The rumors don’t stop and in the midst of a GPU shortage, leaks keep coming on new graphics cards from NVIDIA and AMD. The data just released is as optimistic as it is in principle hard to believe, but if correct, we should expect a massive performance boost from both AMD and NVIDIA, where Ada lovelace It would not be the backbone of the Greens.
Once again, Kopite7Kimi to attack and once again the data provided leaves locals and strangers speechless. While we obviously cannot confirm or deny what has been leaked, this user’s reputation is already legendary and at the moment he has a virtually flawless success rate, but also take the following with a little salt and we’re going to have a mess.
NVIDIA and AMD would improve performance with Ada Lovelace and RDNA 3
-5nm (no matter TSMC or SEC)
-AD102 in transition or GH202 in revolution?
-RDNA3, GFX11, how is AMD achieving its performance target? Double the SIMD in CU?
-And Intel?– kopite7kimi (@ kopite7kimi) May 22, 2021
The data is as concrete as it is astounding, and it looks like the new thing from both companies would happen in a 5nm node that is unclear if it would be TSMC or Samsung, both with EUV and with very similar density, a only mask and a jump in density and efficiency more than interesting.
But the data doesn’t end there, since NVIDIA’s latest would arrive with Ada Lovelace, but it wouldn’t be the most powerful, but a transitional architecture in front of the main course: Hopper. Instead, AMD would focus solely on RDNA 3, its new architecture which, as we already know, promises 50% more performance / watt.
The leaked figures are really hard to believe, at least initially: NVIDIA would increase performance 2.2 times with Ada lovelace and 3 times the performance with Hopper over the current ampere, while AMD would achieve 2.5 times the performance of RDNA 2 with RDNA 3. Does this sound difficult? Well, a priori, of course, they seem very optimistic figures, but in addition, another figure has been filtered out which makes the previous ones pale: an FP32 performance of 100 TFLOPS.
Does NVIDIA have a trump card up its sleeve with Hopper?
Taking into account that the current performance of the RTX 3090 is nothing less than 35.58 TFLOPS (unreal and marked by the controversy of dual-function Shaders), we are talking about three times the current performance, as unreal as they may be.
How could this be possible? Behind the scenes, NVIDIA and AMD have reportedly studied the performance of their new chips with advanced lithographic processes for a while and concluded that it can scale at a good pace in power, but they find a very clear bottleneck. throttling: memory bandwidth.
NVIDIA has mitigated this with GDDR6X and AMD with their Infinity Cache, but we’re not sure exactly how they’re going to fix it. Ada Lovelace and RDNA 3. We are talking about a return to HBM2 or even new cache systems with GDDR6X to ensure that the jump in performance is not reduced by memory and especially not to draw on the final cost of each GPU.
In any case, there is almost a year left for the presentation of the two architectures, where Hopper remains the great unknown, since NVIDIA could advance Ada Lovelace by the end of the year and in 2022 launch Hopper with the architecture. MCM to get ahead of AMD, if RDNA 3 is not already MCM. We will have to wait.
The post The next NVIDIA and AMD GPUs: at 5nm, double the performance and MCM? first appeared on HardZone.