Question 'Ampere'/Next-gen gaming uarch speculation thread

Page 120 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Ottonomous

Senior member
May 15, 2014
559
292
136
How much is the Samsung 7nm EUV process expected to provide in terms of gains?
How will the RTX components be scaled/developed?
Any major architectural enhancements expected?
Will VRAM be bumped to 16/12/12 for the top three?
Will there be further fragmentation in the lineup? (Keeping turing at cheaper prices, while offering 'beefed up RTX' options at the top?)
Will the top card be capable of >4K60, at least 90?
Would Nvidia ever consider an HBM implementation in the gaming lineup?
Will Nvidia introduce new proprietary technologies again?

Sorry if imprudent/uncalled for, just interested in the forum member's thoughts.
 

JasonLD

Senior member
Aug 22, 2017
485
445
136
That's kind of the point.

I kinda don't see the need to be doing that. Node itself is never the central focus of marketing. If it performs well, it doesn't matter if it is based on 8nm or 7nm.
Nvidia never had problem marketing their 12nm based Super GPUs against 7nm Navi. Performance do the most talking, not the node it is based on.
Neither Intel, AMD, or Nvidia ever tried to create confusion simply because the competitor has better node, they will just stay away from that to be the marketing point. People only interested on what node it is based on is people who are posting technical forums like us.
 
Last edited:

DXDiag

Member
Nov 12, 2017
165
121
116
nVidia did this with Pascal - GP107 and GP108 were fabbed at Samsung 14 nm while GP102/GP104/GP106 were fabbed at TSMC 16 nm.
Completely different situation, 14nm and 16nm back then were very close, better yet NVIDIA chose Samsung for low end chips, and TSMC for the high end stuff, if you are basing this off Pascal, then you should be sure that high end Ampere is from TSMC, and low end is from Samsung.
 

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
Completely different situation, 14nm and 16nm back then were very close, better yet NVIDIA chose Samsung for low end chips, and TSMC for the high end stuff, if you are basing this off Pascal, then you should be sure that high end Ampere is from TSMC, and low end is from Samsung.

The original rumor was that they were fabbing GA101-GA106 on SS7 and 107/108 on SS8.
 

Konan

Senior member
Jul 28, 2017
360
291
106
Couldn't Vega pull upwards of 450 with unlocked power? 500 W is believable with ocing. Not sure if an AIB would release a card with 500w tdp though

Sent from my SM-N975F using Tapatalk

The 350W of the 3090 FE becomes 420 watts with the OC slider set to + 20%. I don't think anyone truly thinks a consumer card in normal operation is going to pull 500W.....

The whole 500W thing is just AIB marketing bollocks.. Example: "Our triple fan design can cool great up to 500"


1598820942980.png
 
  • Like
Reactions: lightmanek

CastleBravo

Member
Dec 6, 2019
119
271
96
The 350W of the 3090 FE becomes 420 watts with the OC slider set to + 20%. I don't think anyone truly thinks a consumer card in normal operation is going to pull 500W.....

The whole 500W thing is just AIB marketing bollocks.. Example: "Our triple fan design can cool great up to 500"


View attachment 28908


We should finally be able to implement the ultimate form of temporal anti-aliasing once the cards reach 1.21 gigawatts.
 

Stuka87

Diamond Member
Dec 10, 2010
6,240
2,559
136
I think the 500W thing is coming from it having three 8 pin PCI-E power connections. So in theory it could use that, but I doubt it would. But if it uses 400W, it would still require the three power connectors.
 
  • Like
Reactions: lightmanek

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
I think if it is based on Samsung 8nm, they will say it is 8mm. If it is completely custom variant of 8nm and they decided to call it something else, most likely they won't call it 7nm. Samsung actually has 7nm node which is completely different from their 8nm and will only create unnecessary confusion.
There is still a good chance it is actually based on TSMC 7nm so I wouldn't say what you said for sure without knowing the architecture details.
7 nm LPP process appears to be slightly tuned 8 nm LPP process, and appears to have 7% higher performance than 8 nm LPP.

