$30,000 NVIDIA graphics card loses to integrated graphics in games
It turned out that the most powerful computing system NVIDIA H100 can not even cope with old video games
Although computing systems like the NVIDIA H100 are technically categorized as graphics processors and are video cards, they can hardly generate graphic content – there is not enough component base for the corresponding manipulations. But this does not prevent enthusiasts from comparing the performance of two completely different devices. For example, a user with the nickname Geekerwan decided to test the most powerful NVIDIA solution with a price tag of 30 thousand dollars in modern video games and benchmarks in order to compare this model with “ordinary” video cards, and found out that even integrated graphics can leave the expensive H100 far behind.
The video is in Chinese and without subtitles, but the meaning can be understood by FPS
It is worth noting that the NVIDIA H100 video card is based on the GH100 processor with 14,592 CUDA cores and support for FP64, TF32, FP32, FP16, INT8 and FP8. For example, the AD102 processor found in the RTX 4090 has 18,176 CUDA cores, but only supports the FP32 format. At the same time, NVIDIA’s $30,000 data center doesn’t even have a built-in graphics display solution, plus the manufacturer obviously doesn’t optimize drivers for video games and benchmarks. As a result, the H100 in the 3DMark Time Spy benchmark showed only 2,681 points, which is less than the integrated graphics AMD Radeon 680M, which was introduced last summer (it scores 2,710 points).
But tests in real scenarios with video games turned out to be much more interesting – an NVIDIA video card at the price of a good crossover produced only 9 FPS in 4K resolution in Red Dead Redemption 2. Next, the enthusiast lowered the resolution to 1080p, turned on DLSS, and lowered the graphics settings a bit to eventually see a more or less stable frame rate of 30 FPS. Moreover, in tests aimed at rendering graphics, the RTX 4090 graphics card was much faster – it generated a scene more than twice as fast as the H100. Accordingly, we can safely say that this model is not suitable for either games or professional work with graphics.
However, the H100 graphics card should not be able to handle games or graphics rendering – it was created for something else. NVIDIA designed this solution for artificial intelligence and complex computing processes in data centers, not for running shooters and racing games. And in its field, the H100 is definitely the best solution on the market with insane possibilities, especially when using a stack of multiple graphics cards.