Nvidia’s upcoming RTX 5090D graphics card may boast the same specs as its regular non-D counterpart. This China-exclusive GPU is said to use a firmware limitation to fit within the USA’s performance restrictions.
According to leaker Panzerlied on the Chiphell forum, Team Green plans to use the same hardware on both the regular GeForce RTX 5090 and the China-exclusive RTX 5090D. Seeing how the older RTX 4090 surpasses the US’s trade restrictions – hence the launch of the RTX 4090D – we wonder how Nvidia is planning to limit the RTX 5090D’s performance. As a reminder, the RTX 4090D’s GPU was cut down by around 11% compared to the original in ordr to meet the restricted performance target.
One solution could be a firmware/BIOS lock, stopping the D variants from utilising the full might of Nvidia’s new flagship chip. Another could be degrading performance – especially AI – through the drivers. Lastly, an aggressive power limit to starve the GPU or an artificial boost limit could stop it from reaching higher frequencies. That said, the brand may have found some way to circumvent said restriction via other means. Something similar to the LHR (Lite Hash Rate) GeForce cards back during the crypto mining boom comes to mind.
A firmware/BIOS solution has two advantages, assuming the US authorities don’t counter it. The first simplifies Nvidia’s life by allowing the use of identical chips, reducing manufacturing costs. The second gives Chinese customers a potential way to unlock the card’s full power with some kind of modification, such as a simple BIOS flash if Nvidia allows it, or through some other software trickery. As long as the chip isn’t physically cut, someone will probably find a way. Reminds me of the old days when you could unlock an extra core on chips like AMD’s Phenom II X3.
For now, RTX 5090, and thus RTX 5090D, is expected to house 21,760 CUDA cores alongside 32GB of GDDR7 memory running on a 512-bit bus. The card is said to target a 600W TGP, using one or two 12-2×6 power cables.