Its big enough difference, that I do not see a reason why Samsung cannot call it 7 nm LPP, for marketing purposes, even if it still a tweaked 8 nm LPP.

I never argued, that Nvidia is lying here. I was only interested in the truth, and the truth appears to be that it is 7 nm LPP.

So this 7 nm LPP appears to be the same process that Exynos 9825 was made on. And as TechInsights claim, it is slightly tuned 8 nm LPP process.

So technically. If we go by marketing standards. Nvidia is delivering first EUV GPU :D.
 
Last edited:

JasonLD

Senior member
Aug 22, 2017
485
445
136
7 nm LPP process appears to be slightly tuned 8 nm LPP process, and appears to have 7% higher performance than 8 nm LPP.

Its big enough difference, that I do not see a reason why Samsung cannot call it 7 nm LPP, for marketing purposes, even if it still a tweaked 8 nm LPP.

I never argued, that Nvidia is lying here. I was only interested in the truth, and the truth appears to be that it is 7 nm LPP.

I don't think that is enough evidence that 7nm LPP is just slight tuned 8nm LPP process. Exynos 9825 just seems like a quick port job from their existing 8nm Exynos 9820 just so that they can say they released their first 7nm on Q3 2019.
We got proper 7nm EUV on Exynos 990 which seems to be designed from ground up to be on 7nm. Which explains the density difference.

Samsung's 7nm LPP EUV technical info has been known for a long time, and it is clear that it is half-node step up from their existing 10nm. Samsung would have lied about their entire roadmap if 7nm LPP is just a slightly tuned 8nm LPP, but I don't think that is the case.

SFF2019-1 (9)_575px.jpg


Well, this discussion would be entirely pointless if Ampere is on TSMC 7nm lol. Or will people start talking about Nvidia using TSMC's fake 7nm.
 
Last edited:

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
  • Like
Reactions: lightmanek

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
Samsung's 7nm LPP EUV technical info has been known for a long time, and it is clear that it is half-node step up from their existing 10nm. Samsung would have lied about their entire roadmap if 7nm LPP is just a slightly tuned 8nm LPP, but I don't think that is the case.

View attachment 28912
Technically, they can call this process 7 nm LPN. Further Enhancing 8 nm LPU is absolutely not lying about their roadmap.

The names are extremely confusing though. That 7 nm process, about which we talk, is Further Enhancement of 8 nm LPU.
 

JasonLD

Senior member
Aug 22, 2017
485
445
136
Technically, they can call this process 7 nm LPN. Further Enhancing 8 nm LPU is absolutely not lying about their roadmap.

The names are extremely confusing though. That 7 nm process, about which we talk, is Further Enhancement of 8 nm LPU.

I honestly find this this whole 7nm conspiracy theory ridiculous. I am genuinely curious, is it because 8nm rumor that people really believed in is on the verge of being crumbled now they are starting the idea that 7nm is actually 8nm and Samsung has another 7nm that is nothing more than refined 8nm?
Come on now. Neither Samsung or Nvidia doesn't really have to go through any of this lol. I am pretty sure that regardless of 7nm or 8nm, they are just going to be straightforward on what node it is based on.
 
  • Like
Reactions: Lodix

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
I honestly find this this whole 7nm conspiracy theory ridiculous. I am genuinely curious, is it because 8nm rumor that people really believed in is on the verge of being crumbled now they are starting the idea that 7nm is actually 8nm and Samsung has another 7nm that is nothing more than refined 8nm?
Come on now. Neither Samsung or Nvidia doesn't really have to go through any of this lol. I am pretty sure that regardless of 7nm or 8nm, they are just going to be straightforward on what node it is based on.
Yeah, just like Nvidia and TSMC were straight forward about 12 nm FFN, a slightly refined 16 nm FinFet? ;)
 
