A possible Nvidia RTX 5090 prototype shows what might have been – an absolute monster with nearly 25K CUDA cores and an 800W TDP

a Possible Nvidia Rtx 5090 Prototype Shows What Might Have Been – an Absolute Monster with Nearly 25k Cuda Cores and an 800w Tdp

A possible Nvidia RTX 5090 prototype shows what might have been – an absolute monster with nearly 25K CUDA cores and an 800W TDP

Home » News » A possible Nvidia RTX 5090 prototype shows what might have been – an absolute monster with nearly 25K CUDA cores and an 800W TDP
Table of Contents

The Nvidia GeForce RTX 5090 is already shaping as much as be a beast of a GPU, given the specs unveiled at CES 2025, but when a brand new report is right, it might have been much more of a monster.

A well-regarded rumor miller, HXL, shared a submit on the Chinese language {hardware} discussion board ChipHell that claims to point out the PCB for an early prototype RTX 5090, together with some reasonably eye-watering specs properly past these for the manufacturing mannequin RTX 5090 due out subsequent week.

In accordance with the poster, the prototype was an engineering pattern produced in mid-July 2024 and was despatched to AIB companions to assist them put together their very own variations of the GPU. How the person received their hand on the prototype – assuming it is actual, which isn’t in any respect sure, so take every little thing with a heap of salt – they didn’t say, however they did present a number of the supposed specs on the pattern.

This contains the GPU SKU of GB202-200-A1, a CUDA core rely of 24,576 (or about 13% greater than the 21,760 within the manufacturing RTX 5090), a barely increased clock velocity of two,100MHz base and a couple of,514MHz increase, and barely sooner GDDR7 reminiscence modules clocked at 32Gbps (in comparison with the 28 Gbps chips within the manufacturing RTX 5090). These would have pushed the cardboard’s reminiscence bandwidth to 2TB/s reasonably than 1.79TB/s for the manufacturing 5090.

Given the CUDA core rely, we are able to additionally extrapolate that there would have been 192 SMs for the GPU, so 192 ray tracing cores and 768 Tensor cores for AI workloads.

Probably the most unimaginable spec, nevertheless, is the 800W TDP, which is nearly double the ability draw of the RTX 4090 and about 40% greater than the RTX 5090. As such, it might require two 12VHPWR connectors to produce sufficient energy for the cardboard.

Might it’s a Blackwell Titan RTX?

As our buddies over at Tom’s {Hardware} be aware, this card might additionally match the specs of a Titan RTX card constructed on Blackwell or an RTX 5090 Ti. We’ve not seen a Titan RTX for the reason that Turing period, although the argument could be made (and has) that the RTX 3090 and RTX 4090 graphics playing cards are the successors to the Titan RTX playing cards of previous, and it is undoubtedly potential that an RTX 5090 TI might sport these sorts of elevated specs.

Personally, if the GPU posted to ChipHell is legitimately an early engineering pattern of the RTX 5090 that has made its option to manufacturing, I believe it’s merely that: a pattern. It might be analogous to a first- or second-draft GPU earlier than refining the structure right down to the RTX 5090 that can go on sale subsequent week.

Whereas it is attention-grabbing to see some behind-the-scenes engineering in comparison with the precise manufacturing mannequin, in the end, it in all probability is not way more than that.

share this article.

Enjoying my articles?

Sign up to get new content delivered straight to your inbox.

Please enable JavaScript in your browser to complete this form.
Name