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.

 

[Glo.]

You think the delay will happen WITHOUT Nvidia tweaking the uArch?

Man, not even I'm that optimistic haha.

No, I don't think that . I do believe that next gen, will be fully next gen, with new architecture.

Nailed it! Glad someone else is tired of it. I'm still waiting for the magic drivers that were suppose to make Vega ROFLOL stomp Pascal.

And Vega 64 is not as fast, or faster than GTX 1080, right now?

 

[Adonisds]

All that would mean 50% faster with less than 50% power usage. Not happening.

No. Why would a 20% smaller die with 20% higher clocks use less than 50% the power? What math did you do?

 

[uzzi38]

Fair enough. Then yes, TSMC did not missed their Performance targets, contrary to density and power targets.

Power targets were probably met too, but on the flip side Ryzen is actually the telling story there.

You need to take into account that these targets are defined using the V/f curve. At high clocks, pushing the curve forwards 200mhz or so could actually mean a HUGE drop in power draw, like 30% or so in extreme cases.

At low clocks? Well then, you can find the best case scenario for increasing clocks, as it's pretty easy to do there.

That's what TSMC's figures are based off.

 

[lobz]

He said a 20% bump to clocks and 25% bump to perf per clock. That's not 50% perf with half the power, that's 50% perf at the same power.

Whether or not it's doable will be interesting to see. The 30-50% generation on generation performance improvement is totally doable and standard for Nvidia.

You forget that he said 20% less die area too, and it's 7nm from 12/16.

 

[Adonisds]

Then show proofs for Your bloody claims, before you will call anyone out. I asked you this before, in another topic. You haven't responded because you haven't got any of them.

So I ask you. Show proofs, that Nvidia will have suddenly better process for their GPUs, and that TSMC will not miss power/density targets for only their GPUs. Show proofs that Ampere is not HPC only. Show proofs that next gen Nvidia cards are actually coming this year, that are not based only on GA100 chip.

Show actual proofs, because when you have been shown proofs or confirmation, that is not happening, you dismissed them, or ignored, just like Muhammad did.

And whats worse, based on your belief that Nvidia will actually deliver next gen GPUs for gaming this year, you were advicing in another thread someone to just wait for them.

You may actually be right, if anyone is patient enough to wait circa 9 months for them, in the best possible scenario.

Why is it so unbelievable that Nvidia will release a new generation of cards a bit more than 2 years after the last?

If it does release only next year this gen will be the longest lived by far

 

[Glo.]

Why is it so unbelievable that Nvidia will release a new generation of cards a bit more than 2 years after the last?

If it does release only next year this gen will be the longest lived by far

Because we have had confirmation from certain good, reliable source that Nvidia hasn't taped out next gen Gaming GPUs, and only GA100, which is only HPC chip has been.

There are rumors about GA103, and GA104 chips, which are fake, according to that source. Only GA100 chip is from Ampere architecture, and this is HPC, professional only.

Based on this we expect only: GA100 based Titan A - prosumer GPU, Quadro and Tesla GPUs based on this chip.

But next gen gaming cards are named differently, and are based on different architecture than Ampere. And they are not going to be released anytime soon.

 

[railven]

And Vega 64 is not as fast, or faster than GTX 1080, right now?

Wait, AMD finally got the primitive shaders working? Remember the ones you said would make Vega 64 match/beat GTX 1080 Ti? Oh, you changed it to GTX 1080 now.

Gotcha.

 

[Glo.]

Wait, AMD finally got the primitive shaders working? Remember the ones you said would make Vega 64 match/beat GTX 1080 Ti? Oh, you changed it to GTX 1080 now.

Gotcha.

No, they did stopped the development of those drivers. And yes, I was wrong about it.

But what makes you believe that I am making stuff up about Nvidia next gen Gaming GPUs, and that Ampere is HPC only chip, hm?

 

[happy medium]

Why is it so unbelievable that Nvidia will release a new generation of cards a bit more than 2 years after the last?

If it does release only next year this gen will be the longest lived by far

If your a AMD stock holder , the last thing you want is talk of a new Nvidia card to smack big Navi 's performance and lower its MSRP before it even launches .
If I were a rabid AMD fan/stockholder, I would be saying things like ,we wont see a new Nvidia card till next year or it's going to be huge, miss its power targets and it's not even taped out yet.

