I guess the public will never find out why the monolithic variant of Cooper Lake ("SP" not "AP") was to have 6 UPI ports. It would have made a great 8-way server. (And Ice Lake SP is rumored to only have 4 UPI.)
The impression seems to have been that the 8S monolithic Cooper Lake (SP) is the one that remains for those specific customers, while the multi-die high-core-count Cooper Lake AP was meant to drop into the Whitley platform with Ice Lake SP, as a 2S product.
I don't believe it's there & disabled. If Ice Lake had it in silicon, then they'd have too much to gain by enabling it to leave it dark.
Now, maybe with all of the delays, they've found time for a new spin of Ice Lake that has it... which could be the real reason for cancelling Cooper Lake.
We will see how it stacks up when the new 10nm+ node which is baked in Tigerlake comes out. I am not expecting huge improvements, given it is a complicated process with lots of patterning.
so 10nm / ice lake is supposed to save intel ?? but semi accurate also reported several planned 10nm fabs were re purposed and that they will have very little 10nm capacity at a time they are forced to increase core counts.
is losing market share to samsung/TSMC worse than losing it to them with AMD designs?? i am surprised they are not outsourcing more to other fabs then if these reports are true
Ice lake is not supposed to save Intel. Firstly, Intel doesn't need to be saved at the moment, though disruption because of the virus may change things.
But ice lake is just one step in intel trying to right the ship. They have a faster cadence of meaningful imprivements of products coming out and by the time they are on 7 nm they should be in better shape. They'll likely lose market share in servers but that was bound to happen anyway. Their troubles have accelerated that.
From a related SA article: "Because Intel’s 10nm process is abjectly broken as SemiAccurate exclusively reported months ago, the Tick-Tock process became Tick-Tock-Sproing with the emergency addition of the Kaby Lake rehashes and Tick-Tock-Spoing-Thud with Coffee."
Well at least INtel and Facebook have something in common: they both have fast declining market share!
Cooper Lake moved to a none mass product because not a single large OEM wanted to redesign there volume products to It. HPE, DELL, Lenovo, Cisco were all reluctant to design for it because of the uncertain roadmap and lack of any vision in portfolio by Intel. THey could not bring a dual cpu type strategy into single product lines. All the answers of OEM are the same. If you stick with INtel go now for the R line and just stay there and dont even move to IceLAke if you dont have to.
About Ice LAke its delayed and delayed, it does not start volume production before this summer so its a Very very late this year volume shipping. Lets hope for intel they dont have to delay again and not a paper launch again… it will be 38 ish cores with XCC rated already as 230+W TDP it aint going to be enough to hold against competition.
Why paper launch well cascade lake R line is another fine market strategy. Announce lower pricing higher core count to the market to kill your own skylake - cooperlake portfolio and try to move the focus away from AMD.... but pls wait for 2 months untill we can deliver the product....
Offcourse all this because of very very high demand… sure intel we believe you. Nice joke. It is because you have to deliver high core count parts only which is bad for the 14nm +++++++ volume and 10nm is still horrible. ASP with R line will go DOWN DOWN DOWN.
I think this is for the good, if the mindset inside Intel has changed. They need to stop refreshing 14nm parts and move on. There are quite a few scores on Geekbench with the new Ice Lake server platform and they are quite good. Sure, Ice Lake won't be a resounding success, it is late, it is built on a bad process, etc. But they simply need to focus the best people they got and get 7nm done, build a solid foundation for it and start working on moving your existing IP to 7nm.
If I were Intel, I would bail on 10nm and focus on 7nm EUV. Use existing 14nm parts for now, work out a 7nm optimized design, and focus on delivering 7nm ASAP. Even if they can’t hit density targets with 7nm, it’s worth killing 10nm as it has been nothing but a plague.
But how can they move on? Cooper Lake is canceled. Kaby Lake was end of life'd a couple weeks ago. What Xeon is still actually being produced? It looks like they've just straight up canceled production of Xeon's.
With 10nm still a no show it looks like they just abandoned the server market.
Of course they were. How else do you play canceling the majority of your Xeon products except for 4 and 8 xeon platforms? This decision is just boggling unless they got really strong pushback from the OEM's and didn't have a choice.
> As far as we understand, there was one big customer for bfloat16: Facebook.
The irony of this is that Google actually pioneered the data type. But, I guess google is using their own TPUs for machine learning, so that's probably why they're not requiring it from the CPU.
For those who are Easily-BeDazzled-By/"Fanbois-of" the likes of "ChipZilla", the ONLY reason that guy is "At-His Job" is because "ChipZilla" wishes to "Make-A Living". Namely, that of profitting through Mesmerism/"A-Killing". 'Nuff-Said 'cos The-Mods are trying to "Completely-Eradicate" that looney and 'is posting. Ja? Good-Byeee ....
