Recently, we reported that NVIDIA was creating another China-exclusive GPU in an effort to comply with latest US export laws, and the GPU in question was referred as ‘RTX 5090 DD’ at the time. Per MEGAsizeGPU, the name has since changed, now dubbed ‘RTX 5090 D v2’.
RTX 5090 D v2
Despite a seemingly intuitive naming of appending ‘v2’ in its name, the very nature of this card is rather counter-intuitive. This GPU, while sharing the same core count (21,760 CUDA cores) and TGP (575 watts), has 8GB less VRAM, cutting it down to just 24GB – and this does come with performance penalties due to the reduced memory bandwidth associated with narrower memory bus, which is now at 384-bit.
Thus, the ‘RTX 5090 D v2’ card will perform worse than the original RTX 5090 D, although the silver lining here is that NVIDIA will make it pretty clear right on its branding. The variant with reduced VRAM capacity will now see ‘v2’ suffix next to the D letter (as shown by the leaker), which should be prominently visible on the retail packaging. According to HKEPC, this card will launch in August 2025, with similar pricing expected.
As far as gaming performance is concerned, the RTX 5090 D v2 is likely to see a minor regression in performance in bandwidth-hungry games, though the capacity itself shouldn’t be a concern. With equal amounts of CUDA cores and power envelope to work with, gaming performance should generally be mostly intact; for AI-specific workloads though, this new update may render some bigger AI models impossible, if not at least impractical, to run on this card.
Pokdepinion: Still a pretty odd name if you ask me.
