Nvidia's $40K AI GPU fails at gaming, but that’s to be expected

gamesradar.com:

For whatever reason, someone has decided to use Nvidia’s super expensive AI GPU to play PC games, and the results are predictably dire. While it sounds weird saying that the RTX 4060 could potentially run rings around a $40K graphics card, things are a bit clearer when you look at what the H100 is actually meant to be. Still, it’s fun to playfully roast cutting-edge tech, and new benchmarks naturally help with the dunking. 

It might not be too obvious, but despite being literally called graphics processing units, some of the best graphics cards out there aren’t built with producing visuals in mind. As such, it’s more appropriate to call expensive cards like the H100 Hopper an ‘AI-accelerator’ or ‘general purpose’ GPU, as they’re meant to live within advanced data centers rather than your gaming PC. 

Yet, that’s not to say you can run your Steam library using the Nvidia H100, as even though the card completely lacks HDMI and DisplayPort, Twitter user I_Leak_VN points out that YouTuber Geekerwan has managed to do so anyway. Rather than popping the card onto a PCIe slot and installing some drivers, the enthusiast had to add a fan to the card (as it doesn’t have one) and trick it into using another GPUs video output, but the end result is something that functions like a traditional graphics card.

I say traditional, but if the Nvidia H100’s benchmarks are anything to go by, it acts more like an iGPU than the likes of the GeForce RTX 4090. The card manages to “achieve” a 3DMark TimeSpy score of 2,618, which is less than the AMD Radeon 680M. Again, this is to be expected, as while the expensive part boasts 80GB HBM3 memory and 14,592 CUDA cores, it wields fewer raster operating units and lacks game-optimized drivers. 

Running