While there’s still plenty left for Nvidia to reveal about GeForce RTX 5070 and 5070 Ti, rumourmongers have offered a potential sneak peek at their specifications. Should these leaks prove true, these next generation graphics cards will demand more power, boast slightly higher CUDA core counts, and ride in on a similar memory configuration to their predecessors albeit with a more modern, faster type.
These specs come courtesy of (who else?) kopite7kimi. This marks the second time the leaker has shared details on the upcoming 70 class cards, building on prior insights into GeForce RTX 5070 Ti. This time around, the X user is offering insight into memory specifications.
| RTX 5070 Ti | RTX 5070 | |
|---|---|---|
| Board/SKU | PG147-SKU60 | PG146/147-SKU70 |
| GPU | GB203-300-A1 | GB205-300-A1 |
| CUDA cores | 8,960 | 6,144 |
| Memory size | 16GB | 12GB |
| Memory type | GDDR7 | GDDR7 |
| Memory bus | 256-bit | 192-bit |
| Power | 300W | 250W |
GeForce RTX 5070 Ti is apparently much closer in specification to GeForce RTX 4070 Ti Super than base GeForce RTX 4070 Ti, featuring a 16GB VRAM buffer and 256-bit bus. However, its bandwidth should be much greater thanks to its GDDR7 memory modules. Continuing comparisons to its closest Lovelace relative, this Blackwell successor’s CUDA core count will be 6% greater but it’s impossible to extrapolate how this will affect performance given architectural differences. I should note that all of these bumps come with a 15W bump to power consumption too, which is still thankfully lower than earlier leaks suggested.
Contrastingly, GeForce RTX 5070 appears more closely related to GeForce RTX 4070 rather than GeForce RTX 4070 Super. Given that memory bus and size remains the same at 192-bit and 12GB, respectively, the biggest clue is CUDA core count. The upcoming graphics card features 6,144, relative to its predecessor’s 5,888, a generational increase of 4%. Finally, power consumption jumps up by 50W.
I have mixed feeling about these specifications. While I’m glad to see Nvidia build on its Super foundations for GeForce RTX 5070 Ti, it disappoints me to see that the company seemingly won’t give GeForce RTX 5070 the same treatment. Of course, I can’t make any definitive assessments until I have the card in-hand, as there’s every chance that the next iteration of DLSS could plug the gap for example. That’s not forgetting potential ‘neural rendering‘ capabilities too.
We’ll learn the truth of the matter soon enough when Nvidia takes the stage at CES 2025, on January 7 at 2:30AM (GMT). Club386 will be in Las Vegas for the show, so stay tuned for any and all announcements. In the meantime, give the site a follow on Google News to make sure you don’t miss any of our daily news coverage, as well as reviews and other thoughts on all things tech.