That's a good explanation.

 

[Glo.]

If your a AMD stock holder , the last thing you want is talk of a new Nvidia card to smack big Navi 's performance and lower its MSRP before it even launches .
If I were a rabid AMD fan/stockholder, I would be saying things like ,we wont see a new Nvidia card till next year or it's going to be huge, miss its power targets and it's not even taped out yet.

That's a good explanation.

If that is the best thing you can do, when I asked you directly, where are your proofs that Nvidia will come with new gaming GPUs this year...

Well the outlook for all Nvidia fans, eagerly waiting for next gen GPUs, is not the best, is it?

P.S. I have zero AMD, or other tech stock.

 

[lobz]

No. Why would a 20% smaller die with 20% higher clocks use less than 50% the power? What math did you do?

What is wrong with you people tonight? Because it's on a full-on node shrink as well?

 

[uzzi38]

If your a AMD stock holder , the last thing you want is talk of a new Nvidia card to smack big Navi 's performance and lower its MSRP before it even launches .
If I were a rabid AMD fan/stockholder, I would be saying things like ,we wont see a new Nvidia card till next year or it's going to be huge, miss its power targets and it's not even taped out yet.

That's a good explanation.

Can't speak for Glo, but, I'm personally way too broke for that, no chance I'll be getting into stocks any time soon.

If I cared about whatever is of greatest benefit to me personally, I'd be rooting for Intel in the hopes that they might bring down GPU pricing, because AMD and Nvidia certainly won't. Oh no, pricing will get rough from both.

I just follow this stuff because I find it interesting (yes, I'm something of a freak). The tapeout rumour has weight to it, and it's something that first popped up on the rumour mill months ago that has only seemed more and more likely as time has gone on.

 

[uzzi38]

You forget that he said 20% less die area too, and it's 7nm from 12/16.

Well I don't want to comment much on die sizes, but I mean, 20% alongside the rest sounds perfectly reasonable to me.

 

[Glo.]

If I cared about whatever is of greatest benefit to me personally, I'd be rooting for Intel in the hopes that they might bring down GPU pricing, because AMD and Nvidia certainly won't. Oh no, pricing will get rough from both.

Yep. But that may way too optimistic. We know how Intel thinks about themselves, and what happened during the years of AMD being incompetent. Nothing would stop Intel from giving meaningless performance uplifts with each generation if they would be in the same position as they were then.

Funnily enough I believe its Nvidia who can keep AMD and Intel honest on GPU front. Proofs? Look no further than latest RTX 2060 price drop, and AMD's reaction to it...

 

[Adonisds]

What is wrong with you people tonight? Because it's on a full-on node shrink as well?

A similar die size should have similar power consumption. Don't you know Dennard scaling ended around 2006?

 

[railven]

No, they did stopped the development of those drivers. And yes, I was wrong about it.

But what makes you believe that I am making stuff up about Nvidia next gen Gaming GPUs, and that Ampere is HPC only chip, hm?

It's more because of this:

No, they did stopped the development of those drivers. And yes, I was wrong about it.

You kind of oversell AMD while underselling NV, and consistently you've been wrong.

It's not that I think you're lying, but you kind of cherry pick what you want to accept as a source and it's only those sources you've selected that are valid. And even then it becomes inconsistent.

You said Vega 64 would beat GTX 1080 Ti, just think about that statement in itself. Even AMD didn't think that and never intended for them to compete. But you sure did and FineWine was the supporting meme to keep it going for so long that even in this thread you still wanted to claim victory.

And Vega 64 is not as fast, or faster than GTX 1080, right now?

Vega 64 was "as fast or faster" than GTX 1080 since launch.

 

[Glo.]

A similar die size should have similar power consumption. Don't you know Dennard scaling ended around 2006?

RTX 2060 with 445 mm2 die has almost half the power draw of 484 mm2 GTX 1080 Ti.

Nvidia shocked the world with Pascal GPUs and how high they clocked. They did the same with Turing but with its power draw, in relation to die sizes. I would not draw conclusions based solely on die size alone, for any upcoming GPUs.