  • Like
Reactions: Mopetar

USER8000

Golden Member
Jun 23, 2012
1,542
780
136
I honestly find this this whole 7nm conspiracy theory ridiculous. I am genuinely curious, is it because 8nm rumor that people really believed in is on the verge of being crumbled now they are starting the idea that 7nm is actually 8nm and Samsung has another 7nm that is nothing more than refined 8nm?
Come on now. Neither Samsung or Nvidia doesn't really have to go through any of this lol. I am pretty sure that regardless of 7nm or 8nm, they are just going to be straightforward on what node it is based on.

The numbers are marketing. TSMC 12NM was 16NM.Globalfoundries 12NM was 14NM.

NV being on 8nm would be 1nm behind amd.
 

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
I am not going to discuss further if that is the direction you are going to take. You should know that is not the same thing.
If that 7 nm process is slightly refined 8 nm LPP by Samsung, how can it be completely different thing, than 12 nm FFN being slightly refined 16 nm FinFet?

Have you ever, for a second, considered, that Kopite was right, in his info, considering how much Nvidia info he nailed? Or do you want to only confirm your belief that it is a completely different process, and that Kopite was wrong, alongside people who followed him?
 

JasonLD

Senior member
Aug 22, 2017
485
445
136
If that 7 nm process is slightly refined 8 nm LPP by Samsung, how can it be completely different thing, than 12 nm FFN being slightly refined 16 nm FinFet?

7nm is node jump from 10nm, while 8nm is refined 10nm. Saying 7nm is slightly refined 8nm will mean that 7nm is just a further refinement from their 10nm, but that is clearly not the case, unless Samsung is flat out lying about their processes.
I said it is different because when TSMC introduced 12nm, its name didn't overlap with any of their other existing node. What you are suggesting is that Samsung is creating another 7nm variation which is completely different from what their 7nm is, and foundries don't do that, and they will create another name for it if that is the case.

Have you ever, for a second, considered, that Kopite was right, in his info, considering how much Nvidia info he nailed? Or do you want to only confirm your belief that it is a completely different process, and that Kopite was wrong, alongside people who followed him?

As always with any leakers, he could be right or wrong, may get some right, may get some wrong. Never trust 100%.
I don't think I completely dismissed 8nm. I said, if it is 8nm, Nvidia will say it is 8nm. If it is 7nm, they will say it is 7nm. They will not fabricate some refined custom 8nm as 7nm.
 
  • Like
Reactions: Lodix

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
7nm is node jump from 10nm, while 8nm is refined 10nm. Saying 7nm is slightly refined 8nm will mean that 7nm is just a further refinement from their 10nm, but that is clearly not the case, unless Samsung is flat out lying about their processes.
I said it is different because when TSMC introduced 12nm, its name didn't overlap with any of their other existing node. What you are suggesting is that Samsung is creating another 7nm variation which is completely different from what their 7nm is, and foundries don't do that, and they will create another name for it if that is the case.



As always with any leakers, he could be right or wrong, may get some right, may get some wrong. Never trust 100%.
I don't think I completely dismissed 8nm. I said, if it is 8nm, Nvidia will say it is 8nm. If it is 7nm, they will say it is 7nm. They will not fabricate some refined custom 8nm as 7nm.
sff2019-1-9-_575px-jpg.28912

This is the very image you have brought here.

It blatantly there stays: Further Enhancement, under 10 LPE. It was refined to 10 LPP, 8 LPP, 8LPU, and then... "Further Enhancement" which suggests that such process was already in development.

If Further Enhancement of 8 nm LPU is 7 nm is there any lie?

We talk about naming of such nodes, as if it matter more than actual process technology.

Secondly. If that 7 nm process is further enhancement of 8 nm LPP, Kopite was correct.

And Nvidia is not lying calling them 7 nm GPUs. Everything falls into place. Mystery solved.

So why is it a problem?
 

