News [Toms] Beyond Rome: AMD's EPYC and Radeon to Power World's Fastest Exascale Supercomputer

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Thunder 57

Senior member
Aug 19, 2007
723
226
136
#26
Why would AMD have a larger chance of being unable to deliver? Their track record since 2017 has been nearly spotless, at least on the CPU side. On the GPU side, things are a bigger unknown, though they came through for PS5/Xbox 2(?) so I would not dismiss their ability to get hardware ready for a specific client. If they have neglected anyone, it's the consumer PC market. Now that AMD is using TSMC for nearly everything, they'll be ready on the fab side.

Intel has absolutely no recent record on delivering on the dGPU front. They did have to cancel an entire uarch upgrade (Cannonlake) on the CPU side, along with Gen10 graphics, so that certainly isn't good. Intel could improve their prospects by using an outside fab for Xe. We still haven't seen them deliver a working dGPU product, though.



NV was always an option. POWER/NV wound up in Summit, after all.
I saw the POWER/NVIDIA combo so that's why I don't get the whole AMD was their only option nonsense.

Also, as far as being unable to deliver, well according to Tomshardware Intel already has had a bit of an issue regarding that:

"Intel’s followup Knights Hill products, which never came to market, have borne the blame for the delay of the Aurora supercomputer, which will now feature Intel’s Xe Graphics Architecture. "
 

Thunder 57

Senior member
Aug 19, 2007
723
226
136
#27
AMD can lose priority at TSMC, which can lead to them being unable to deliver. Which was why AMD went to GlobalFoundries exclusively. No fellow customers means they would be the only customer.

Semi-custom however has extra weight in TSMC's eyes.
That seems highly unlikely. AMD is giving them a ton of business. With smartphone sales in decline, it's not like Apple would or needs to pay them untold billions for fab space.

AMD went to GloFo because of their unique relationship/history and the WSA.
 

jpiniero

Diamond Member
Oct 1, 2010
6,500
295
126
#28
Lol, sure. There’s a much larger chance AMD won’t be able to deliver. As a matter of fact I say it’s 100x more likely AMD will fail than intel.

There, I made a statement just as valid as yours, based upon the same amount of information.

Anyone can make stuff up.
If you think about it, Intel's already failed twice basically. Aurora was supposed to be using Xeon Phi, then perhaps an AP line product. Now it's Xe.

The issue is more 10 nm, that is if that's where Xe is being fabbed.
 
Mar 13, 2006
10,129
130
126
#29
That seems highly unlikely. AMD is giving them a ton of business. With smartphone sales in decline, it's not like Apple would or needs to pay them untold billions for fab space.

AMD went to GloFo because of their unique relationship/history and the WSA.
Revenue from AMD is a single digit percentage of TSMCs sales. It’s not very difficult to look this stuff up.
 
Mar 13, 2006
10,129
130
126
#30
If you think about it, Intel's already failed twice basically. Aurora was supposed to be using Xeon Phi, then perhaps an AP line product. Now it's Xe.

The issue is more 10 nm, that is if that's where Xe is being fabbed.
When is Purch renaming these the AMD zone forums? It’s all that’s left of them.




Your incessant trolling of AMD threads is going to stop.
One way or the other.


esquared
Anandtech Forum Director
 
Last edited by a moderator:

Thunder 57

Senior member
Aug 19, 2007
723
226
136
#31
Revenue from AMD is a single digit percentage of TSMCs sales. It’s not very difficult to look this stuff up.
So glad you provided a link then, because surely you looked it up just now. Also, AMD just switched to TSMC as their primary partner. What happens when Zen 2 sells like hotcakes?
 

JDG1980

Golden Member
Jul 18, 2013
1,657
172
136
#32
AMD can lose priority at TSMC, which can lead to them being unable to deliver.
TSMC is a pure-play foundry. This means that their business model is dependent on third parties being willing to trust them to fab chips reliably and on time. Arbitrarily having one customer "lose priority" would jeopardize TSMC's standing with other customers as well.

Now, what potentially *can* happen is that Zen 2 (or subsequent) sales are significantly more than AMD expected, and all the existing capacity is already allocated at TSMC for a period of time so there would be a lag before they could further increase supply to meet demand. In that case, AMD would presumably use their wafers to fab the highest margin chips, and reduce production of low margin parts until more fab capacity opens. But owning your own fab doesn't prevent against this kind of thing happening - it is happening to Intel right now with regards to 14nm; there have been shortages for months.