You kind of oversell AMD while underselling NV, and consistently you've been wrong.

It's not that I think you're lying, but you kind of cherry pick what you want to accept as a source and it's only those sources you've selected that are valid. And even then it becomes inconsistent.

Where did I talked about Nvidia GPUs before, eh?

When I talked about Turing I hoped they will use Volta's architecture, and you know what? They did! When Talked about Volta, I said it was my favorite Nvidia GPU arch for very long time. Which it was, until Turing.

How can I undersell Nvidia, when mainly I post in AMD threads, because of the nature of their product releases? You could've seen me in other threads that asked consumer related questions posting about AMD and Nvidia from both sides. You have simply made up your mind about me, and now you go along with it.

And that is the problem.

 

[railven]

RTX 2060 with 445 mm2 die has almost half the power draw of 484 mm2 GTX 1080 Ti.

Nvidia shocked the world with Pascal GPUs and how high they clocked. They did the same with Turing but with its power draw, in relation to die sizes. I would not draw conclusions based solely on die size alone, for any upcoming GPUs.

Where did I talked about Nvidia GPUs before, eh?

When I talked about Turing I hoped they will use Volta's architecture, and you know what? They did! When Talked about Volta, I said it was my favorite Nvidia GPU arch for very long time. Which it was, until Turing.

How can I undersell Nvidia, when mainly I post in AMD threads, because of the nature of their product releases? You could've seen me in other threads that asked consumer related questions posting about AMD and Nvidia from both sides. You have simply made up your mind about me, and now you go along with it.

And that is the problem.

Welps, nothing good will come from this so, I'll just refrain from interacting with you. Least that way we'll both be happy

 

[Glo.]

Welps, nothing good will come from this so, I'll just refrain from interacting with you. Least that way we'll both be happy

You started it. It was you who turned on personal side this discussion.

 

[Adonisds]

RTX 2060 with 445 mm2 die has almost half the power draw of 484 mm2 GTX 1080 Ti.

Nvidia shocked the world with Pascal GPUs and how high they clocked. They did the same with Turing but with its power draw, in relation to die sizes. I would not draw conclusions based solely on die size alone, for any upcoming GPUs.

Where did I talked about Nvidia GPUs before, eh?

When I talked about Turing I hoped they will use Volta's architecture, and you know what? They did! When Talked about Volta, I said it was my favorite Nvidia GPU arch for very long time. Which it was, until Turing.

How can I undersell Nvidia, when mainly I post in AMD threads, because of the nature of their product releases? You could've seen me in other threads that asked consumer related questions posting about AMD and Nvidia from both sides. You have simply made up your mind about me, and now you go along with it.

And that is the problem.

I'll bite, but just this time:
The 2060 has less memory, more efficient memory, lower clocks and a lot of its die area is dark silicon (is that what's called?) doing nothing before of AI and RT.

A fair comparison would be GTX 680 and GTX 1080: similar die size and power consumption

 

[DXDiag]

I don't know why some people are constantly downgrading anything NVIDIA does with their next gen, I would like to quote a famous Beyond3D post about this:

If we follow past trends, comparing the biggest gaming dies NVIDIA releases, a certain pattern emerges:

580 to 780Ti is a 85% uplift (new node)
780Ti to 980Ti is a 45% uplift (same node)
980Ti to 1080Ti is a 75% uplift (new node)
1080Ti to 2080Ti is a 40% uplift (same node)

So at the very least I expect a 3080TI to be around 70% faster than 2080Ti at the same TDP, that would make it close to two times as efficient as Turing, it all depends on what NVIDIA can extract from the new node.

Nvidia Post-Volta (Ampere?) Rumor and Speculation Thread

1- 16/12nm -> 10nm 2- 10nm -> 7nm 10nm was never an option for high performance SoCs, it's designed only for mobile chips, so I wouldn't count it as a separate node when talking about GPUs

forum.beyond3d.com

I too expect a 3080Ti to be 70% faster than 3080Ti at the new node, based on past trends, and the latest rumors about HPC Ampere are also empowering this prediction.

As for the timeline, If NVIDIA is releasing HPC next gen soon, chances are they are releasing gaming soon as well, just like they did with Pascal, where they released GP100 alongside the GTX 1080.

