Nvidia is secretly developing the RTX 4070, but no one else seems to be doing so. We have obtained and mulled over a full spec list and proof of what is claimed to be a full spec list and third-party models from MSI and Palit (opens in new tab) ahead of the graphics card's anticipated release.

Videocardz (opens in new tab) obtained a slide claiming to be from Nvidia. We do not vouch for it, but it seems certain. [This slide shows that Nvidia will be using an AD104 GPU for the RTX 4070. The rest of the specs are no surprise. The card is reported to have 12GB of GDDR6X memory that can provide 504GB/s of bandwidth across the card, which works in conjunction with a large 36MB L2 cache to boost performance. This is 9 times the L2 cache of comparable RTX 30 series cards.

As for performance, here is a graph comparing the RTX 4070 to the RTX 3070 Ti (open in new tab) and RTX 3070 (open in new tab). Looks good, but what we are looking at is the RTX 4070's lead with DLSS Frame Generation enabled. This is what the big chunk cited as "300 OFA" is really about: the Ada optical flow accelerator, along with the 4th generation tensor cores, can perform frame generation to significantly improve performance, but the Ampere's cannot.

For a more apples-to-apples comparison, you can see the RTX 4070's shader TFLOPs marked as "29 shaders" on this graph, an improvement over the RTX 3070 Ti's 22 TFLOPs, so there should be a reasonable improvement in non-DLSS game performance.

This 29 TFLOPs figure actually puts the RTX 4070 at about the same 29.7 TFLOPs of the RTX 3080 10GB and just slightly below the 30.6 TFLOPs of the RTX 3080 12GB.

Raw computing performance as measured in TFLOPs is not an entirely accurate measure of gaming performance. Memory is also extremely important.

Nvidia appears to be deploying the same memory configuration on the RTX 4070 as on the RTX 4070 Ti. Compared to the previous generation RTX 3080, the RTX 4070 outperforms the RTX 3070 in overall bandwidth, although the memory system looks a bit weaker due to a thinner memory bus and lower overall bandwidth. Also, while the RTX 3070 used GDDR6 memory, the RTX 4070 sticks to GDDR6X memory, which is preferred by the high-end RTX 40 series cards.

And even though the overall core architecture remains largely conventional, the additional pool of L2 cache on the new Ada card delivers very high performance.

If the RTX 4070's price tag is the $599 (open in new tab) that has been reported in previous rumors, this specification will likely make it one of the more affordable RTX 40 series cards. That said, it is still a significant amount of money for a fourth-tier graphics card.