One of the benchmarks which can maximize a GPU's potential is Octanebench. It is used for rendering purposes and can maximises a graphic cards throughput with almost no driver optimization required. In other words, you can think of it as the best case scenario for any given GPU. The NVIDIA RTX 4070 Ti was recently spotted (via LaptopVideo2Go via Videocardz) in the Octanebench database and beats out the older RTX 3090 Ti by a solid 5%.
NVIDIA RTX 4070 Ti is 5% faster than the RTX 3090 Ti in the first Octanebench aggregate score
Now since we are dealing with a benchmark that maximizes the theoretical throughput of a GPU, let us look at the theoretical numbers first. The RTX 3090 Ti has 10,752 CUDA cores clocked at 1860 MHz which yields a FP32 performance of 39.99 TFLOPs. The RTX 4070 Ti on the other hand has 7680 CUDA cores clocked at 2610 MHz which yields a FP32 performance of 40.09 TFLOPs.
Now Octanebench can also supposedly take into consideration the RT cores on the card, of which the NVIDIA RTX 4070 Ti has 60 compared to the 3090 Ti's 84. All said and done, the RTX 3090 Ti is very close, on paper, to the RTX 4070 Ti and it is interesting to see that there is at least one benchmark out there that can take full advantage of this.
The RTX 4070 Ti gets a score of 725.3 points while the RTX 3090 Ti gets a score of 688.17 points (or 5% faster). However, it is worth noting there that this score is a combination score with the results heavily aggregated and weighted. I noticed that the RTX 4070 Ti is significantly powerful in the info channels test (probably because of the upgraded VRAM) where it scores 953 compared to the RTX 3090 Ti's 697. On the Direct Lightning and Path Tracing side however, it scores roughly the same with the scores being 705 and 696 compared to the RTX 3090 Ti's 687 and 687.
This looks like a pretty good performance upgrade from NVIDIA and a solid replacement to the RTX 3090 Ti at face value. However, that is where the comparison breaks down. There are a lot of caveats associated with this. 1) This is a render test, not a gaming test. Games are heavily affected by driver optimizations and only have a partial correlation to peak theoretical FP32 performance. 2) We do not know if the card used here was a heavily OC-ed version or the vanilla stock version. It is always better to have a larger sample size before drawing any conclusions.
All this means that in real world scenarios, the RTX 3090 Ti should continue to remain faster than the RTX 4070 Ti on an aggregate gaming basis for the foreseeable future. The increased clocks speeds and architectural improvements however, could make this a desirable card to have for rendering enthusiasts and video professionals.