Gaming Tested on NVIDIA H100 Hopper GPU: Slower Than Integrated GPU

With the least attention for graphics-intensive gaming tasks, NVIDIA’s H100 Hopper GPU is a chip built for pure AI and other computing workloads. Even still, it will be intriguing to see how this 30,000 USD GPU compares to other gaming GPUs and whether or not it can actually run games. As the Chinese YouTube channel Geekerwan points out, it turns out that it is technically possible but does not make much sense. The H100 PCIe version tested here can produce 204.9 TeraFLOPS at FP16, 51.22 TeraFLOPS at FP32, and 25.61 TeraFLOPS at FP64 based on the GH100 GPU SKU with 14,592 CUDA, with its inherent power lying in accelerating AI applications.

But how does it do in gaming tests? The testing demonstrates that not very effectively. In 3DMark Time Spy, it received 2681 points, which is less than AMD’s built-in Radeon 680M, which earned 2710 points. It’s interesting to note that the GA102, the highest-end gaming GPU SKU, has 112 ROPs compared to the GH100’s 24 ROPs. It follows naturally from this and paints a clear picture of why the H100 GPU is exclusively utilized for computing. Since it lacks any display outputs, the system required a second conventional GPU to supply the image while the H100 GPU handled the calculation.