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.

 

[exquisitechar]

To be honest, I don't see what's the point of Titan when there's Quadro lineup. For those that do not need full blown Quadro, RTX 3080 Ti should be enough.

Thing is, it seems like the 3080ti/3090(?) won’t be ready until 2021. Releasing a GA100 Titan could be Nvidia’s only way of keeping the gaming performance crown once AMD releases their fastest Navi 2x card. Not sure how feasible it is for them to do so for several reasons, though.

 

[Tup3x]

Thing is, it seems like the 3080ti/3090(?) won’t be ready until 2021. Releasing a GA100 Titan could be Nvidia’s only way of keeping the gaming performance crown once AMD releases their fastest Navi 2x card. Not sure how feasible it is for them to do so for several reasons, though.

If they can release Titan, they can also release 3080 Ti. If they would target gamers with it, after all it would be the same but likely few cores disabled and stuff nerfed.

 

[Qwertilot]

Can, maybe. They tend to stagger it quite a long time though - some of that is just waiting for manufacturing costs to mature over time, some of it after early adopter money.

Also if they're doing things like Volta/Turing they'll have a rather different architecture for the compute and gaming chips, and the 3080ti will definitely be a gaming chip. We'll see.

 

[Glo.]

If they can release Titan, they can also release 3080 Ti. If they would target gamers with it, after all it would be the same but likely few cores disabled and stuff nerfed.

Is this based on hope?

Titan V cost 3000$, for cut-down GPU. What makes you believe that Titan A based on the same type of chip(HPC, massive die) will not cost 5000$ because of astronomical design cost of the chip?

If Nvidia's only Ampere GPU that we will see this year is GA100 - they can not release 3080 Ti based on it, anytime soon, because it will be "unsellable" to customers with its astronomical price.

 

[Veradun]

If they can release Titan, they can also release 3080 Ti.

Like that Volta 80Ti?

 

[Tup3x]

Is this based on hope?

Titan V cost 3000$, for cut-down GPU. What makes you believe that Titan A based on the same type of chip(HPC, massive die) will not cost 5000$ because of astronomical design cost of the chip?

If Nvidia's only Ampere GPU that we will see this year is GA100 - they can not release 3080 Ti based on it, anytime soon, because it will be "unsellable" to customers with its astronomical price.

Not hoping anything and you are mixing thing. If they would release a Titan that would be targeted for gamers, then the gamer oriented Ti version would be likely just cut down version of that and it would be different from compute oriented versions (like how it has been, usually). I'm not expecting to see Titan though, might happen but I think they just should keep GeForce, Quadro and Tesla cards properly separate.

The leaked cards are likely compute oriented ones. Though, there are Quadros with massive amounts of VRAM so... They could be anything (likely Teslas).

Not really liking how you are setting your "questions" by the way, that's unnecessary and your assumptions were incorrect.

 

[Glo.]

Not hoping anything and you are mixing thing. If they would release a Titan that would be targeted for gamers, then the gamer oriented Ti version would be likely just cut down version of that and it would be different from compute oriented versions (like how it has been, usually). I'm not expecting to see Titan though, might happen but I think they just should keep GeForce, Quadro and Tesla cards properly separate.

The leaked cards are likely compute oriented ones. Though, there are Quadros with massive amounts of VRAM so... They could be anything (likely Teslas).

Not really liking how you are setting your "questions" by the way, that's unnecessary and your assumptions were incorrect.

Just Like there was XX80 Ti based on GV100, eh?

Once again let me repeat. If there will be Titan A, based on GA100, there will not be any XX80 Ti based on this GPU. Period. Its way too expensive, from manufacturing point of view.

 

[reb0rn]

With the Navi2x I bet NV will have to come out guns blazing so 70% perf is defo realistic!!
Also power use is quite possible if Navi2 managed to cut power use up to 50% from Navi1
It all looks like 7nm node matured a lot and the one used for RX5700 is quite bad in power use

So let say 3080TI defo will be HBM!

 

[lobz]

Not hoping anything and you are mixing thing. If they would release a Titan that would be targeted for gamers, then the gamer oriented Ti version would be likely just cut down version of that and it would be different from compute oriented versions (like how it has been, usually). I'm not expecting to see Titan though, might happen but I think they just should keep GeForce, Quadro and Tesla cards properly separate.

The leaked cards are likely compute oriented ones. Though, there are Quadros with massive amounts of VRAM so... They could be anything (likely Teslas).

Not really liking how you are setting your "questions" by the way, that's unnecessary and your assumptions were incorrect.

You said, and I quote, 'a few cores disabled'.

Please don't say @Glo. is mixing thing.

 

[Glo.]

So let say 3080TI defo will be HBM!

If RDNA2 really is living up to the hype AMD created, Nvidia has no other option .

 

[maddie]

With the Navi2x I bet NV will have to come out guns blazing so 70% perf is defo realistic!!
Also power use is quite possible if Navi2 managed to cut power use up to 50% from Navi1
It all looks like 7nm node matured a lot and the one used for RX5700 is quite bad in power use

So let say 3080TI defo will be HBM!

