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.

 

[CP5670]

To be honest, the target market for the 3090 won't be concerned with 350-400W power usage if the performance is there. It's like buying a Ferrari and looking at the MPG. Many of the existing AIB 2080ti cards already go up to 350W on OCs. I think the idle power consumption is more important, especially since the 2080/2080ti had some issues with that.

 

[Glo.]

And one more thing. Hasn't 102 die shot already leaked suggesting made by Samsung, and not by TSMC?

 

[n0x1ous]

And one more thing. Hasn't 102 die shot already leaked suggesting made by Samsung, and not by TSMC?

its said TW which doesnt tell us anything. Just means packaged in Taiwan

 

[TESKATLIPOKA]

"Damage control"? Wow. 350W with 35b xtors, 920gb/s and 24GB is allright. AMD has 300W with 12b xtors.

I find It a bit sad that you have to compare Ampere to an old Vega albeit on 7nm, just so you can say 350W is alright.
In my opinion 350W is too high and I don't care If It's a card from AMD or Nvidia.
With such a high power consumption in desktop what can we expect in laptops?

 

[A///]

Most of those people have cases or cooling solutions that solve the issue of that thermal output. At peak the 3090 will be dumping more heat into your case than 490 did over a decade ago.

 

[Glo.]

With such a high power consumption in desktop what can we expect in laptops?

Intel finds 400W is pretty OK, these days.

 

[jpiniero]

I find It a bit sad that you have to compare Ampere to an old Vega albeit on 7nm, just so you can say 350W is alright.
In my opinion 350W is too high and I don't care If It's a card from AMD or Nvidia.
With such a high power consumption in desktop what can we expect in laptops?

nVidia did refresh the laptop line in April. Perhaps they won't bother with Ampere, maybe wait for the Refresh or even Hopper?

 

[Glo.]

Uhhh, guys, why do you worry about Laptops?

Nvidia's lineup in laptops tops out at 104 die. TU104 based RTX 2080 Super was 260W TDP GPU.

The same dies, with full 48 SM's landed in laptops, and nobody blinked even once about it. Why? Because those GPUs still were underclocked, to fit in laptops.

Current outlook says that 104 die still has 250W. So why worry about it?

 

[A///]

That's weird. He said bulk of GPUs, not a specific class set of GPUs. He didn't lie, but he also wasn't specific. Professional and scientific could get the 7nm bulk mentioned, with regular consumer getting Samsung.

Did JHH name drop GeForce GPUS or just GPUS in general?

 

[TESKATLIPOKA]

Uhhh, guys, why do you worry about Laptops?

Nvidia's lineup in laptops tops out at 104 die. TU104 based RTX 2080 Super was 260W TDP GPU.

The same dies, with full 48 SM's landed in laptops, and nobody blinked even once about it. Why? Because those GPUs still were underclocked, to fit in laptops.

Current outlook says that 104 die still has 250W. So why worry about it?

It will fit, but the performance increase over the previous generation will be smaller compared to desktop.

Edit: maybe I am mistaken, considering this is built on better newer process, so It can have better frequency/voltage characteristic at lower clockspeeds compared to previous generation than at higher clockspeeds.

 

[Glo.]

That's weird. He said bulk of GPUs, not a specific class set of GPUs. He didn't lie, but he also wasn't specific. Professional and scientific could get the 7nm bulk mentioned, with regular consumer getting Samsung.

Did JHH name drop GeForce GPUS or just GPUS in general?

He said majority of 7 nm PRODUCTS is going to TSMC.

 

[A///]

He said majority of 7 nm PRODUCTS is going to TSMC.

Ok. So Quadros, Titan (if there is one) and anything DC based. Should make up at least half of their 7/8nm products. People losing their minds over a claim when the wording is enough to throw them off. He he said GeForce products most people would have a point. I could be wrong and will admit to such, but I've seen enough smart marketing talk in my professional worklife to pick up on subtle crap like this.

 

[AtenRa]

Ok. So Quadros, Titan (if there is one) and anything DC based. Should make up at least half of their 7/8nm products.

Quadros and Titans using the same chips as Consumer products

 

[AtenRa]

So if GA102 at 7nm TSMC with 34.5B transistor at 628mm2 , then they use the same libraries as A100.

And since A100 clocks are way low, increasing the clocks in GA102 also increase the power Consumption dramatically, hence the higher TDP

 

[sontin]

That is still wrong. TSMC's 7nm is about density and not clock rate. Look at every SoC produced on it. All companies have been gone wide with a low clock rate.

And 1,7GHz with 34,5B xtors is very high. That is Turing level with an density increase of 2,2x.

 

[A///]

Quadros and Titans using the same chips as Consumer products

And the RTX6000 and the 2080ti use the same TU102, however, the former outperforms the latter in performance and memory size while having a minimal increase in energy use at peak load. You also missed the point of my post. So I'll reiterate it.

Geforce: Samsung
Quadro/Titan/DC: TSMC

 

[A///]

So if GA102 at 7nm TSMC with 34.5B transistor at 628mm2 , then they use the same libraries as A100.

And since A100 clocks are way low, increasing the clocks in GA102 also increase the power Consumption dramatically, hence the higher TDP

Technically, nothing is stopping them from requiring a 2x 12 pin power plus board power for a theoretical 675 watts available to a card. Maybe next generation. Hopper is supposedly going to be chiplet based, which should be interesting if the rumors are true.

 

[Glo.]

If 8 nm theoretical maximum Transistor Density is 61 mln/mm2, then it is still possible to achieve 627 mm2 die, with 35 bln transistors.

Its still 55 mln xTors/mm2.

That density however explains the power draw of 3090.

 

[DXDiag]

The denial in this thread is strong.

 

[Kenmitch]

Dang....Ran out of popcorn again.

This thread is like a broken record.

 

[A///]

If 8nm Samsung is as bad as the rumors for the last year and a half have been, then I can't imagine them using that for their higher end professional cards if Geforce gets the limited TSMC order. You don't put leaky chips in solutions that aren't mean to suck up power. Even in a workstation, extra watts are simply a waste, let alone DC.

 

[DDH]

40% increase in tgp for a 20% increase in SMs? Maybe they went all in on RT cores.
A almost 30% increase in tgp for the same SMs on the 3080?

3080 with only 10gbs vram? If they could do 2gb stacks on the 3090, why not the 3080?



Sent from my SM-N975F using Tapatalk

 

[DDH]

The denial in this thread is strong.

A number of reliable leakers stated 8nm with confidence.

Now another source has said 7nm. You are doing the same as those in denial" if you believe one over the other


Sent from my SM-N975F using Tapatalk

 

[Glo.]

A number of reliable leakers stated 8nm with confidence.

Now another source has said 7nm. You are doing the same as those in denial" if you believe one over the other


Sent from my SM-N975F using Tapatalk

Even Kopite, himself is not sure about it. Also, he is sceptical about RTX 3070 info:

https://twitter.com/x/status/1299350091795329025

 

[reb0rn]

3070 no gddr6x ???? shite 8nm, for first time in past 20 years AMD here I come!

NV root in hell