Be aware that True-Awarenesss is using The-True-Heart and lack "Thought-Processing" [TP] elements [as opposed to "Aggressively-'Finking'-About-Eet" via Mental-Deliberation, for example, something "Thinkers" call being {Mis}Guided by TP, "The-Evil" element within self/The-Individual-Thinker].
"Customers moving from old Broadwell Xeon (Q1 2016) systems or Skylake Xeon (Q3 2017), who normally have a 3-5 year update cycle, are desperately looking at something to update to."
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.
38 Comments
Back to Article
Elstar - Tuesday, March 17, 2020 - link
I guess the public will never find out why the monolithic variant of Cooper Lake ("SP" not "AP") was to have 6 UPI ports. It would have made a great 8-way server. (And Ice Lake SP is rumored to only have 4 UPI.)anonomouse - Tuesday, March 17, 2020 - link
The impression seems to have been that the 8S monolithic Cooper Lake (SP) is the one that remains for those specific customers, while the multi-die high-core-count Cooper Lake AP was meant to drop into the Whitley platform with Ice Lake SP, as a 2S product.dersteffeneilers - Wednesday, March 18, 2020 - link
How does a round number of UPI ports make sense? If you have a 4-socket mobo, you need CPUs with 3 ports to use all of them.Elstar - Wednesday, March 18, 2020 - link
I’d imagine that with a quad processor motherboard, the extra UPI port on Ice Lake would either go unused to be attached to an FPGA accelerator.JayNor - Tuesday, March 17, 2020 - link
Any word on bfloat16 or dual avx512 going into the Ice Lake Server chips?Any word on a Cooper Lake XE or an Ice Lake XE?
Ian Cutress - Tuesday, March 17, 2020 - link
First question is covered in the article.FreckledTrout - Tuesday, March 17, 2020 - link
It wasn't super clear to me that bfloat16 was going to make it into IceLake.Ian Cutress - Wednesday, March 18, 2020 - link
It isn't.FreckledTrout - Wednesday, March 18, 2020 - link
Well that is interesting. Intel is a mess right now.yeeeeman - Wednesday, March 18, 2020 - link
Officially no.mode_13h - Thursday, March 19, 2020 - link
I don't believe it's there & disabled. If Ice Lake had it in silicon, then they'd have too much to gain by enabling it to leave it dark.Now, maybe with all of the delays, they've found time for a new spin of Ice Lake that has it... which could be the real reason for cancelling Cooper Lake.
ksec - Tuesday, March 17, 2020 - link
Still nothing to suggest 10nm is on track and good enough for Server and higher end Laptop. Let just hope AMD deliver Milan on time.yeeeeman - Wednesday, March 18, 2020 - link
We will see how it stacks up when the new 10nm+ node which is baked in Tigerlake comes out. I am not expecting huge improvements, given it is a complicated process with lots of patterning.remosito - Wednesday, March 18, 2020 - link
let's hope in addition amd will finally tackle the low core high freq niche...azfacea - Tuesday, March 17, 2020 - link
so 10nm / ice lake is supposed to save intel ?? but semi accurate also reported several planned 10nm fabs were re purposed and that they will have very little 10nm capacity at a time they are forced to increase core counts.is losing market share to samsung/TSMC worse than losing it to them with AMD designs?? i am surprised they are not outsourcing more to other fabs then if these reports are true
Yojimbo - Saturday, March 21, 2020 - link
Ice lake is not supposed to save Intel. Firstly, Intel doesn't need to be saved at the moment, though disruption because of the virus may change things.But ice lake is just one step in intel trying to right the ship. They have a faster cadence of meaningful imprivements of products coming out and by the time they are on 7 nm they should be in better shape. They'll likely lose market share in servers but that was bound to happen anyway. Their troubles have accelerated that.
boozed - Tuesday, March 17, 2020 - link
From a related SA article: "Because Intel’s 10nm process is abjectly broken as SemiAccurate exclusively reported months ago, the Tick-Tock process became Tick-Tock-Sproing with the emergency addition of the Kaby Lake rehashes and Tick-Tock-Spoing-Thud with Coffee."Hahaha
pepoluan - Tuesday, March 17, 2020 - link
I always love S|A's colorful use of language :DThat said, Charlie was shaking his head really hard as he commented incredulously how Intel spun the quenching of Cooper Lake...
(He also have juicy tidbits about Intel having problems with Aurora... I have to research about this.)
casperes1996 - Tuesday, March 17, 2020 - link
Love that video. "You can say it for me" is a perfect way of saying yes without technically saying itVitor - Tuesday, March 17, 2020 - link
Quite crazy how much momentum Intel seems to be losing.yeeeeman - Wednesday, March 18, 2020 - link
Quite crazy how important a process can be for a company. Still, I think Intel internal policies have made this situation much much worse.Machinus - Tuesday, March 17, 2020 - link
Dusted? what are you talking aboutCasper42 - Wednesday, March 18, 2020 - link
GoneFinito
Dead
Not going to happen.