JasonLD

Senior member
Aug 22, 2017
485
445
136
sff2019-1-9-_575px-jpg.28912

This is the very image you have brought here.

It blatantly there stays: Further Enhancement, under 10 LPE. It was refined to 10 LPP, 8 LPP, 8LPU, and then... "Further Enhancement" which suggests that such process was already in development.

If Further Enhancement of 8 nm LPU is 7 nm is there any lie?

We talk about naming of such nodes, as if it matter more than actual process technology.

Secondly. If that 7 nm process is further enhancement of 8 nm LPP, Kopite was correct.

And Nvidia is not lying calling them 7 nm GPUs. Everything falls into place. Mystery solved.

So why is it a problem?


If Samsung didn't have 7LPP, then calling the further enhancement of 8nm LPP "7nm" wouldn't be a problem. But Samsung already defined 7nm LPP as node jump from 10LPE.
I am pretty sure Samsung will not call Further enhancement from 8LPU "7nm" to avoid confusion, as they are based on completely different nodes.
 

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
If Samsung didn't have 7LPP, then calling the further enhancement of 8nm LPP "7nm" wouldn't be a problem. But Samsung already defined 7nm LPP as node jump from 10LPE.
I am pretty sure Samsung will not call Further enhancement from 8LPU "7nm" to avoid confusion, as they are based on completely different nodes.
Looking at the differences between Exynos 9825 and 990 are you absolutely sure about that?

They are both made on 7 nm nodes by Samsung, that have large differences between them.
 

moonbogg

Lifer
Jan 8, 2011
10,635
3,095
136
If Nvidia wanted to set the world on fire they'd release 3080 at $600 and 3080Ti 6 months from now at $700. That would require they actually have enough cards to meet demand though, and I'm pretty sure that isn't happening no matter what, lol. Like I said before, you got 30 seconds to buy this thing and if you fail to do so, you're screwed for a year.
 
  • Haha
Reactions: GodisanAtheist

JasonLD

Senior member
Aug 22, 2017
485
445
136
Looking at the differences between Exynos 9825 and 990 are you absolutely sure about that?

I see that being a quick port job vs designed from ground up difference. I don't see that as evidence of being 2 different variations of 7nm. I am pretty sure chip design has a huge influence on density as well.
 

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
I see that being a quick port job vs designed from ground up difference. I don't see that as evidence of being 2 different variations of 7nm. I am pretty sure chip design has a huge influence on density as well.
I will quote it for everyone who is lazy enough to not bother with looking at that blog post:

TechInsights said:
Last year, Samsung announced the introduction of EUV into their 7LPP process used in the Exynos 9825. Through analysis of the part, we found little difference between their 7LPP process in the 9825 and their 8LPP process in the Exynos 9820.

Its still named 7 nm LPP, by Samsung, despite being absolutely no different than 8 nm LPP, and vastly different from 7 nm LPP(EUV)...
 

Konan

Senior member
Jul 28, 2017
360
291
106
I see that being a quick port job vs designed from ground up difference. I don't see that as evidence of being 2 different variations of 7nm. I am pretty sure chip design has a huge influence on density as well.

The whole quick port job reference, Is positioning language trying to make it a negative for deviant reasons I would just ignore it. (Trying to make out that somehow Ampere is bad in some way because finer details of node process haven’t been disclosed yet)

The chip is designed to do what it is designed to do. It doesn’t really matter what the process is.
If it is all marketing as just the same 16nm/12nm Turing still came out ahead Despite process marketing naming differences. It’s not like there is a standard across all fabrications so there’s no point really arguing as public information about all of the details regarding each of the node processes aren’t fully disclosed for competitive reasons.

What can probably be said fairly from recent news is that Nvidia most likely has some form of custom node. To me doesn’t really matter what the process node number is, Hasn’t especially for the last couple of generations.
 
Last edited:
  • Like
Reactions: spursindonesia