 

[AtenRa]

I don't know why some people are constantly downgrading anything NVIDIA does with their next gen, I would like to quote a famous Beyond3D post about this:

Nvidia Post-Volta (Ampere?) Rumor and Speculation Thread

1- 16/12nm -> 10nm 2- 10nm -> 7nm 10nm was never an option for high performance SoCs, it's designed only for mobile chips, so I wouldn't count it as a separate node when talking about GPUs

forum.beyond3d.com

I too expect a 3080Ti to be 70% faster than 3080Ti at the new node, based on past trends, and the latest rumors about HPC Ampere are also empowering this prediction.

As for the timeline, If NVIDIA is releasing HPC next gen soon, chances are they are releasing gaming soon as well, just like they did with Pascal, where they released GP100 alongside the GTX 1080.

From GTX980Ti (8B transistors) to GTX1080Ti (11.8B transistors) we had a 48% increase of the transistor budget all dedicated to Raster performance.
GTX980Ti = 601mm2 @ 28nm
GTX1080Ti = 471mm2 @ 16nm

So if we want to reach 50-70% increase of rasterizing performance vs RTX2080Ti , then we have to increase the transistor budget by almost 50%. And we also have RayTracing that NVIDIA will also allocate more transistors than Turing.
If we increase the transistor budget of theRTX2080Ti (18.6B transistors) by 50% to 3080Ti (27.9B transistors) and make a die at 7nm it will be close to 600-700mm2.

Since 7nm wafers are 2x more expensive than 12nm , the new 3080Ti should cost almost 2x the price of 2080Ti in order for NVIDIA to keep the same margins.

Just simple to put, 600-700mm2 dies at 7nm for consumer products is not going to happen in 2020.

As i have said before, im expecting a 30% increase in Raster performance and 2x increase in RayTracing performance vs Turing , with smaller dies and same prices.
Also to point out that im not certain that NVIDIA will have access to high volume 7nm wafers in 1H 2020 in order to release lots of 7nm products in H2 2020.

So to sum up,

Dont expect 70% Raster performance increase over Turing and we still dont know when NV will release its next consumer products.

 

[happy medium]

From GTX980Ti (8B transistors) to GTX1080Ti (11.8B transistors) we had a 48% increase of the transistor budget all dedicated to Raster performance.
GTX980Ti = 601mm2 @ 28nm
GTX1080Ti = 471mm2 @ 16nm

So if we want to reach 50-70% increase of rasterizing performance vs RTX2080Ti , then we have to increase the transistor budget by almost 50%. And we also have RayTracing that NVIDIA will also allocate more transistors than Turing.
If we increase the transistor budget of theRTX2080Ti (18.6B transistors) by 50% to 3080Ti (27.9B transistors) and make a die at 7nm it will be close to 600-700mm2.

Since 7nm wafers are 2x more expensive than 12nm , the new 3080Ti should cost almost 2x the price of 2080Ti in order for NVIDIA to keep the same margins.

Just simple to put, 600-700mm2 dies at 7nm for consumer products is not going to happen in 2020.

As i have said before, im expecting a 30% increase in Raster performance and 2x increase in RayTracing performance vs Turing , with smaller dies and same prices.
Also to point out that im not certain that NVIDIA will have access to high volume 7nm wafers in 1H 2020 in order to release lots of 7nm products in H2 2020.

So to sum up,

Dont expect 70% Raster performance increase over Turing and we still dont know when NV will release its next consumer products.

That's a good summary.
I'm going for 50% faster raster , 80% faster with RT , 20% smaller ,at about the same prices.

 

[Bouowmx]

Die area is a gamble. Using 27900 M transistors (1.5x TU102):

• Could be like AMD Navi, 41 M/mm^2: 680 mm^2
• Could be 50 M/mm^2 (close to TSMC 7 nm high-performance): 558 mm^2
• Could be 75 M/mm^2 (close to TSMC 7 nm high-density): 372 mm^2

 

[lobz]

A similar die size should have similar power consumption. Don't you know Dennard scaling ended around 2006?

?????????????????????????????????

That's at least as wrong as saying 12nm - 7nm results in 3.3x density. I'd advice that you look for better info again...