Nvidia Blackwell Architecture Speculation

  • Thread starter Deleted member 2197
  • Start date
DegustatoR said:
Every result where it's more than +15% (half of flops gain) is highly likely to be a result helped by the additional memory b/w.
Click to expand...
So in a hypothetical world where the bandwidth had increased by the same amount as the FLOPS gain (~32%) to maintain the same ratio as the 4090, what performance difference would we see?
 
Subtlesnake said:
So in a hypothetical world where the bandwidth had increased by the same amount as the FLOPS gain (~32%) to maintain the same ratio as the 4090, what performance difference would we see?
Click to expand...
I'm sure someone reading this will get a 5090. It would be very interesting to downclock the memory and see how performance scales. But it's only interesting if you post the results here :yes:
 
Subtlesnake said:
So in a hypothetical world where the bandwidth had increased by the same amount as the FLOPS gain (~32%) to maintain the same ratio as the 4090, what performance difference would we see?
Click to expand...
On average? Doubt that the results would be much different. Some games which are showing +50% now though would start showing something closer to +30.
 
DegustatoR said:
On average? Doubt that the results would be much different. Some games which are showing +50% now though would start showing something closer to +30.
Click to expand...
Ok so the question is whether on average the 5070 Ti and 5070 will see the ~20% improvements suggested by the slides, or Nvidia just cherry picked games that benefit from the extra bandwidth.
 
Subtlesnake said:
Ok so the question is whether on average the 5070 Ti and 5070 will see the ~20% improvements suggested by the slides, or Nvidia just cherry picked games that benefit from the extra bandwidth.
Click to expand...
We'll know the answer to that in several days since what 5080 will show will likely be very similar to what 5070Ti will show and - probably - 5070 too.
 
Just purely speculating here but I wonder if GB202 has some experimentation in terms of how to approach a MCM design. This would be akin to how they approached preceeding GPUs before actually moving to MCM with GB200.
 
It's interesting it seems like the CNN model uses less VRAM on the above? I'm wondering if the transformer models leverage FP8 (or INT8) more which wouldn't benefit Ampere (and presumbly Turing as well).
 
Tried the new transformer model with an old game that had a weak DLSS implementation (well that, and I'm currently replaying it) - Rise of the Tomb Raider. 4K, custom settings, DLSS performance, motion blur off.

Transformer DLSS4:


CNN (DLSS 3.8.10):


And uh, quite a downgrade unfortunately. Basically Performance mode is equivalent to CNN's Balanced in fps for this title on my card, but with significantly increased aliasing/shimmering to boot.

So the performance cost on lower-end cards may be significant. Not too bothered by the quality downgrade as it's early days and obviously not every title will play well with a new model (and especially one that didn't have even TAA to begin with and DLSS was slapped in with a patch) hence the toggle we're going to get.
 
Last edited:
An interesting connotation to be sure, thank you for posting it! It does indeed make us wonder if it's an app specific issue, or maybe an AI training issue, or something else entirely... I'm sure more examples will surface over the next few weeks and I'm intrigued by what we are about to discover.
 
trinibwoy said:
I did not. Please read again.
Click to expand...
'in games where it matters' is the only way you could wiggle out of this, except that's not how it works. In games where the gains are lower, it doesn't mean those games matter less in context of a GPU review. Come on now. That same reasoning could be used to present all manner of dishonest claims. I could even make the exact opposite claim and say that examples where the gains are only like 15% are actually the ones that matter, and so there's no meat on the bones whatsoever. I wouldn't do that though, cuz it wouldn't be an honest portrayal.
 
trinibwoy said:
That is actually how English works. Please move on.
Click to expand...
You literally cut off my whole comment to ignore the part where I say you cannot just conveniently choose which games matter when making performance claims. Not helping yourself at all in the 'honesty' arena here. smh

If you would like to move on and deflect from your misleading claims, you are free to ignore my posts. Dont tell me what I should or should not be responding to, ffs.
 
DSOG review of DLSS4 Frame Gen.

Overall, DLSS 4 Multi-Frame is one of the best new features of the Blackwell RTX 50 series GPUs. By using it, you will finally be able to enjoy path-traced games at super high framerates. And you know what? I’d take 200FPS with the response times of 50-60FPS any day over simply gaming at 50-60FPS.
Click to expand...
And good luck getting a similar experience with “TV interpolation” or Lossless Scaling. DLSS 4 is on an entire next-gen level.
Click to expand...
So go ahead and cope all you want, make all the excuses or memes you can. Personally, after actually getting my hands on it and gaming with DLSS 4 X4, I’ll be using it in pretty much all the games that support it!
Click to expand...

www.dsogaming.com

NVIDIA DLSS 4 Multi-Frame Gen Benchmarks & Impressions

Overall, DLSS 4 Multi-Frame is one of the best new features of the NVIDIA GeForce Blackwell RTX 50 series GPUs.
www.dsogaming.com www.dsogaming.com
 
The new model while improved has a different set of artifacts and problems that are very clearly visible. Also there is an increased performance cost on the 4080 super.
 
You must log in or register to reply here.
Back
Top