NVIDIA RTX 3070 3DMark Performance Leaked – Actually Faster Than An RTX 2080 Ti

So NVIDIA's upcoming RTX 3070 has gotten pretty much all performance metrics leaked courtesy of Whycry over at Videocardz. As expected the card is actually faster than the RTX 2080 Ti. The non-linear downscaling of NVIDIA's RTX 3000 lineup is something we explained in our NVIDIA fine wine article a while back and these results prove our point. The RTX 3070 is just slightly faster than the RTX 2080 Ti flagship and as drivers and APIs improve over the next year or so, the performance of the RTX 3000 series will greatly increase.

The RTX 3070 scores 8,749 points in Fire Strike Ultra graphics benchmark and 17,115 points in Fire Strike Extreme. Time Spy Extreme clocks in at around 6,907 points. In the newly launched ray tracing DX12 Port Royal benchmark, the RTX 3070 scores 8324 points.

The RTX 3070 will have a GA104 GPU equipped with 8GB GDDR6 memory. It is also currently based on Samsung's 8nm (N8) process and is the second-fastest (uncut) chip in NVIDIA RTX 3000 lineup. The GPU spans 392.5mm² and features 17.4 Billion transistors which is almost the same number as the TU102 GPU (93% to be precise). Considering the die size is almost half as small, the density increase is pretty impressive.

For the GeForce RTX 3070, NVIDIA has enabled a total of 46 SM units which result in a total of 5888 CUDA cores. In addition to the CUDA cores, NVIDIA's GeForce RTX 3070 also comes packed with next-generation RT (Ray-Tracing) cores, Tensor cores, and brand new SM or streaming multi-processor units. However, due to the bottlenecks in software which we explored in our fine wine article, performance appears to scale nonlinearly over the 2080 Ti mark.

ProShop is a large e-tailer that regularly deals with the DIY market and unlike most other retailers they have kindly taken to publishing the current status of their supply line with various manufacturers. This allows us to glimpse the state of the supply chain for NVIDIA GPUs - at least for small-to-mid-sized retailers in Germany/Denmark - and estimate future conditions.

According to the data provided by them, they have ordered a total of 15117 NVIDIA RTX 3000 series GPUs from various AIBs and have only received (or are going to receive at the time of writing) 1112 GPUs. This means that the company is only getting 7.4% of the demand from various NVIDIA AIBs. The complete set of data from ProShop (at the time of writing) is given below along with an explanation of the variables.

Explanation:

For the RTX 3090, ProShop ordered a total of 1912 GPUs from various AIBs, out of which they have only received (or are going to receive at the time of writing) a total of 200 GPUs. This equals a supply vs demand fulfillment of 10.5% on an aggregate basis. On the retail side, ProShop has customer preorders amounting to 290 right now (which means they are already unable to fulfill 31% of preorders).

For the RTX 3080, ProShop ordered a total of 8925 GPUs from various AIBs, out of which they have only received (or are going to receive at the time of writing) a total of 621 GPUs. This equals a supply vs demand fulfillment of 7.0% on an aggregate basis. On the retail side, ProShop has customer preorders amounting to 3733 right now (which means they are unable to fulfill 84% of preorders).

For the newly launched RTX 3070, ProShop ordered a total of 4280 GPUs from various AIBs, out of which they have only received (or are going to receive at the time of writing) a total of 291 GPUs. This equals a supply vs demand fulfillment of 6.8%.

So it seems NVIDIA is currently only able to meet under 10% of demand based on this data but there are two obvious caveats associated with this. 1) The numbers could theoretically look very different for huge retailers like Amazon, BestBuy, Walmart, etc. and 2) We do not know what the previous demand baseline is - although I would hazard a guess that it was at least, not this low.

There have also been rumors that the company may be planning a shift back to TSMC 7nm - which if true would indicate unforeseen yield issues with Samsung's 8nm - although I personally [caution: opinion] think that is an unlikely eventuality [/opinion]. As per our reports, supply should improve significantly in early 2021 as the volume blocks ordered by NVIDIA kick in and should be able to meet a significant chunk of all this unmet demand - provided of course, the miners don't have anything to say about it.