processors. For example, if I had done some research on the original B2 Phenom's overclockability, I would have steered away, but unfortunately that was not the case.
I am looking at this from a rational perspective, not being negative. Clock speed, as seen in the days of the Pentium 4 vs. Athlon 64 means less than performance when running applications.
I am an AMD fan, but when it comes right down to it, if Windows 7 by default shows that an 8-core Bulldozer processor doesn't win hands down compared to a Phenom 2 at the same clock speed, that shows that there are problems. The CORE design of Bulldozer is supposed to be better, clock for clock compared to the previous generation, and we are not seeing that. A 4-core processor based on a Bulldozer core won't perform as well clock for clock compared to a 4-core Phenom 2 due to modules.
I think that is AMD's propaganda. While W8 is still in pre-beta state, the tests have shown that it won't really improve the performance. Of course, it may improve later on as W8 develops.
However, what is the point? Intel will have IB out before W8 and AMD should also be out with Piledriver soon.
The exact opposite is true according to AMD, because if you leave two modules asleep, they get power gated off, turbo kicks in, and the two modules you're loading get a much higher clock speed.
Clock speed does NOT equal performance, which is why so many people are posting negative comments about the performance of Bulldozer. Better performance per clock means that Turbo Core is a nice idea, but the "module" concept is currently acting as a performance bottleneck.
4-cores active with 1 core/module being used provides better performance than 4 cores with 2 cores/module being used, even with Turbo.
Too bad he was banned for life for cheating last year, oh wait, that got reduced to just a year because of the influence he has amongst HWBOT mods. Just goes to show how much resources can influence the overclocking competition community. Need a good chip buy a truck load, want to win a competition, cheat, get caught, pay to have the sentence reduced. Give me a break. He could break 9Ghz and I wouldn't care, it means nothing. He did nothing for the community to redeem himself.
They have the temp for liquid helium correct, it does boil at 4K (-269°C), but they are totally off about LN2. It's boiling point is -196°C (77K) and according to wiki, it solidifies at -210°C (63K). I don't know where they are getting the -253 from. As far as I was aware, all of these super low temp overclocks are done with evaporative cooling. They essentially just pour the LN2 into a large pipe that is sitting over the CPU and let it boil off if it needs to.
Sorry about that. I was looking at a Finnish science book (has all these characteristics and formulas etc) and mixed nitrogen and hydrogen (they were next to each other). I was also a bit surprised as I remember LN2 being around -195 too.
Thanks for the heads up. Next time I'll be sure to double-check!
The facts are simple, just quite simple: BD with ~2x elements & ~1.5x die size - performance less than SB!!! That's I call BIG FAILURE!!! BD SB NV* Speed 3.6\4.2~16% 3.4\3.8~12% 0.7\1.4\1.8 GHz Max TDP 125 95 50 W Process Technology 32 32 40 nm Approximate Die Size 315 216 116 mm^2 Approximate Transistor count ~2*10^9 995*10^6 585*10^6 # *NV GT530 is only for reference.
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.
44 Comments
Back to Article
XZerg - Thursday, November 3, 2011 - link
Post some real performance numbers too to droll at otherwise just another thrash-talk...Articuno - Thursday, November 3, 2011 - link
Maybe if they get it to 12GHz it'll actually be able to equal the 2500k. /sarcasmtiro_uspsss - Thursday, November 3, 2011 - link
that an AMD CPU should get closer to the 10Ghz that Netburst promised..JarredWalton - Thursday, November 3, 2011 - link
Ah, but NetBurst was from 2000 or so, and it was retired in 2006. I would guess that had Intel kept pushing with NetBurst and transitioned to 45nm and now 32nm, they too could be hitting ~8.5GHz while sucking down a ton of power. Still not sure what AMD was thinking with this long pipeline for Bulldozer. :-\rns.sr71 - Thursday, November 3, 2011 - link
the pipeline is longer than phenom 2, yes, but it isn't any longer than SB.there are just a lot of other issues that are limiting performance.
