A well-known hardware leaker has spilled a bit more info about Nvidia’s next-gen graphics cards known as Blackwell.
As VideoCardz spotted, Kopite7kimi has shared a leak on Blackwell’s GB202 GPU, what will in theory be the RTX 5090 and successor to AD102 (that powers Lovelace’s current-gen flagship, the RTX 4090).
As I mentioned before, GA100 is 8*8, and GH100 is 8*9. GB100 will have a basic structure like 8*10. GB202 looks like 12*8.September 28, 2023
The key info imparted is that the RTX 5090 may run with 24,576 CUDA cores – or at least that’s Nvidia’s current plan (and not the first time that figure has been floated).
That would be a chunky uplift from 18,432 CUDA cores with the AD102 chip, indeed it would represent an increase of a third, no less. However, it should be noted that the RTX 4090 tops out at 16,384 CUDA cores (and there’s no sign of a 4090 Ti that might push the full core allocation yet).
On top of this, Kopite7kimi reasserts a previous leak with a claim that the RTX 5090, or GB202, will employ a 512-bit memory bus.
As VideoCardz adds with its spec guesses, we could well be looking at GDDR7 video memory for Nvidia’s next-gen, and clocks that push up towards 3GHz, all of which should make for a surefire entry in the best graphics card list on the performance chops alone.
Analysis: Elevated top-end expectations
We are still a long way from next-gen Blackwell, so at this point, leaks very much need to be treated with even more seasoning than normal. Yes, go crazy with that saltshaker, because the RTX 5090 and siblings may not arrive until 2025 (most likely). Although some corners of the rumor mill have suggested earlier, the grapevine whispers still feel at a very early stage here.
However, hearing Nvidia is planning to ‘go big’ with the RTX 5090 is not exactly a surprise.
For one thing, we’ve already heard rumors of a 70% gen-on-gen performance increase for the next-gen flagship. Just dump the entire contents of the saltshaker on that one, you think? Although, in fairness, it’s true enough that given how much of a generational leap the RTX 4090 made, Nvidia has kind of set up expectations for huge uplifts at the top end of its GeForce line-up (with smaller upgrades down the line).
With great performance, though, will doubtless come a greatly eye-watering price. Nvidia may well want to squeeze profits for all they’re worth, given that there is so much money to be made on the AI side of the GPU market.
And then come the inevitable questions about how much frame rate juice a top-end desktop graphics card really needs. Isn’t the RTX 4090 already more than enough?
We wouldn’t argue with that assertion, but sometimes the very highest-end GPUs are not about what gamers need, so much as what they can brag about – and wallets get pretty deep at the enthusiast end of the scale, even for gaming rigs.