There's no guarantee for anything. Sometimes you just lose. AMD knows this a lot from not too long ago. Not putting down Nvidia, as they have proven to be and are formidable innovators, but don't let recent history blind you to the fact that change can happen.

 

[reb0rn]

Change how, NV is not Intel that sleep as bear for 10 years
AMD GPU is very bad only reason they sell is NV is overpricing their cards by 2x at least!

 

[Glo.]

Change how, NV is not Intel that sleep as bear for 10 years
AMD GPU is very bad only reason they sell is NV is overpricing their cards by 2x at least!

Considering that the likes of RX 5600 XT are faster, cheaper and more efficient than direct competitor: RTX 2060 I would not write with straight face that AMD GPUs are bad, even in a Nvidia thread .

And no, lets not turn this thread into AMD vs Nvidia war

 

[lobz]

Change how, NV is not Intel that sleep as bear for 10 years
AMD GPU is very bad only reason they sell is NV is overpricing their cards by 2x at least!

UUUuuhhhhhhhhhh this comment made me cringe so bad, that it's best if I don't reply more than this.

 

[Adonisds]

When announcing Volta, the man in the cool leather jacket said that it would be a huge jump compared to the previous gen in AI because of the tensor cores. But recently we see machines being sold to be used in AI using AMD GPUs that, as far as I know, don't have AI acceleration. How important and good for AI exactly are those tensor cores?

 

[Glo.]

When announcing Volta, the man in the cool leather jacket said that it would be a huge jump compared to the previous gen in AI because of the tensor cores. But recently we see machines being sold to be used in AI using AMD GPUs that, as far as I know, don't have AI acceleration. How important and good for AI exactly are those tensor cores?

They are good for anything Nvidia hardware related(Nvidia specific). Everything else looks to be running on BFloat16.

 

[Det0x]

KittyCorgi with some more leaks/rumors on twitter

https://twitter.com/CorgiKitty/status/1235508177363300353

In fact the GA102 still doesn’t tape out but specifications have been determined.

https://twitter.com/x/status/1235244837969879041

https://twitter.com/x/status/1235508987472130048

 

[Glo.]

Only 40% faster than RTX 2080 Ti?

That's laughable...

For this very reason I very much doubt in this rumor, because it means that Nvidia genuinely ******d up.

 

[maddie]

Only 40% faster than RTX 2080 Ti?

That's laughable...

For this very reason I very much doubt in this rumor, because it means that Nvidia genuinely ******d up.

Why? Maybe they want to to have users upgrading more often, plus they might have decided for smaller die sizes. The danger with this is overconfidence and ignoring your competitor as real opponent. After all, this was Intels playbook for many years.

 

[xpea]

KittyCorgi with some more leaks/rumors on twitter

https://twitter.com/CorgiKitty/status/1235508177363300353

In fact the GA102 still doesn’t tape out but specifications have been determined.

Another random twitty guy looking for followers.... I can guarantee 100% that:
v1- Number of shaders is totally wrong
2- GA102 taped out last year

 

[happy medium]

Or mabe because that's how it's been the past 8 years?
980ti to 1080ti, about 45% faster,1080ti to 2080ti,about 40% faster, 2080ti to 3080ti should be about 40% faster and much faster ray tracing,sounds about right.

And the big Navi will be close to the performance of the rtx3080 and about 40% slower.

 

[exquisitechar]

Or mabe because that's how it's been the past 8 years?
980ti to 1080ti, about 45% faster,1080ti to 2080ti,about 40% faster, 2080ti to 3080ti should be about 40% faster and much faster ray tracing,sounds about right.

And the big Navi will be close to the performance of the rtx3080 and about 40% slower.

1080ti is a quite a bit more than 45% over the 980ti, I think. For a node shrink + microarchitectural innovations 40% seems to be a small improvement relative to what we’ve seen before. I also don’t trust KittyCorgi for several reasons. Calling BS on this one.

TU102 is huge, though. Maybe Nvidia opted for a 40% uplift over it overall and significantly higher RT performance, without creating too big of a die. It’s not totally out of the question.

 

[CakeMonster]

980ti to 1080ti was two years, for consumers and gamers who couldn't care less about the technical details of the chip, what they saw was 980ti -> 1080 -> 1080ti at a steady 1 year release cycle and ~25% performance increase for each.

 

[sontin]

If RDNA2 really is living up to the hype AMD created, Nvidia has no other option .

Because the competition has finally catched up with a two year old architecture? Bravo.

 

[uzzi38]

1080ti is a quite a bit more than 45% over the 980ti, I think. For a node shrink + microarchitectural innovations 40% seems to be a small improvement relative to what we’ve seen before. I also don’t trust KittyCorgi for several reasons. Calling BS on this one.

Thank you for being sensible. CorgiKitty has done nothing to prove temselves yet aside from repeat already existing rumours and claim them as their own until now. There's still no reason to trust them.