Arnulf - Friday, November 4, 2011 - link
Any numbers to back this up ? I don't think AMD has spilled the beans on particular details of Bulldozer architecture yet (unlike Intel does and AMD did for their previous architectures).DigitalFreak - Thursday, November 3, 2011 - link
8.58Ghz with only one core and liquid nitrogen. Call me when the system can actually be used for something at that speed.prophet001 - Thursday, November 3, 2011 - link
You can use it at that speed... you just have to do it really fast :-Dtulx - Thursday, November 3, 2011 - link
Two cores.hsew - Thursday, November 3, 2011 - link
One ModuleShining Arcanine - Thursday, November 3, 2011 - link
How does AMD's SMT implementation equal two cores? It is one core.SlyNine - Thursday, November 3, 2011 - link
Because much of the chip logic including the integer core is doubled. Saying one core is further off then saying to cores. More like 1.5 cores.DigitalFreak - Thursday, November 3, 2011 - link
Two cores in marketing terms. No different than saying an Intel SB chip with 4 physical cores + hyper-threading = 8 cores. Of course, Intel doesn't do that.SlyNine - Thursday, November 3, 2011 - link
Oh yea, Way different.N4g4rok - Thursday, November 3, 2011 - link
They aren't the same. Each BD module contains two physical integer cores, as opposed to the two logical cores in SB's hyperthreading. Big difference.Four modules, 2 physical cores a piece = 8 cores.
DigitalFreak - Thursday, November 3, 2011 - link
So at what point does a core stop being a true core?silverblue - Thursday, November 3, 2011 - link
It's an octo-core if you're running integer workloads, and a quad-core for floating point (albeit eight threads).cfaalm - Thursday, November 3, 2011 - link
So what are you doing here?We're counting cores </rainman>
Belard - Friday, November 4, 2011 - link
Who cares? This 8 core AMD is still slower than a dual core intel 2600k.No more waiting... I'm going intel.
Landiepete - Thursday, November 3, 2011 - link
What is it exactly all these experiments are supposed to tell me ? That, yes, the standard performance on bulldozer is significantly slower than on i7, but I can (and am supposed to) easily overclock it to 4GHz or more, thus making it competitive ? Or that I have to treat the extra cores as spare tires in case I get a flat ?I simply fail to see the real world significance. Please enlighten me.
Kristian Vättö - Thursday, November 3, 2011 - link
Well, I don't think there is any real world significance. Only one Bulldozer module (i.e. two "cores") was enabled, so this was just aiming for as high frequency as possible. With all cores enabled, you will be looking at much lower frequencies.Basically, this is one kind of a sport. We don't benefit from it but a world record is a world record. You don't benefit if someone breaks the 100m sprint record, do you? ;-)
r3loaded - Thursday, November 3, 2011 - link
There is no real-work significance. All we learn from this is that Bulldozer can be clocked really high in theory (much like NetBurst). It still sucks in normal usage, even overclocked under water cooling against Sandy Bridge.Targon - Thursday, November 3, 2011 - link
I agree about these experiments, but it has been shown that the performance issues are due to how Windows is currently assigning threads to the cores. It's a scheduler problem as much as it is a design issue.Basically, if Windows(or any other OS) is set to assign threads to one core per module as the way to assign threads, you get higher performance than if you load up core 1, 2(module 1), then 3 and 4(module 2), leaving the remaining two modules unused. The whole "module" approach is a big part of the performance problem. Some have even wondered if you force the Core i7 scheduler to be used instead of the normal "default" scheduler if that would work better with Bulldozer at this point.
Assimilator87 - Thursday, November 3, 2011 - link
Jeez, what's with all the negativity? While the max overclock isn't 24/7 stable, it IS an indicator of how well the architecture overclocks in general, at all levels of cooling. It could help someone avoid a really terrible line of processors. For example, if I had done some research on the original B2 Phenom's overclockability, I would have steered away, but unfortunately that was not the case.Targon - Friday, November 4, 2011 - link
I am looking at this from a rational perspective, not being negative. Clock speed, as seen in the days of the Pentium 4 vs. Athlon 64 means less than performance when running applications.I am an AMD fan, but when it comes right down to it, if Windows 7 by default shows that an 8-core Bulldozer processor doesn't win hands down compared to a Phenom 2 at the same clock speed, that shows that there are problems. The CORE design of Bulldozer is supposed to be better, clock for clock compared to the previous generation, and we are not seeing that. A 4-core processor based on a Bulldozer core won't perform as well clock for clock compared to a 4-core Phenom 2 due to modules.
