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.

 

[Gideon]

But it will never look as good, the tradeoff is going to be a janky ride, from almost nice upscaling, to quite bad. We are cheating ourselves to higher framerates, but it looks good on paper though... "4k with RTX and DLSS 3.0" WOWZA! But all i hear is a fancy phrase for "1440p upscaled with crappy image and higher FPS" But we make it look like it's all great features and clearly superior 4k perf. to the competition.

Well not quite, did you actually look at the videos (especially the last one) ? For Instance at times Death Stranding looks better with DLSS than native resolution with Temporal AA (that's built into the engine). Even in that game though it has artefacts with smoke and some buildings.

I agree that this tech is sometimes overhyped but to say: "crappy image" is definitely hyperbole.

Plenty of people would accept the image-quality loss if this allows you to play on High/Ultra quality in upscaled 1440p/1080p instead of Low 4K, or allows to turn raytracing effects on.

I'm personally really looking forward to some kind of image-reconstruction on RDNA2 hardware as well (considering DLSS 1.9 worked on shader cores and still had considering perfomance uptick, there's bound do be something on consoles at some point).

 

[coercitiv]

It pretty much indicates, that 40-45% Rasterization increase may be spot on.

Just like I predicted...

Careful there, the Y axis has no unit of measure. What if it's... watts?!

I'll see myself out.

 

[AtenRa]

Wow, so a ~2x perfomance increase with DLSS and RTX @ 4K.

Impressive for the settings, but considering it has more than 4x the tensor cores,new RTX cores and 50% more bandwidth (that RT really needs) it's the best-case scenario ...

Should make it abundantly clear that pure rasterization improvement can't really be much more than ~50%, at best.

Actually GA102 has less Tensor cores (384) vs TU102 (544), but due to 4x Higher Throughput per Tensor Core in Ampere it makes the GA102 have 2.8x the Tensor Core Throughput vs TU102.
Ampere only has a 4x Tensor Cores per SM when Turing has 8x Tensor Cores per SM.

edit: correction on Tensor cores per SM between Ampere and Turing.

 

[Gideon]

Actually GA102 has less Tensor cores (348) vs TU102 (544), but due to 4x Higher Throughput per Tensor Core in Ampere it makes the GA102 have 2.8x the Tensor Core Throughput vs TU102.
Ampere only has a single Tensor Core per SM when Turing has 2x Tensor Cores per SM.

Thank you for the correction, makes sense.

 

[tviceman]

2x increase in RT performance is great, but it needs to be backed up with a 60+% rasterization performance increase too. The TDP is up 40% from Turing, which is up 10% from Pascal.

A 40-50% perf/w improvement would being two back to back generations of little to no perf/w improvements, which would be incredibly terrible.

 

[Krteq]

https://twitter.com/x/status/1300373469843120133

Well, seems to be a fake

- source: Chiphell forums
- card render seems to be much smaller than RTX 3090 card on photos leaked before
- "2080Ti" instead of "2080 Ti"
- "Minecraft RTX" instead of "Minecraft with RTX"

 

[DiogoDX]

Well, seems to be a fake

- source: Chiphell forums
- card render seems to be much smaller than RTX 3090 card on photos leaked before
- "2080Ti" instead of "2080 Ti"
- "Minecraft RTX" instead of "Minecraft with RTX"

Nvidia calls it Minecraft RTX. Quake 2 too.

 

[exquisitechar]

Well, seems to be a fake

- source: Chiphell forums
- card render seems to be much smaller than RTX 3090 card on photos leaked before
- "2080Ti" instead of "2080 Ti"
- "Minecraft RTX" instead of "Minecraft with RTX"

Yeah, that's a fake. Not a very good one either.

 

[Krteq]

Nvidia calls it Minecraft RTX. Quake 2 too.

Nope, it's "Minecraft with RTX" officially

NVIDIA - PLAY MINECRAFT WITH RTX

 

[DXDiag]

But it will never look as good, the tradeoff is going to be a janky ride, from almost nice upscaling, to quite bad.

Check again, DLSS 2 provides equal or better IQ than native resolution + TAA, DLSS 2 achieves near perfect AA quality, is temporally stable, and is sharper during motion, it's the trifecta.

 

[DiogoDX]

Nope, it's "Minecraft with RTX" officially

NVIDIA - PLAY MINECRAFT WITH RTX

I remember in the old videos they calling Minecraft RTX but seems that the oficial release name is "with RTX". Thanks.

 

[Stuka87]

Check again, DLSS 2 provides equal or better IQ than native resolution + TAA, DLSS 2 achieves near perfect AA quality, is temporally stable, and is sharper during motion, it's the trifecta.

Yeah, we will just ignore that things like text are often unreadable, or that the games visuals are being manipulated away from what the original game makers intended. Or that it offers up to 20% worse performance than regular upscaling because regular upscaling is free, DLSS is not. Most games take a 15% performance hit when compared to upscaling, some games are closer to 20%.

