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

[Ottonomous]

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.

 

[Stuka87]

12nm is 16nm. The same process. And nVidia is beating AMD with 16nm.

And that something i guess people just ignore: AMD is not using a new and advanced process. They have done this step last year. nVidia is going from 16nm to something with >2x the density.

TSMC had stated that they worked with nVidia to create the 12nm process, it is no renamed 16 or anything.

 

[Glo.]

What? It is exactly what you have written:



So yes, you are saying that nVidia wont have a chance to beat RDNA2 with a new architecture on a new node while AMD is using RDNA on the same node.

Can you point me to a post in which I SPECIFICALLY say that Next gen Nvidia architecture will not improve, or you are just putting words into my keyboard that I have not written?

RDNA2 is not RDNA1. Put that idea out of your head.

TSMC had stated that they worked with nVidia to create the 12nm process, it is no renamed 16 or anything.

Sontin is correct saying it is 16 nm process. Because it has all of those physical optimizations I talked about, that resulted in Turing's efficiency, at the expense of die size. It was interplay of Nvidia Engineers with TSMC engineers ending up in masterpiece of a physical design. However...

8 nm Samsung's Process does not have those optimizations.

 

[sontin]

Can you point me to a post in which I SPECIFICALLY say that Next gen Nvidia architecture will not improve, or you are just putting words into my keyboard that I have not written?

RDNA2 is not RDNA1. Put that idea out of your head.

I have pointed you to one of your posting in which you have written that Ampere on Samsung's 8nm process would be inferior to RDNA2.
We have more information about RDNA2 than Ampere and yet you are claiming that nVidia has no chance.

 

[Glo.]

I have pointed you to one of your posting in which you have written that Ampere on Samsung's 8nm process would be inferior to RDNA2.
We have more information about RDNA2 than Ampere and yet you are claiming that nVidia has no chance.

So you confirm that I have not written what you imply I have.

No, you know nothing about RDNA2, especially based on what you posted in this thread.

 

[Konan]

and only product that is taped out on 7 nm TSMC is GA100 chip.

Got a source for that or is it an assumption?

 

[Konan]

What we have for sure now is two things:
Ampere HPC is 7nm TSMC
Ampere Gaming high end is 7nm TSMC
Ampere Gaming low end is 8nm Samsung

FYI - https://finance.technews.tw/2020/05/07/nvidia-ampere-gpu-for-tsmc-foundry/ - this says "The entry-level Ampere architecture GPU is expected to be produced on Samsung's 7-nm EUV process or 8-nanometer process technology."

Edit: here is Nvidia officially saying Samsung 7nm http://www.koreaherald.com/view.php?ud=20190702000692#cb (so looks like 7nm EUV for the low end not 8nm?)

Edit 2:

• Ampere HPC is 7nm TSMC
• Ampere Gaming high end is 7nm TSMC

Could both be N7+

https://twitter.com/x/status/1112053410860290049

 

[sontin]

So you confirm that I have not written what you imply I have.

No, you know nothing about RDNA2, especially based on what you posted in this thread.

How can we know "nothing about RDNA2" when Sony, Microsoft and even AMD have put out information? I mean we know that RDNA2 wont have Tensorcores, is +-50% more efficient than RDNA1, the RT performance is around Turing and a 12TFLOPs Xbox One Series X doesnt perform so much better than a RTX2080.

On the other hand we have no information about Ampere. Maybe you can give us some? Otherwise you claims are not even assumptions...

 

[Glo.]

How can we know "nothing about RDNA2" when Sony, Microsoft and even AMD have put out information? I mean we know that RDNA2 wont have Tensorcores, is +-50% more efficient than RDNA1, the RT performance is around Turing and a 12TFLOPs Xbox One Series X doesnt perform so much better than a RTX2080.

On the other hand we have no information about Ampere. Maybe you can give us some? Otherwise you claims are not even assumptions...

Even based on that information, you know nothing about RDNA2 .

 

[Konan]

Even based on that information, you know nothing about RDNA2 .

But this is not AMD thread, so lets not make this into a Nvidia vs AMD fight.

^^ that would be great thanks if you stop derailing this thread.

