Recently, NVIDIA spawned a second variant of the RTX 5090 D (intended for Chinese markets as part of US export controls) aptly named RTX 5090 D v2, which notably chops the memory capacity down to two-thirds of the original, at 24GB. As it turns out, even this didn’t meaningfully change the gaming performance of this GPU, according to tests carried out by Chinese-based Yesky.
RTX 5090 D v2 Is Just As Powerful
The reviewer took the INNO3D’s iChill version of the card to evaluate the GPU performance – this is the part where we remind you that despite the difference in name, the RTX 5090 D v2 effectively uses the same silicon as the original RTX 5090 D (and the RTX 5090), with all 21,760 CUDA cores onboard and 575W of TGP to work with. In other words, if a game is not bottlenecked by the lack of memory capacity or bandwidth in the export-compliant version (since the bus is cut down to 384-bit), the performance should behave exactly the same.
As expected, the results from the 3DMark synthetic tests and some of the gaming tests showed the performance of the RTX 5090 D v2 is virtually the same as the RTX 5090 D (which is known to match RTX 5090’s performance sans AI), losing just single-digit’s worth of framerates in all cases. The reviewer further noted (translated): “put it this way, there is virtually no performance difference between the RTX 5090 D and v2 version, the differences seen in games might be less than what a user may gain from manual overclocking.”
Still, Yesky pointed out that in scenarios like running games in 8K resolution or using high-resolution textures, the performance difference between 32GB and 24GB of VRAM could be manifested; for any scenario that doesn’t use beyond 24GB, both cards are virtually the same. Suffice to say, it looks like games aren’t getting close to saturating the 384-bit memory bus just yet.
Pokdepinion: Quite surprised, to be honest.