Semi-custom however has extra weight in TSMC's eyes.
Nonsense. Why would TSMC care?
 

NostaSeronx

Platinum Member
Sep 18, 2011
2,385
166
126
#33
Arbitrarily having one customer "lose priority" would jeopardize TSMC's standing with other customers as well.
Nonsense. Why would TSMC care?
AMD is already in TSMC's bad graces. However, the semi-custom partners Google, Microsoft, Sony, and etc are not in TSMC's bad graces. So, Semi-custom is safe while official mainline is back to previous policy within TSMC. Pure TSMC products or no benefits, capisce?!

If a company fabs at Samsung, UMC, HLMC, SMIC, and or GlobalFoundries.
=> No benefits, no priority, no price cuts, etc.

As far as my ROC contact goes... AMD is not on the Loyalty N7 node.
 
Last edited:

Thunder 57

Senior member
Aug 19, 2007
723
226
136
#34
TSMC is a pure-play foundry. This means that their business model is dependent on third parties being willing to trust them to fab chips reliably and on time. Arbitrarily having one customer "lose priority" would jeopardize TSMC's standing with other customers as well.

Now, what potentially *can* happen is that Zen 2 (or subsequent) sales are significantly more than AMD expected, and all the existing capacity is already allocated at TSMC for a period of time so there would be a lag before they could further increase supply to meet demand. In that case, AMD would presumably use their wafers to fab the highest margin chips, and reduce production of low margin parts until more fab capacity opens. But owning your own fab doesn't prevent against this kind of thing happening - it is happening to Intel right now with regards to 14nm; there have been shortages for months.



Nonsense. Why would TSMC care?
Thank you for explaining better than I could.

AMD is already in TSMC's bad graces...
Says who, you? Even if that were true, I don't think TSMC would be overly upset with making money off of AMD. Also, what are Google, Microsoft, Sony and others have fabbed at TSMC on 7nm, or any process for that matter? I expect a hard to interpret non-reply, as you seem to believe you know what every fab is up to and never back up your claims.
 

Dayman1225

Senior member
Aug 14, 2017
904
90
96
#35
If you think about it, Intel's already failed twice basically. Aurora was supposed to be using Xeon Phi, then perhaps an AP line product. Now it's Xe.

The issue is more 10 nm, that is if that's where Xe is being fabbed.
Aurora was never going to use -AP
 
Jan 26, 2004
76
40
91
#36
Yeah, it isn't Genoa based after all, I guess.
Just because its "custom" doesnt mean that it doesnt use the same architecture, could be even same chips.Microsoft uses custom, playstation uses custom.Still, its the same architecture.This chinese console uses custom etc.
 
Apr 27, 2000
12,365
1,320
126
#37
"Intel’s followup Knights Hill products, which never came to market, have borne the blame for the delay of the Aurora supercomputer, which will now feature Intel’s Xe Graphics Architecture. "
Ouch. No way is Intel going to rely on 10nm or 7nm for their dGPUs. They're going to have to hit up an outside fab to get the parts delivered.

Aurora was never going to use -AP
One must wonder who does want to use -AP.
 

moinmoin

Senior member
Jun 1, 2017
849
363
106
#38
AMD is to work together with Cray on the by far fastest and most expensive supercomputer, and somebody claims TSMC essentially wants no part of that? AMD finally has a competitive x86 design, Intel fails with their new node, and TSMC wouldn't prioritize AMD on their HP 7nm node (that by all reports is far from at full capacity) to show the world who's the leading foundry now? Don't make me laugh.

One must wonder who does want to use -AP.
Isn't it essentially a limited quantity product to show they have something, just like Cannon Lake?
 
Apr 27, 2000
12,365
1,320
126
#39
Isn't it essentially a limited quantity product to show they have something, just like Cannon Lake?
Honestly, I don't know how limited in quantity it will be. As with all things Xeon, a significant number of these chips were likely made available to the usual suspects before anyone in the public could get ahold of them. Did anyone actually commit to buy the things? Charlie seemed to think that nobody much wanted them. And Intel won't even sell the chips individually, or even in BGA format. You have to buy an entire chassis. I can't see ODM customers much liking that arrangement. Regardless, I do not think I would want to try to build a supercomputer using Cascade Lake-AP. Maybe Intel should have used contrarevenue to get people to buy them!
 