 

[sontin]

Yeah, but the fun part is you know nothing about Ampere, too. And yet you think that nVidia cant put out a product which will be much better than RDNA2 which doesnt have Tensorcores and a worse Raytracing performance.

But only six days and a few hours to go.

 

[Glo.]

But only six days and a few hours to go.

Its High Performance Computing chip. What do you expect GA100 to be?

The only way it lands on consumer market is as a Titan A, replacement for Titan V.

 

[Glo.]

^^ that would be great thanks if you stop derailing this thread.

It was not I who was pushing the discussion into this direction. But thanks for reminding Sontin, and DXDiag to keep it straight.

 

[sontin]

Try again: https://forums.anandtech.com/thread...arch-speculation-thread.2572510/post-40155409

Its High Performance Computing chip. What do you expect GA100 to be?

The only way it lands on consumer market is as a Titan A, replacement for Titan V.

Hm, a GPU which would be not really better than GV100. I mean why should the HPC chip lightyears ahead of the gaming GPUs.

 

[Glo.]

Try again: https://forums.anandtech.com/thread...arch-speculation-thread.2572510/post-40155409



Hm, a GPU which would be not really better than GV100. I mean why should the HPC chip lightyears ahead of the gaming GPUs.

Can you stop manipulating and twisting my words?

As far as I know this is considered trolling and is bannable offense on this very forum.

 

[sontin]

What is "manipulating and twisting"? Your claims?
You are claiming that Samsung's 8nm is so bad that GA100 should be lightyears ahead of the gaming chips. Which doesnt make any sense because nVidia would have gone only with TSMC and would not have waited two years to scrap their plans.

Maybe you should stop posting baseless claims.

 

[Glo.]

What is "manipulating and twisting"? Your claims?
You are claiming that Samsung's 8nm is so bad that GA100 should be lightyears ahead of the gaming chips. Which doesnt make any sense because nVidia would have gone only with TSMC and would not have waited two years to scrap their plans.

Maybe you should stop posting baseless claims.

This is exactly what I mean when I say: "can you stop manipulating and twisting my words".

Thank you. Because that is what you have just done with my words .

 

[sontin]

But you said it. You said that Samsung's 8nm process is so far away from TSMC's 7nm process that it forced nVidia to scrap their lineup:

And why would all of this happen?
Because Nvidia knows that with 7 nm products on SS process, they wouldn't have a chance against AMD's RDNA2. Retaping their architecture, on another process gives them the opportunity to give AMD a fight.

No manipulation and twisting.

 

[Glo.]

But you said it. You said that Samsung's 8nm process is so far away from TSMC's 7nm process that it forced nVidia to scrap their lineup:


No manipulation and twisting.

You said that based on what I am posting, GA100 will not be better than GV100, in this very post:

Hm, a GPU which would be not really better than GV100. I mean why should the HPC chip lightyears ahead of the gaming GPUs.

No manipualtion and twisting, eh?

P.S. You've just got a dose of what you are selling in this thread. Lets end it here. I have nothing to prove to anybody. What I have been writing about Ampere has been confirmed by articles on past few pages. Lets wait those 6 days, for the rest of my claims, and then lets wait to the end of this year for the rest to be confirmed as well.

 

[sontin]

Yes, because why would nVidia use such an inferior process from Samsung when they could just produce smaller chips to save money? Even nVidia is downplaying the importance of smaller nodes. So it doesnt matter for them. And i think everyone here will agree that 50% performance increase vom Volta to Ampere is "not really better" when nVidia went from 16 to 7nm. At the same time you are claming that AMD will archive more with RDNA2 even without a new architecture and a smaller node.

And nothing has been confirmed by any article. It isnt even confirmed that the HPC chip will be coming from TSMC. Orin is the only confirmed product (Samsung's new automobile 8nm process).

/edit: And there is another point: nVidia is using mostly "Geforce"-Dies for Quadro cards and their inference business is growing faster than the training part. So why would they cutting off their own nose with an inferior Samsung process when they can easily increase their position in these professionell markets...

 

[DXDiag]

Ah, yes the famous TechpowerUp, measuring overall performance, not the IPC of GPUs. I thought we talked about IPC, not overall performance?

