Everyone's favourite Nvidia specs leaker, Kopite7kimi, has just done the deed on Nvidia RTX 5070 Ti with a fairly comprehensive spec dump (via Videocardz).
To cut to the chase, we're talking 8,960 CUDA cores and a board power of 300 W. For context, the current RTX 4070 Ti sports 7,680 cores and 285 W rating. On paper, that makes the new GPU a fairly marginal upgrade.
The CUDA core increase represents a mere 17% bump, which is an awfully modest upgrade given a two-year wait. However, the critical detail missing here is clockspeed. The 4070 Ti has a 2,610 MHz Boost speed. The new 5070 Ti could up that significantly.
The 5070 Ti is also said to be based on the same GB203 silicon as the upcoming and similarly rumoured RTX 5080, though that chip is expected to have more of the die enabled, with 10,752 CUDA cores.
The arguable catch here is that the whole RTX 50 family is expected to be built on TSMC's N4 silicon node. That's a derivation of the N5 technology used for the RTX 40 family, not a properly new node like TSMC's various N3 nodes. That doesn't necessarily bode well for a huge frequency uplift, even if N4 ought to enable at least some improvement over N5 for clocks.
Of course, the other major imponderable is the Blackwell architecture inside the RTX 50 cards. For now, little is known about any fundamental changes that could, for instance, see a Blackwell CUDA core do more work per clock cycle than the Ada Lovelace spec cores in the 4070 Ti.
In other words, if you combine the 17% CUDA core bump with a decent clockspeed improvement and more work done per clock, you could end up with a fairly hefty overall gen-on-gen performance uplift. Or not. It's all to be proven.
Keep up to date with the most important stories and the best deals, as picked by the PC Gamer team.
All that said, what this will all likely come down to is pricing. If the 5070 Ti ends up dramatically faster than the 4070 Ti but also costs dramatically more, that won't be great. On the other hand, even
Read more on pcgamer.com