Although no benchmark or similar has been leaked, reality invites us to think that the performance jump that both NVIDIA and AMD pose is going to be, possibly, the highest in history compared to predecessor GPUs. In the case of NVIDIA we already saw that it would be a monolithic chip taken to the maximum with the 5 nm of TSMC, but in the case of AMD everything turns towards a modular architecture.
The leaked gives concrete details of what we can expect and for the first time we have a global perspective on which to stick, so let’s get to that and do a review on what had already been leaked before entering the matter.
As we well know, the chip will be called Navi 31 and will belong to the family GFX11It will be an MCM system where apparently we will have three chips, the two dies and GPUs to use and an I / O die that would do the functions of assigning resources and work as well as managing them between both chips. Why do we know this? Because AMD already has the first MI200 on the market and it is also leaked that the die will reach 5 nm while the I / O die would reach 6 nm, both under the TSMC umbrella.
This is theory, since AMD will dispense with HBM2E in these gaming GPUs and therefore would require, in principle and except for surprise, said I / O die. What seems to be equally confirmed are the 120 WGP with 15,360 Shaders that will be joined by some surprising 32 GB from GDDR6 at 18 Gbps already a bus of 256 bits.
AMD repeats formula and will integrate Infinity Cache, except that in this case it will be the new 3D version with up to 512 MB. Doesn’t this impress you? Well, the frequency will go up between 2.4 GHz and 2.5 GHz to finally give an approximate yield of 75 TFLOPS in FP32, which would be awesome.
Consumption of Navi 31, more efficient than AD102?
As for N31, I have no data and can only guess that the TBP is about 350-550w.🤔
– Greymon55 (@ greymon55) November 9, 2021
The data is enormous and although it could change to a greater or lesser extent, since there is still a lot of time left, the last thing that has been leaked would make AMD get a “win-win” against NVIDIA. The raw performance data already shows some advantage, but now we also know the consumption data, at least approximate.
Comparatively speaking and to place ourselves, the new NVIDIA AD102 chip should be between 450 or 650 watts, where we have already seen how apparently the latest rumors sentence it to 600 watts Finally. What can you achieve? Well, the first rumor with base figures the consumption of Navi 31 between 350 watts and 550 watts, surely and seeing the data described above much more towards the second than towards the first.
In any case, AMD would not only be more powerful on paper, but it could be more efficient and by a lot, which would put NVIDIA in check for the first time in the two main sectors where until now it was a leader: performance and consumption.
Are we facing a turn in the gaming GPU market? Can NVIDIA go as far as 85 TFLOPS according to the latest rumors by the filtered increase of the frequencies to 2.5 GHz? It is complicated, it seems that it will have more Shaders, more bandwidth, but it seems difficult to think that with those numbers it will reach the frequency that AMD will implement in two chips with lower voltages, lower Shaders per unit and better overall dissipation.