Overall performance is the IPC of the GPUs across different workloads. Fact is Turing often does more with less. And it doesn't even use any of the advanced features it has yet: Variable Rate Shading, Mesh Shaders, Sampler Feedback, Tesnor Cores, Ray Tracing. And When it does it literally will blow RDNA1 out of the water, just wait and see how RDNA1 crumbles under the pressure of the new wave of consoles games.

P.S. You've just got a dose of what you are selling in this thread. Lets end it here. I have nothing to prove to anybody. What I have been writing about Ampere has been confirmed by articles on past few pages. Lets wait those 6 days, for the rest of my claims, and then lets wait to the end of this year for the rest to be confirmed as well.

Sorry, all you did is prove yourself woefully wrong and embarrassed yourself in the process.

You claimed NVIDIA won't have a share in 7nm from TMSC, WRONG.
You claimed NVIDIA next gen gaming GPUs will be 8nm from Samsung, WRONG.
You claimed gaming Ampere hasn't taped out yet WRONG.

 

[DXDiag]

Turing is 12nm, not 16. And Turing isn't beating AMD, competing chips are nearly identical to each other.

12nm is a slightly enhanced 16nm process, 7nm is vastly better than both.
The fact that Turing is still using 12nm and is better than the 7nm RDNA1 speaks volumes about the kind of architectural efficiency and power efficiency advantages NVIDIA is enjoying right now.

No, you know nothing about RDNA2, especially based on what you posted in this thread.

We know enough now, a 52CU RDNA2 part (in Xbox Series X) is equal to an RTX 2080 in Gears 5, that means clock for clock, RDNA2 is about the same as RDNA1, or very slightly better (~5%), AMD only claimed 50% performance per watt improvements, nothing more. Most of that is spent on reducing the power draw of RDNA2 compared to RDNA1. Granted that will mean AMD can now build bigger chips, instead of the 40CU part consuming 225W, AMD can now build a 80CU part while consuming about 250W.

 

[maddie]

12nm is a slightly enhanced 16nm process, 7nm is vastly better than both.
The fact that Turing is still using 12nm and is better than the 7nm RDNA1 speaks volumes about the kind of architectural efficiency and power efficiency advantages NVIDIA is enjoying right now.

We know enough now, a 52CU RDNA2 part (in Xbox Series X) is equal to an RTX 2080 in Gears 5, that means clock for clock, RDNA2 is about the same as RDNA1, or very slightly better (~5%), AMD only claimed 50% performance per watt improvements, nothing more. Most of that is spent on reducing the power draw of RDNA2 compared to RDNA1. Granted that will mean AMD can now build bigger chips, instead of the 40CU part consuming 225W, AMD can now build a 80CU part while consuming about 250W.

I know I really should not be responding but I'll give it one shot.

GloFlo put out a release a few months ago about a 12nm process they had that would deliver most of the power benefits of 7nm. I was really surprised to see that was possible and seeing that none of us know what transpired with this custom 12 nm for Nvidia, I would not make definitive claims.

 

[DXDiag]

GloFlo put out a release a few months ago about a 12nm process they had that would deliver most of the power benefits of 7nm. I was really surprised to see that was possible and seeing that none of us know what transpired with this custom 12 nm for Nvidia, I would not make definitive claims.

We know the 7nm TSMC is vastly better than 12nm TSMC, all the facts (process parameters) are out there already.

 

[beginner99]

And yet you think that nVidia cant put out a product which will be much better than RDNA2

No, he said using the Samsung process will hamper NV so badly that they switched to TSMC. And it's entirely possible that Samsung botched the process. Just look at intel 10nm. So either the processes had too poor yields (for the big dies needed) or simply didn't clock high enough. And because NV switched high-end gaming 3000-series will be late, eg Q1 2021.

I'm not saying that is true at all, I'm just saying that he is being very clear in what he says and you seem to kind of intentionally (or unintentionally) not understanding it.

Ulimtaltey no discussion needed as time will tell who was right.

 

[sontin]

I understood it. But this means that Samsung's 8nm process is worse than TSMC's 16nm process and nVidia cant produce anything on it. At the same time they will release a +-600mm^2 Die on TSMC's 7nm process without problems...
.