There average Joe Server Admin won't get Cooper Lake from Dell or HPE.
duploxxx - Wednesday, March 18, 2020 - link
Well at least INtel and Facebook have something in common: they both have fast declining market share!Cooper Lake moved to a none mass product because not a single large OEM wanted to redesign there volume products to It. HPE, DELL, Lenovo, Cisco were all reluctant to design for it because of the uncertain roadmap and lack of any vision in portfolio by Intel. THey could not bring a dual cpu type strategy into single product lines. All the answers of OEM are the same. If you stick with INtel go now for the R line and just stay there and dont even move to IceLAke if you dont have to.
About Ice LAke its delayed and delayed, it does not start volume production before this summer so its a Very very late this year volume shipping. Lets hope for intel they dont have to delay again and not a paper launch again… it will be 38 ish cores with XCC rated already as 230+W TDP it aint going to be enough to hold against competition.
Why paper launch well cascade lake R line is another fine market strategy. Announce lower pricing higher core count to the market to kill your own skylake - cooperlake portfolio and try to move the focus away from AMD.... but pls wait for 2 months untill we can deliver the product....
Offcourse all this because of very very high demand… sure intel we believe you. Nice joke. It is because you have to deliver high core count parts only which is bad for the 14nm +++++++ volume and 10nm is still horrible. ASP with R line will go DOWN DOWN DOWN.
yeeeeman - Wednesday, March 18, 2020 - link
I think this is for the good, if the mindset inside Intel has changed. They need to stop refreshing 14nm parts and move on. There are quite a few scores on Geekbench with the new Ice Lake server platform and they are quite good. Sure, Ice Lake won't be a resounding success, it is late, it is built on a bad process, etc. But they simply need to focus the best people they got and get 7nm done, build a solid foundation for it and start working on moving your existing IP to 7nm.supdawgwtfd - Wednesday, March 18, 2020 - link
I thought they were skipping 7nm and going straight to 5???goatfajitas - Wednesday, March 18, 2020 - link
I am pretty sure everything after Skylake wasn't supposed to exist.eek2121 - Wednesday, March 18, 2020 - link
If I were Intel, I would bail on 10nm and focus on 7nm EUV. Use existing 14nm parts for now, work out a 7nm optimized design, and focus on delivering 7nm ASAP. Even if they can’t hit density targets with 7nm, it’s worth killing 10nm as it has been nothing but a plague.cheshirster - Wednesday, March 18, 2020 - link
https://www.youtube.com/watch?v=pTZaNHZGsQorahvin - Wednesday, March 18, 2020 - link
But how can they move on? Cooper Lake is canceled. Kaby Lake was end of life'd a couple weeks ago. What Xeon is still actually being produced? It looks like they've just straight up canceled production of Xeon's.With 10nm still a no show it looks like they just abandoned the server market.
mode_13h - Thursday, March 19, 2020 - link
Cascade Lake was just refreshed.rahvin - Wednesday, March 18, 2020 - link
Of course they were. How else do you play canceling the majority of your Xeon products except for 4 and 8 xeon platforms? This decision is just boggling unless they got really strong pushback from the OEM's and didn't have a choice.mode_13h - Wednesday, March 18, 2020 - link
> As far as we understand, there was one big customer for bfloat16: Facebook.The irony of this is that Google actually pioneered the data type. But, I guess google is using their own TPUs for machine learning, so that's probably why they're not requiring it from the CPU.
vishwa108 - Sunday, March 22, 2020 - link
For those who are Easily-BeDazzled-By/"Fanbois-of" the likes of "ChipZilla", the ONLY reason that guy is "At-His Job" is because "ChipZilla" wishes to "Make-A Living". Namely, that of profitting through Mesmerism/"A-Killing". 'Nuff-Said 'cos The-Mods are trying to "Completely-Eradicate" that looney and 'is posting. Ja? Good-Byeee ....Be aware that True-Awarenesss is using The-True-Heart and lack "Thought-Processing" [TP] elements [as opposed to "Aggressively-'Finking'-About-Eet" via Mental-Deliberation, for example, something "Thinkers" call being {Mis}Guided by TP, "The-Evil" element within self/The-Individual-Thinker].
Qasar - Sunday, March 22, 2020 - link
wth ???peevee - Monday, March 23, 2020 - link
"<sign> * 1 + <fraction> * 2<exponent>"If it would be so numbers below -1 could not be represented.
peevee - Monday, March 23, 2020 - link
"Customers moving from old Broadwell Xeon (Q1 2016) systems or Skylake Xeon (Q3 2017), who normally have a 3-5 year update cycle, are desperately looking at something to update to."Why desperately when EPIC is out there?
melgross - Monday, March 30, 2020 - link
“Keeping the accuracy close to zero”. Well, that’s a goal all computers should have, zero accuracy. How difficult is it to achieve that?