Apr 27, 2000
12,365
1,320
126
#41
Ouch. No way is Intel going to rely on 10nm or 7nm for their dGPUs. They're going to have to hit up an outside fab to get the parts delivered.
I am quoting myself to amend this statement. It looks like Intel intends to use their 7nm process for datacenter Xe in 2021. Which is interesting, and somewhat unexpected. They're going to use Xe as a pipecleaner for 7nm. Reminds me of AMD in 2018/2019 with Vega20.
 

moinmoin

Senior member
Jun 1, 2017
849
363
106
#42
AMD finally has a competitive x86 design, Intel fails with their new node, and TSMC wouldn't prioritize AMD on their HP 7nm node (that by all reports is far from at full capacity) to show the world who's the leading foundry now?
And Intel was even teasing with their "process leadership" again as we were discussing that. Don't know about you lot, but I'm looking forward to Intel and TSMC fighting for their "process leadership" in desktop x86 CPUs.
 

itsmydamnation

Golden Member
Feb 6, 2011
1,840
223
136
#43
Remember there is a chance amd could be 14nm limited not 7nm limited if things are selling like hot cakes, with depressed phone sales it seems TMSC has capacity, the 14nm i/o dies arent small and GF does have limited capacity.
 

NostaSeronx

Platinum Member
Sep 18, 2011
2,385
166
126
#44
The only way the common chiplet production can be impacted is from;
- KAIST/KIPB include GlobalFoundries into Round 2.
- GlobalFoundries cans 14LPP(dual-gate)/12LP(tri-gate)... since it doesn't follow the planar road; 55LPx -> 40LP -> 28SLP -> 22FDX -> 12FDX -> 7FDX
- AMD doesn't move to another GlobalFoundries process. If there is no FinFET, then there is no barrier for FDSOI to be funded for big dies.
 
Feb 29, 2008
42
37
91
#46
AMD is to work together with Cray on the by far fastest and most expensive supercomputer, and somebody claims TSMC essentially wants no part of that? AMD finally has a competitive x86 design, Intel fails with their new node, and TSMC wouldn't prioritize AMD on their HP 7nm node (that by all reports is far from at full capacity) to show the world who's the leading foundry now? Don't make me laugh.
Isn't it essentially a limited quantity product to show they have something, just like Cannon Lake?
Thumbs Up for one of the most spot-on posts so far, well, actually among a fair number.

Regardless of whether or not AMD was on TSMC's Christmas card list before vis-a-vis AMD's lock-in with GloFo, TSMC has definitely welcomed AMD into the 'club' because with AMD it now has a foothold in the x86 market. Being able to crow about having the best ARM main-liner Apple, AND a not-so-secret Giant Killer AMD onboard has got to be quite heady days for many at TSMC.
The bragging rights as much as the money are important, especially in the East.
Even thinking about knocking AMD's allocations about because they have the temerity to have a back-up plan in place is borderline grade-school mentatility.
If they ever did do such, all they would do is shoot themselves in the foot as every one of their other customers would recognize it for what it was and make haste to get their own backup fabber in place. Like Samsung, GloFlo, etc. Every other fabber would love to see TSMC do this.
 
Feb 6, 2011
1,840
223
136
#47
Thumbs Up for one of the most spot-on posts so far, well, actually among a fair number.

Regardless of whether or not AMD was on TSMC's Christmas card list before vis-a-vis AMD's lock-in with GloFo, TSMC has definitely welcomed AMD into the 'club' because with AMD it now has a foothold in the x86 market. Being able to crow about having the best ARM main-liner Apple, AND a not-so-secret Giant Killer AMD onboard has got to be quite heady days for many at TSMC.
The bragging rights as much as the money are important, especially in the East.
Even thinking about knocking AMD's allocations about because they have the temerity to have a back-up plan in place is borderline grade-school mentatility.
If they ever did do such, all they would do is shoot themselves in the foot as every one of their other customers would recognize it for what it was and make haste to get their own backup fabber in place. Like Samsung, GloFlo, etc. Every other fabber would love to see TSMC do this.
You just don't understand..... everything Nosta says comes to pass, he is like the jesus of the internet semiconductor world!
 
Apr 27, 2000
12,365
1,320
126
#48
You just don't understand..... everything Nosta says comes to pass, he is like the jesus of the internet semiconductor world!
Juangra has competition. Heaven help us if they worked together. It would be worse than Thanos.
 

moinmoin

Senior member
Jun 1, 2017
849
363
106
#49
In related news, HP buys Cray. I can just hope Cray influences HP and not the other way.
 


ASK THE COMMUNITY