Nvidia is conserving the RTX 4070 underneath wraps nevertheless it seems to be like completely nobody else is. We have got what’s claimed to be a whole specification listing, and proof of third-party fashions from MSI and Palit (opens in new tab), to ponder over forward of the graphics card’s anticipated launch.
Videocardz (opens in new tab) has acquired maintain of a slide that’s claimed to be from Nvidia. We have no assure of that, nevertheless it seems to be prefer it holds water.
The slide exhibits Nvidia will probably be utilizing the AD104 GPU for the RTX 4070—the identical GPU used within the RTX 4070 Ti (opens in new tab). That checks out with what we’re anticipating, and the remainder of the spec is not any shock, both. There’s reportedly 12GB of GDDR6X reminiscence able to pumping out 504GB/s bandwidth throughout the cardboard, working in tandem with a big L2 cache dimension of 36MB to push efficiency. That is 9x the L2 cache of the equal RTX 30-series playing cards.
When it comes to efficiency, there is a graph evaluating the RTX 4070 to the RTX 3070 Ti (opens in new tab) and RTX 3070 (opens in new tab). It seems to be good, however what we’re taking a look at is an RTX 4070’s lead with DLSS Body Technology enabled. That is what that large chunk cited as ‘300 OFA’ is. Ada’s Optical Stream Accelerators, alongside these 4th Gen Tensor Cores, can run body technology to massively enhance efficiency, whereas those in Ampere can not.
For a extra apples-to-apples comparability, you’ll be able to see the shader TFLOPs famous as ’29 shader’ for the RTX 4070 in that graph. An enchancment over the 22 TFLOPs for the RTX 3070 Ti, which suggests we should always see an affordable step-up in non-DLSS sport efficiency; that is your conventional rasterized stuff.
That 29 TFLOPs quantity really places the RTX 4070 at across the identical because the RTX 3080 10GB at 29.7 TFLOPs, and only a contact underneath the RTX 3080 12GB’s 30.6.
Uncooked compute efficiency measured in TFLOPs is not a wholly correct measure of sport efficiency. Reminiscence can be essential.
Nvidia has seemingly rolled out the identical reminiscence configuration on the RTX 4070 as on the RTX 4070 Ti. That is a reminiscence system that does appear a bit weak subsequent to final gen’s RTX 3080, with a slimmer reminiscence bus and decrease total bandwidth, however the RTX 4070 could be roundly topping the RTX 3070 for bandwidth total. Additionally the place the RTX 3070 used GDDR6 reminiscence, the RTX 4070 is sticking to the GDDR6X reminiscence favoured by the higher-end RTX 40-series playing cards.
After which there’s that additional pool of L2 cache on the brand new Ada playing cards that has seen them ship a lot additional efficiency when the general core structure has remained largely acquainted.
If the RTX 4070’s price ticket is the $599 sum (opens in new tab) thus far reported by rumours, this specification seems to be to be one of many higher worth RTX 40-series playing cards for the cash. Nonetheless, a severe sum for a fourth-tier graphics card—will we ever see a value correction for graphics playing cards?
