Early RTX 5090D V2 benchmarks suggest that maybe 384-bit memory is just fine after all — 'nerfed' China-market GPU runs within 2.2% of 5090D in some tests

Early RTX 5090 V2 benchmarks have been published overseas, revealing a surprising trait of Nvidia's latest Chinese flagship GPU. Chinese review site Yesky benchmarked the RTX 5090D and RTX 5090D V2 head-to-head and discovered the latter performs on par with the former in gaming performance, despite the V2 variant featuring 33% less memory bandwidth.

Yesky benchmarked the RTX 5090D and RTX 5090D V2 in several titles, featuring the full gauntlet of 3DMark benchmarks: Steel Nomad, Speed Way, Port Royal, Fire Strike Ultra, Fire Strike (the vanilla version), TimeSpy Extreme, TimeSpy (the vanilla version), and several games Cyberpunk 2077, Hogwarts Legacy, some mysterious PC game Google Translate is dubbing "Marvel Brawl", Borderlands 3 and Far Cry New Dawn.

In all of the benchmark runs Yesky performed, the RTX 5090D V2 maintained virtually identical performance to the RTX 5090D, with any gaps being pretty marginal. In 3DMark Speed Way, the RTX 5090D was 3.5% faster than the RTX 5090D V2, and in Fire Strike, the RTX 5090D was 2.2% quicker. Speed Way and Fire Strike represented some of the best results for the RTX 5090D, with benchmark results from the rest of the 3DMark suite producing even lower margins.

In Cyberpunk 2077, the RTX 5090D reportedly produced 424 FPS with DLSS 4, and in the same test, the RTX 5090D V2 produced a nearly identical 418 FPS. This behavior was present in the rest of Yesky's testing suite. Hogwarts Legacy achieved 300 FPS on the RTX 5090D and 297 FPS on the 5090D V2 with DLSS 4. "Marvel Brawl" produced 577 FPS on the RTX 5090D and 569 FPS on the RTX 5090D V2.

Even without DLSS 4 or any DLSS upscaling, the two GPUs are closely tied. Borderlands 3 at 4k resolution — a resolution that is normally sensitive to memory bandwidth, the RTX 5090D produced 164 FPS, and the RTX 5090D V2 produced 161 FPS. The same behavior was also present in Far Cry: New Dawn.

Yesky's benchmarking results demonstrate that the RTX 5090D V2 suffers virtually no performance loss from its neutered 384-bit 24GB memory configuration compared to the RTX 5090D's 32GB 512-bit configuration. More interestingly, Yesky's benchmarks appear to suggest that Nvidia's Blackwell architecture cannot take advantage of the RTX 5090 and RTX 5090D's 512-bit memory interface for gaming, even at 4k resolution.

This could be the reason Nvidia has not dabbled with a 512-bit memory interface until now with the RTX 5090 series. Past Nvidia flagships, including the RTX 4090, RTX 3090, and RTX 2080 Ti, all featured 384-bit memory interfaces. Even the highly regarded GTX 1080 Ti only came with a 352-bit interface. By contrast, AMD was dabbling with 2048-bit interfaces at the time of Pascal's debut in cards such as the RX Vega 64.

We already know that architectural advancements (like Ada Lovelace's L2 cache) and speedier memory ICs can make up for skinnier bus widths, but it is very interesting to know that Nvidia's architectures might not be capable of taking advantage of 512-bit interfaces for gaming.

The RTX 5090D V2 is Nvidia's third RTX 5090 variant and the second RTX 5090D model tailor-made to comply with U.S. export trade restrictions aimed at China. The RTX 5090D V2 is a neutered version of the RTX 5090D with "just" 24GB of GDDR7 memory and a 384-bit interface, matching the memory configuration of past Nvidia flagships like the RTX 4090 and RTX 3090. Potential buyers will be buoyed by these early performance scores, given that the V2 costs the same as the outgoing 5090D, despite the lowered specs.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.