RTX 5090 lastly again on prime of GPU leaderboard, however it’s not the variant you anticipate

Final week, PassMark’s video card benchmark leaderboard confirmed the RTX 5090 struggling in opposition to its last-gen competitor, the RTX 4090, with the latter sitting on the prime with 23 extra factors. It wasn’t simply the RTX 5090 that carried out slower than anticipated, because the RTX 5070 Ti additionally couldn’t do a lot to dethrone the RTX 4070 GPUs. This made it seem to be the RTX 50 sequence both struggled in 3D rendering or one thing was hindering their efficiency.

It turns on the market was certainly a difficulty with the take a look at. PassMark was utilizing a 32-bit framework for testing, and for individuals who don’t know, Nvidia eliminated assist for 32-bit purposes on RTX 50 sequence GPUs, stating, “32-bit CUDA purposes are deprecated on GeForce RTX 50 sequence GPUs.” Since then, PassMark has launched a patch, updating the GPU Compute take a look at to make use of 64-bit elements, and the RTX 50 sequence GPUs have already taken their rightful positions on the leaderboard.

RTX 5090 takes the highest spot, however it’s the 5090 D variant

In keeping with the up to date leaderboard, the RTX 5090 has proven an enormous enchancment in comparison with the place it beforehand stood. The final time we reported on the GeForce RTX 5090 on this benchmark, its common PassMark rating was 38,407, however it has now gone as much as 38,827. For comparability, the RTX 4090 beforehand had a mean rating of 38,430 however presently sits at 38,421.

Curiously, whereas the RTX 5090 has taken the lead in opposition to the RTX 4090, it isn’t precisely on the prime spot, or at the least not the worldwide RTX 5090. As a substitute, the highest place is presently held by the China-only RTX 5090 D, which has outperformed the worldwide 5090 by over 7,000 factors. We aren’t precisely positive how the 5090 D has managed to create such an enormous hole, as each fashions have the identical specs other than the 5090 D having lowered AI efficiency, which shouldn’t have an effect on this benchmark in any manner.