The rumors a few slimmed-down model of the RTX 5090D for the Chinese language market are intensifying. Now the board associate MANLI has formally confirmed this in a direct person contact: The revised GeForce RTX 5090D will solely be outfitted with 24 GB of VRAM – 8 GB lower than the unique mannequin. The background to that is the US export coverage, which now prohibits extra highly effective GPUs for sure markets.
Much less reminiscence, an identical GPU base?
In keeping with the newest info, the core GPU configuration stays the identical, even when additional restrictions are to be anticipated. The brand new RTX 5090D is now mentioned to be primarily based on the specs of the RTX Professional 5000 and makes use of 24 GB GDDR7 reminiscence, linked by way of a 384-bit interface. The ensuing reminiscence bandwidth is due to this fact considerably beneath the extent of the unique 5090D with 32 GB and over 1.4 TB/s bandwidth – a transparent measure to adjust to US pointers.
CUDA core downgrade in prospect
Along with the VRAM capping, a discount within the variety of computing models can also be being mentioned. Insider sources converse of solely 14,080 CUDA cores as an alternative of the 21,760 beforehand rumored, as used within the common RTX 5090. If that is confirmed, the efficiency degree of the brand new 5090D could be considerably lowered not solely within the skilled phase, but in addition within the gaming phase. Official confirmations from NVIDIA are nonetheless pending, however MANLI – as a board associate with good connections – has no less than reported the VRAM discount as confirmed. The opposite specs might nonetheless change relying on the political state of affairs and market methods, notably when it comes to clock charges and vitality consumption.
Market launch deliberate from July
In keeping with the present state of affairs, the “new” RTX 5090D is about to go on sale on the finish of July or starting of August. It should exchange the unique mannequin, which was discontinued within the second quarter of 2025. It stays to be seen whether or not the board companions in China will truly use a bodily totally different board or just adapt the BIOS. The cooling can also be prone to stay largely an identical to the total RTX 5090 – however the energy consumption is doubtlessly decrease.
Technical and political compromises
The background to the adjustment is obvious: the US authorities prohibits the export of sure high-performance elements to nations reminiscent of China. NVIDIA is responding to those laws with so-called D variants (“degraded”), which frequently solely differ visually and functionally in particulars – however these particulars are ample to fulfill the export necessities. The discount to 24 GB reminiscence and simultaneous shrinking of the computing models means a clearly outlined efficiency restrict – which is clearly politically fascinating. On the identical time, NVIDIA stays current on the Chinese language market with out fully abandoning the high-performance phase.
Successor: RTX 5080 Ti/Tremendous on the horizon
Whereas the 5090D is being tailored, there may be already hypothesis about upcoming fashions within the RTX 50 collection with Blackwell structure. In keeping with unofficial sources, an RTX 5080 Ti or RTX 5080 Tremendous must also supply 24 GB of VRAM and be launched on the finish of the yr. Whether or not it is going to be positioned as a globally obtainable different to the RTX 5090 or is merely an intermediate step in direction of the RTX 5090D within the new format stays to be seen. General, it’s clear how strongly geopolitical components at the moment are influencing product planning and mannequin variants within the GPU market – with very noticeable results for skilled customers and fanatics alike.
Supply Weibo
