It’s surprising how a 6% reduction can help circumvent regulations that have significant geopolitical implications surrounding military use and AI research.

The first review of NVIDIA’s GeForce RTX 4090D graphics card, exclusive to China, has been published, and it’s not surprising that it’s unquestionably slower than the previous model, which is now restricted in various regions.

The RTX 4090D continues to utilize the AD102 GPU, but with a reduced core count of 14592, marking an 11% decrease compared to the standard RTX 4090 it replaces in China, all while maintaining the same MSRP.

The GPU retains the identical memory setup, featuring 24GB GDDR6X memory coupled with a 384-bit memory interface.

However, there have been alterations in power specifications, with the default power reduced from 450W to 425W. Additionally, there is a decrease in the number of Tensor cores, aiming to limit its AI capabilities in compliance with US regulations.

Expreview conducted a thorough review of the GALAX GeForce RTX 4090D Metal Master graphics card. The card boasts a robust 3.5-slot design and is equipped with a triple-fan solution, which is basically the starting point when it comes to RTX 4090 aftermarket models.

In their comprehensive testing, Expreview evaluated both gaming performance and synthetic benchmarks, with a primary focus on comparing the RTX 4090D against its RTX 4090 FE counterpart.

While there are slight differences in effective cooling capacity, the Metal Master closely aligns with NVIDIA’s reference specifications for the RTX 4090D, providing a near apples-to-apples comparison between it and the RTX 4090.

Despite the GeForce RTX 4090D featuring an approximately 11% reduction in total CUDA cores, Expreview’s performance testing indicates a performance decrease of around 5% to 7%, varying based on the specific workload.

The largest performance drops are attributed to synthetic workloads such as 3DMark’s arsenal of benchmark offerings, with the largest gap strangely enough attributing to Fire Strike Ultra.

While the standard RTX 4090 graphics card managed a score of 24,860 points on Fire Strike Ultra, the gimped RTX 4090D managed just 22,695 points, a performance loss of -8.7%.

While other synthetic workloads showcase declines in performance hovering around the 4% figure, on average in terms of synthetic workloads the RTX 4090D is 6.7% slower on average across 3DMark’s benchsuite, with another major performance loss coming from Time Spy Extreme and their Port Royal benchmarks, with the latter being more than understandable given how it incorporates ray tracing capabilities.

When it comes to rasterized gaming performance, the only real measurement of gaming workloads for the time being, Expreview have benchmarked the following games in order:

Cyberpunk 2077, Alan Wake 2, Forza Horizon 5, Ratchet and Clank: Rift Apart, F1 22, Warhammer 40K Darktide, Dying Light 2, Shadow of the Tomb Raider, Avatar: Frontiers of Pandora, Watch Dogs Legion, Assassin’s Creed (likely Mirage), Far Cry 6, Call of Duty (likely MW3), Final Fantasy 14, Rainbow Six Siege and Counter-Strike 2.

And again we see a general figure, despite effectively replacing the RTX 4090 for the exact same numerical retail pricing, the downgraded RTX 4090D showcases inferior performance figures across the board, with many games showcasing sharper performance losses of around 6.8% to as high as 7.6% inferior performance.

Across the board the performance drops are more severe and consistent with real world gaming workloads versus synthetical performance figures.

When Ray Tracing / Path Tracing is brought into the debate the RTX 4090D falls further behind the original RTX 4090, with differences in performance still varying wildly but generally increasing across the board.

Contrary to previous reports of a TGP (Total Graphics Power) lock, the GeForce RTX 4090D does not have such a restriction. It can reach up to 480W, exceeding NVIDIA’s specified limit by 55W. However, existing software options lack the capability to modify this limit.

Although Expreview managed to increase the GPU core frequency, achieving only a 200 MHz increase (with no mention of undervolting), resulting in a 2.3% improvement in synthetic test scores.

NVIDIA is set to launch the GeForce RTX 4090D later this month at the same retail price as the outgoing and now restricted GeForce RTX 4090. Gamers in China are left with little choice but to accept NVIDIA effectively offering them an inferior product for the same numerical value.

For them, this should be viewed as a price increase for lower specs. With over 5% inferior performance, the card doesn’t appear as bad as initially anticipated when rumors about the 4090D first surfaced. However, other factors need consideration, such as the selection of custom models and their overclocking potential, which is currently limited.