GDC is approaching and so it is expected that as with the AMD financing event, NVIDIA will make its Ampere / Hooper build community visible to the public. We have seen many leaks but at the moment nothing is possible to verify 100% of the data.
Now, yet another filtration is making the appearance based on its opposition to the GA100 chip, which should be very strong for the NVIDIA card range.
The NVIDIA GA100 beast will consume 300 watts
The new GA100 will be a hit from the start, when the company launches it in the market and for the first time as an HPC GPU, that is, a graphics card for data centers and so on.
It will achieve a 7 nm lithographic process made by TSMC, which matches its AMD competitor and where it will have a much higher wavelength depending on the SM count. And all indicates there will be 128 units of NVIDIA minimum for this chip, which would give us a total and maximum total of 8192 Shaders.
As for the VRAM configuration, it seems that two different variants and configured in completely different packages will be offered first, since they can be purchased at 24 and 48 GB of HBM2E. This is evidenced by the lack of SK Hynix's catalog, that Samsung will be responsible for supplying the stack to the company.
Due to this type of VRAM, usage will approach only 300 watts, a high value but it is contained if we look at the card type and its benefits.
Ampere will arrive more often than expected
And NVIDIA plans to boost frequency too, as AMD has done with Navi. Sources reveal this GA100 speed of 2.2 GHz as usual, which is quite impressive considering that the size of the complexity and features, reduced the final speed.
This means that versions of the game can go through this period as usual, which may be a bit difficult to match with AMD. They came back to give a theory of performance theory, where this GA100 based on Ampere as its construction could reach 32 TFLOPS performance on FP32.
To sum it up, the current NVIDIA TITAN RTX runs 16.31 TFLOPS at 4608 Shaders. That is, the new GA100 with less than double Shaders is almost double the floating power.
Undoubtedly, AMD should upgrade its rDNA 2 to mount NVIDIA if this specification is true in the end. In any case, it does not matter if the prices are up again and the price tags go unnoticed, a major barrier to the sale of current models such as the RTX 2080 Ti.
