Why Nvidia’s new GPU performs worse than built-in graphics


One would assume {that a} $40,000+ GPU can be the perfect graphics card for gaming, however the reality is extra difficult than that. Actually, this Nvidia GPU can’t sustain with built-in graphics options.

Now, earlier than you get too upset, it is best to know that I’m referring to Nvidia’s H100, which homes the GH100 (Grace Hopper) chip. It’s a robust knowledge middle GPU designed to deal with high-performance computing (HPC) duties – not PC gaming. It doesn’t have any show outputs, and regardless of its huge capabilities, it additionally doesn’t have any coolers. It is because, once more, you’ll discover this GPU in an information middle or server setup, the place it is going to be cooled utilizing {powerful} exterior followers.

Whereas it has “solely” 14,592 CUDA cores (which is lower than the RTX 4090), it additionally has an insane quantity of VRAM and an enormous bus. In complete, the GPU accommodates 80 GB of HBM2e reminiscence, divided into 5 HBM clusters, every linked to a 1024-bit bus. Not like Nvidia’s shopper GPUs, this card additionally nonetheless has NVLink, which implies it may be linked to work seamlessly in multi-GPU methods.

The query stays: why precisely is such a GPU so dangerous generally and gaming use?

To show the case, YouTuber Gamerwan 4 of those H100 graphics playing cards for testing, and I made a decision to place one in a daily Home windows system to test its efficiency. This was a PCIe 5.0 mannequin, and needed to be paired with the RTX 4090 as a consequence of a scarcity of show outputs. Gamerwan additionally 3D printed an exterior cooler particularly designed to maintain the GPU working easily.

It takes a bit of labor for the system to acknowledge the H100 as a good GPU, however as soon as Gamerwan managed to recover from the hurdles it additionally managed to get ray tracing assist working. Nevertheless, as we found later throughout testing, there isn’t a lot assist for anything on a non-datacenter platform.

On a 3DMark Time Spy benchmark check, the GPU solely hit 2,681 factors. For comparability, the typical rating for the RTX 4090 is 30,353 factors. This rating places the H100 someplace between the GTX 1050 and GTX 1060 for customers. Extra importantly, it’s nearly the identical as AMD’s Radeon 680M, which is an built-in GPU.

Gaming checks had been additionally poor, with the graphics card hitting a mean of 8 frames per second (fps) within the recreation Purple Lifeless Redemption 2. The shortage of software program assist rears its ugly head right here — though the H100 can run at a most of 350 watts, the system by no means appears to exceed 100 watts, which ends up in a big drop in efficiency.


There are a number of totally different causes for this poor show of gaming powers. For instance, whereas the H100 is a super-powerful graphics card on paper, it’s very totally different architecturally from the AD102 GPU powering the RTX 4090. It solely has 24 ROPs, which is kind of a bit. A big drop from the 160 ROPs that the RTX 4090 has. Moreover, solely 4 out of the 112 texture processing (TPC) clusters can render graphics workloads.

Nvidia’s shopper GPUs obtain plenty of assist from the software program aspect to be able to run effectively. This contains drivers, but additionally system assist from builders – each in video games and in modular software program. There aren’t any drivers that optimize the efficiency of this GPU for gaming, and the consequence, as you may see, is fairly dangerous.

We’ve already seen the ability of drivers with the Intel Arc, the place the {hardware} remained the identical, however improved driver assist offered efficiency positive factors that made the Arc an appropriate possibility should you’re shopping for a finances GPU. With no Nvidia Recreation Prepared drivers and a scarcity of entry to the remainder of Nvidia’s software program suite (together with the at all times superb DLSS 3), the H100 is a $40,000 GPU that has no enterprise working any type of recreation.

At its core, this can be a computing GPU and never a graphics card in the identical method most of us realize it. Constructed for every type of HPC duties, with a powerful deal with AI workloads. Nvidia maintains a powerful lead over AMD in the case of AI, and playing cards just like the H100 play an enormous half in that.

Editors’ suggestions