Kristian Vättö - Thursday, November 3, 2011 - link
I think that is AMD's propaganda. While W8 is still in pre-beta state, the tests have shown that it won't really improve the performance. Of course, it may improve later on as W8 develops.However, what is the point? Intel will have IB out before W8 and AMD should also be out with Piledriver soon.
http://www.tomshardware.com/reviews/fx-8150-zambez...
piroroadkill - Thursday, November 3, 2011 - link
The exact opposite is true according to AMD, because if you leave two modules asleep, they get power gated off, turbo kicks in, and the two modules you're loading get a much higher clock speed.Targon - Friday, November 4, 2011 - link
Clock speed does NOT equal performance, which is why so many people are posting negative comments about the performance of Bulldozer. Better performance per clock means that Turbo Core is a nice idea, but the "module" concept is currently acting as a performance bottleneck.4-cores active with 1 core/module being used provides better performance than 4 cores with 2 cores/module being used, even with Turbo.
Targon - Friday, November 4, 2011 - link
I found the article I was thinking of that you might find interesting:http://techreport.com/articles.x/21865
JMC2000 - Thursday, November 3, 2011 - link
I'm waiting for someone to clock BD at 9GHz, just for the eventual flood of "OVER 9000!" memes...utnorris - Thursday, November 3, 2011 - link
Too bad he was banned for life for cheating last year, oh wait, that got reduced to just a year because of the influence he has amongst HWBOT mods. Just goes to show how much resources can influence the overclocking competition community. Need a good chip buy a truck load, want to win a competition, cheat, get caught, pay to have the sentence reduced. Give me a break. He could break 9Ghz and I wouldn't care, it means nothing. He did nothing for the community to redeem himself.QChronoD - Thursday, November 3, 2011 - link
They have the temp for liquid helium correct, it does boil at 4K (-269°C), but they are totally off about LN2. It's boiling point is -196°C (77K) and according to wiki, it solidifies at -210°C (63K). I don't know where they are getting the -253 from.As far as I was aware, all of these super low temp overclocks are done with evaporative cooling. They essentially just pour the LN2 into a large pipe that is sitting over the CPU and let it boil off if it needs to.
MrSpadge - Thursday, November 3, 2011 - link
Agreed. Glad someone else noticed it!Kristian Vättö - Thursday, November 3, 2011 - link
Sorry about that. I was looking at a Finnish science book (has all these characteristics and formulas etc) and mixed nitrogen and hydrogen (they were next to each other). I was also a bit surprised as I remember LN2 being around -195 too.Thanks for the heads up. Next time I'll be sure to double-check!
csimon - Thursday, November 3, 2011 - link
All of this overclocking in the wake of an AMD 10% layoff?http://tech.icrontic.com/news/the-axe-falls-at-amd...
fic2 - Thursday, November 3, 2011 - link
AMD is the overclocking king.Intel is the IPC/compute/low power king (i.e. the things that actually matter).
TC2 - Thursday, November 3, 2011 - link
The facts are simple, just quite simple:BD with ~2x elements & ~1.5x die size - performance less than SB!!!
That's I call BIG FAILURE!!!
BD SB NV*
Speed 3.6\4.2~16% 3.4\3.8~12% 0.7\1.4\1.8 GHz
Max TDP 125 95 50 W
Process Technology 32 32 40 nm
Approximate Die Size 315 216 116 mm^2
Approximate Transistor count ~2*10^9 995*10^6 585*10^6 #
*NV GT530 is only for reference.
Ronakbhai - Thursday, November 3, 2011 - link
*yawn*shbdf - Friday, November 4, 2011 - link
New era cap $12Bikini (Ed hardy,polo) $20
accept paypal and free shipping
== ( http://www.amtata.com )====
== ( http://www.amtata.com )====
fdfgsfdf - Friday, November 4, 2011 - link
= ( http://www.hapous.com )====== ( http://www.hapous.com )====
== ( http://www.hapous.com )====
SanX - Sunday, November 6, 2011 - link
Can anyone run scientific parallel algebra test on Bulldozer?Take 32-bit benchmark for Windows compiled with Intel Fortran at
http://www.equation.com/servlet/equation.cmd?fa=la...
Interesting is to compare it to Phenom/Intel i7 done in Anandtech comments here
http://www.anandtech.com/Show/Index/4481?cPage=5&a...
dfsd - Sunday, November 6, 2011 - link
http://www.hapous.com/dfsd - Sunday, November 6, 2011 - link
http://www.hapous.comfdfsxcvdcfdh - Monday, November 7, 2011 - link
ONLINE STOR good shopping