Budget was reallocated to the RT core. Might even call it a level 4 RT core with the features.
Interesting how they basically went back to Maxwell. Why is that?The oversll INT throughput is doubled but the capability to concurrently issue INT that Turing brought is seemingly lost. I guess just...
Chiplets and so on don't add performance right? So on 4NP, Nvidia would have had to make a chip the size of GB200 and sell for a tenth of the price. Not happening.
The only alternative for performance is to use TSMC N3E, which I assume was too late (Apple and Intel didn't get satisfactory...
Grounding AI in proper physics with proper behaviour is challenging. Doing it in real time, something else. So using a ground truth of the frames you are already generating seems like a good compromise. Then you can add speedups like denoising, neural radiance caching, etc. That's how I...
We're just back to regular scheduled doom posting. Personally I say 30-45% for 5090 and 25-40% for 5080. Primarily from architecture and minor spec bumps in some cases. I doubt clocks are any better in practice vs 2.8ghz for the 4090. Would interpolating the first frame and extrapolating the...
He said that the SM changed. How they changed does not necessarily increase TFLOPs (Remember Turing HALVED TFLOPS per SM compared to Pascal gtx 10 series)
That's not a blanket thing. That's why we always talk about "light and "heavy" RT. We have a game where 7900XTX surpasses 4090 even with RT on and when we look at the RT, its only 1 effect, with not even full resolution, and all the GPUs are very close together in performance
Jensen said something interesting there about the changes in the SM.
"Two dual shaders, one is for floating point and one is for integer" Which he also clarified saying something like 'concurrent execution. Equal performance for both'. Which sounds like an evolution of Turing and Ampere, with...
Makes sense since using current rumors as an indication, that's just GB202 and GB203 covering all these cards, although GB205 seems absent from rumors. Maybe 5060ti alone is on GB205 and 5060 on GB206. Or 5070 ends up on GB205 with 12GB after all
Too big an update for a refresh. No need (has nothing to do with AI focus or whatever), there is simply no need for a generational uplift in memory bandwidth when the normal strategy of using your salvaged chips at a slight hit to priceshas worked so well so far
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.