We’re definitely looking at a match for the RTX 3080 with Ada’s frame gen pushing it beyond if these stats are to be believed.
The RTX 4070 is being kept a secret by Nvidia, but it appears that no one else is. Before the graphics card’s anticipated debut, we have what is said to be a complete specification list and evidence of third-party models from MSI and Palit.
Slide purportedly from Nvidia has been obtained by Videocardz (opens in new tab). Although we cannot verify it, it appears to be true.
The slide indicates that Nvidia will use the same AD104 GPU used in the RTX 4070 Ti for the RTX 4070. It matches what we anticipated, and the rest of the specification comes as no surprise. A massive 36MB L2 cache and 12GB of GDDR6X memory are said to work together to boost performance. This memory is purportedly capable of delivering 504GBs of bandwidth across the card. That is nine times the L2 cache of comparable RTX 30-series GPUs.
Comparison RTX 4070 & RTX 3070 Ti
A graph comparing the RTX 4070’s performance to that of the RTX 3070 Ti and RTX 3070 is available. Although it appears attractive, what we are really seeing is the lead of an RTX 4070 with DLSS Frame Generation turned on. That large portion referred to as 300 OFA is exactly that. In contrast to those in Ampere, Ada’s Optical Flow Accelerators may run frame generation to significantly boost performance along with those 4th Gen Tensor Cores.
The RTX 4070’s shader TFLOPs are marked in that graph as 29 shader for an apples-to-apples comparison. This increase over the RTX 3070 Ti’s 22 TFLOPs means that non-DLSS games, or those that use raster graphics, should perform reasonably well.
This 29 TFLOPs figure actually places the RTX 4070 somewhat behind the 30.6 TFLOPs of the RTX 3080 12GB and roughly on par with the 29.7 TFLOPs of the RTX 3080 10GB.
Game performance cannot be fully gauged by raw computing performance expressed in TFLOPs. Moreover, memory is crucial.
The RTX 4070 and RTX 4070 Ti appear to share the same memory setup from Nvidia. With a narrower memory bus and less total bandwidth than the RTX 3080 from last generation. The memory system does appear to be lacking. The RTX 4070 would easily surpass the RTX 3070 in terms of overall bandwidth. The RTX 4070 sticks with the GDDR6X memory used by the higher-end RTX 40-series cards. Hence replacing the GDDR6 memory used by the RTX 3070.
The new Ada cards also include an additional pool of L2 cache. Which is what has allowed them to perform so much better. Even though if the fundamental architecture as a whole has essentially remained the same.
The RTX 4070 appears to be one of the more cost-effective RTX 40-series cards if its $599 price tag, as previously claimed via rumours, holds true. Nonetheless, that is a significant amount to pay for a graphics card that is in the fourth tier.
Suggested Reads: Pair up a Supreme Monitor with the new RTX