Nvidia's best graphics card for gaming, the RTX 4090 was recently struck with an export ban preventing the GPU from being sold to China. In response, Nvidia is reportedly developing a China-exclusive variant dubbed the RTX 4090 D (for "Dragon") — that's according to WCCFTech. The RTX 4090 D will supposedly feature lower performance and specifications to comply with U.S. export regulations. Sprinkle liberally with salt, obviously, though the idea absolutely makes sense.
The rumored GPU's specifications are unknown, but due to the US export guidelines, the RTX 4090 D will need to be less powerful than its outgoing counterpart. As a result, we will likely see the 4090 D coming with a lower TDP and a lower SM/core count to separate itself from the RTX 4090. Given how big the performance gap is between the RTX 4080 and the RTX 4090, this neutered 4090 could be quite a bit slower than the outgoing part (if the RTX 4080 Super doesn't come out). In a worst-case scenario, it might not even come with the AD102 GPU die.
The main metric that the 4090 D will need to meet is TPP, Total Processing Power. This is calculated by the maximum compute for a given bit-depth, using TFLOPS (or TOPS for integer work) multiplied by the number of bits. For the RTX 4090, TPP is 660.8 * 8 = 5,286 for FP8 work running on the Tensor cores (sparsity doesn't count). Also note that the value is the same for FP16: 330.4 * 16 = 5,286. The allowed limit is 4,800, so the RTX 4090 is about 10% "too powerful."
This is why the RTX 4080 can still be sold (and manufactured / assembled) in China. It has a TPP of 389.6 * 8 = 3,117. There's a secondary metric for data center parts called Performance Density (TPP divided by die area), but that doesn't come into play for consumer GPUs. It's the reason the Nvidia L4 also falls under the restrictions, as it has a PD rating of 6.6, using the AD104's 294 mm^2 die size and a TPP value of 1,936. (It's a good thing PD doesn't apply to consumer products, as the entirety of Nvidia's desktop RTX 40-series has a PD above 6.0.)
So what would Nvidia need to do for an RTX 4090 D to be allowable under U.S. export controls? Technically, anything below 4,800 TPP would suffice, but Nvidia will likely build in some wiggle room — to ensure overclocking as an example doesn't become a problem. Assume a clock speed of 2.7 GHz and we get a maximum number of SMs of 108. That would still have a TPP score of 4,778, however, so Nvidia will likely be aiming a bit lower, like around 100 SMs (4,428 TPP), just to be safe.
Making a neutered RTX 4090 is a smart move from Nvidia, as the Chinese market is huge, with a total population over 4x greater than the United States. As a result, it's worth Nvidia's time and effort to create a high-end replacement for the Chinese market that is similarly priced, so it can reclaim the ultra-high-end GPU market — 7900 XTX cards are reportedly doing well in China right now.
According to WCCFTech, the RTX 4090 D will be available from add-in board partners in China, similar to the RTX 4090, with no indication of a Founders Edition model. Pricing is also reported to be "similar" to the RTX 4090's old price in China — i.e. before the restrictions caused prices to shoot up — though it may be a tad less expensive to reflect its lower performance.
The demise of the RTX 4090 and the rise of the RTX 4090 D is the result of the U.S. Commerce Departments recent attempts to deny China access to advanced technologies. The RTX 4090 is an unfortunate casualty of the recent bans, though given what we're now seeing with RTX 4090 cards being retrofitted with blower-style coolers, its understandable that even consumer parts may need restrictions.
The RTX 4090 D is one of several GPUs Nvidia is creating to comply with the U.S. regulations. Nvidia is also preparing several additional AI and HPC-focused GPUs that will comply with the United States' performance restrictions. The RTX 4090 D will reportedly launch in early 2024.