The only purpose of DLSS is marketing. And some people buy into it hook, line, and sinker.

 

[kurosaki]

are being manipulated away from what the original game makers intended. Or that it offers up to 20% worse performance than regular upscaling because regular upscaling is free, DLSS is not. Most games take a 15% performance hit when compared to upscaling, some games are closer to 20%.

The only purpose of DLSS is marketing. And some people buy into it hook, line, and sinker.

"But, but, look at these specific, highly optimized, clips I found on the internet. It seems more than perfect! "

 

[Gideon]

DLSS 2 achieves near perfect AA quality, is temporally stable, and is sharper during motion, it's the trifecta.

It sure is a very impressive tech when it works (which seems to be more and more often) But occasionally it still has odd artifacts in some places, I personally find very irritating:


1. On the posts and all around:

2. The overhanging wire:

3. The edge behind the glass is flickering in the middle of the screen. Scare warning!
https://youtu.be/9jRR8zB4U5g?t=705

Now I'm sure many people won't probably even notice these and many do claim it works flawlessly for them. Bugt being an Image Quality freak I really hate seeing incorrect behaviour like that.

Granted, considering the huge improvement Nvidia has made with DLSS since 1.0, it will certainly improve in newer games.

I'd still enable it in most games that support it, as the upsides are too good to ignore (especially if it meant 4K playable with Raytracing vs unplayable), but Death Stranding is an example of one that doesn't quite cut it (for me) yet.

 

[Dribble]

But it will never look as good, the tradeoff is going to be a janky ride, from almost nice upscaling, to quite bad. We are cheating ourselves to higher framerates, but it looks good on paper though... "4k with RTX and DLSS 3.0" WOWZA! But all i hear is a fancy phrase for "1440p upscaled with crappy image and higher FPS" But we make it look like it's all great features and clearly superior 4k perf. to the competition.

It looks better in death stranding then native, the horse you are still attempting to flog died a while ago.

 

[A///]

I saw it earlier on Twitter. It looked sketch.

No /sarcasm tag, so I presume you are serious.
Well, if that attitude is common then AMD might as well exit the GPU business.
If "I wish AMD could compete at the high end" is just code for "I want them to compete to bring the prices down so I can buy Nvidia cheaper" then that's not a viable business model.

Yeah, I get it. I don't want to waste days, weeks or months waiting for AMD to fix their mistakes. I'm not the only one. I don't mind AMD CPU teething issues. Those are addressed rather quickly. Their RTG dept. doesn't get that same love.

 

[Glo.]

https://twitter.com/x/status/1300449249449172992

 

[Krteq]

Hmm, if that's related to Ampere customer cards I really don't understand those TDP values

 

[NTMBK]

It sure is a very impressive tech when it works (which seems to be more and more often) But occasionally it still has odd artifacts in some places, I personally find very irritating:


1. On the posts and all around:

2. The overhanging wire:

3. The edge behind the glass is flickering in the middle of the screen. Scare warning!
https://youtu.be/9jRR8zB4U5g?t=705

Now I'm sure many people won't probably even notice these and many do claim it works flawlessly for them. Bugt being an Image Quality freak I really hate seeing incorrect behaviour like that.

Granted, considering the huge improvement Nvidia has made with DLSS since 1.0, it will certainly improve in newer games.

I'd still enable it in most games that support it, as the upsides are too good to ignore (especially if it meant 4K playable with Raytracing vs unplayable), but Death Stranding is an example of one that doesn't quite cut it (for me) yet.

Yikes... that sort of failure case worries me. That's the problem with these DL systems- at the end of the day it's a black box. You can't look at the algorithm and use logic to derive why it's failing in specific cases- you're at the whims of whatever dataset the thing was trained on.

 

[MrTeal]

https://twitter.com/x/status/1300449249449172992

That's pretty vague and appears to be talking about Xe. Do you have a reason to think it's about Ampere?

 

[linkgoron]

That's pretty vague and appears to be talking about Xe. Do you have a reason to think it's about Ampere?

https://twitter.com/x/status/1300451298945503232

 

[sontin]

https://twitter.com/x/status/1300449490416144384

But i think is Intel, too.

 

[MrTeal]

https://twitter.com/x/status/1300449490416144384

But i think is Intel, too.

LOL... Thanks. The guy's twitter feed didn't exactly make his meaning clear.

 

[TESKATLIPOKA]

https://twitter.com/x/status/1300449249449172992

If this wa true for rasterization performance and the performance was 50% better than the previous generation, then Ampere should have ~200W TDP.
I think It's for RT including DLSS or something.

 

[Konan]

Yeah baby 24 hours to go!
DLSS + RTX Vs same on Turing

https://twitter.com/x/status/1300449953387601921

https://twitter.com/x/status/1300449249449172992