If the information is true, it will not be long before NVIDIA presents Hopper for the first time, a GPU architecture for HPC that brings with it many unknowns and novelties. The first of them is to know if we are facing the first graphics card built by chiplets and, therefore, by several chips instead of just one. The second is about the rumors of its internal organization, which will undergo a major change, so much so that one like this had not been seen since the days of the GTX 8800.
Hopper’s specifications according to the NVIDIA leak
The latest rumors about the Hopper architecture is that it will be completely monolithic and therefore composed of a single chip. In any case, we should clear up doubts in a few weeks, since NVIDIA will soon celebrate its GTC where its new and mammoth graphics for HPC is expected to be presented. And what does the NVIDIA leak bring us regarding the GH100? Well, for starters we can know its configuration.
- 8 CPGs in total.
- 9 TPC per CPG.
- 2 SM per TPC.
which gives us a total of 144 sm, the same figure as the GA100, but we must bear in mind that these GPUs are not designed for the gaming market and within their SM they replace some elements with others compared to those used in the RTX. For example, the intersection unit for Ray Tracing, the RT Core, is of little use in scientific calculations, so it is usually removed. At the other extreme we have 64-bit floating point units, which are superfluous in gaming, but not in areas where data accuracy is very important.
A configuration of 144 SM may seem small compared to the 220 Compute Units of the MI200 and its dual GPU configuration. However, it is more than possible that NVIDIA will do the same than with the RTX 30 and double the amount of floating point units, commuting them with those of integers. So in the end the amount of compute unit in the GPU would end up being higher than even AMD’s design. The performance that is expected? 3 times larger than the current GA100, something impossible without doubling these units.
New unit and organization at Hopper: the CPC
However, there could be changes in the organization in Hopper according to the NVIDIA leak, since it would be introduced a unit between the TPC and the GPC called CPC, which would be composed of 3 CPCs. Which makes us suppose that for the new architecture they could implement an additional cache level between the L1 of the SM and the global L2 of the GPU. Something that would not be impossible from the moment that is what AMD has done with its RDNA and it is a way to relieve the stress on the L2, especially in terms of read requests. A change that would make the architecture totally different from what we expect in the RTX 40 for gaming.
NVIDIA Hopper: very limited gaming capabilities
Although we call them graphics cards, these pieces of hardware use the GPU cores for large-scale parallel calculations, as they are inherently more efficient than a CPU in those parts of the algorithms. Thus, instead of increasing the number of processor cores, what is done is to use the card as a coprocessor, not to generate images on the screen, but to achieve mathematical results at a much higher speed.
We have already seen how, for example, AMD has removed all the graphics capabilities from CDNA, well, according to the NVIDIA leak in Hopper, the same thing will happen in all GPCs except for the first one, which will be the only one that can generate graphics. This makes us think that it will have asymmetric GPCs where the fixed function units will only be present in the first one. So it will have basic graphics capabilities, but when it comes to playing games it will be quite limited and at a much worse level than an RTX 4050, therefore, NVIDIA Hopper is not intended or intended for gaming.
