A few weeks ago, @Kopite7kimi took to Twitter to share the expected performance benchmarks for the NVIDIA GeForce RTX 4080. On Friday, he listed the expected specs and 3DMark Time Spy performance data for the RTX 4080 / 4070 cards again. First, he expects the RTX 4080 to use a lean-spec AD103-300 GPU, enabling 80 of the 84 SM groups / 10240 CUDA cores.
Complemented by 224 ROP grid units + 64MB L2 cache, with 256-bit @ 16GB of 21Gbps GDDR6X memory (672 GB/s bandwidth).
The RTX 4080 is expected to have a nominal TBP announcement of around 420W, close to the RTX 4090. However, since the memory frequency is not as aggressive, Nvidia is likely to adjust the GPU main frequency upwards to obtain higher graphics performance.
As for the RTX 4070, @Kopite7kimi speculates that the card will be based on a streamlined AD104-275 GPU with 56 SM units / 7168 CUDA cores enabled (slightly higher than the RTX 3070 Ti’s 6144 CUDA cores).
Complemented by 160 ROP grid units / 48MB L2 cache, with 160-bit @ 10GB of 18Gbps GDDR6 memory, the nominal TBP power consumption may be around 300W, and the expected pricing will not be lower than RTX 3070 / 3070 Ti.
In terms of predicted performance, @Kopite7kimi speculates that the RTX 4080’s “3DMark Time Spy Extreme” score is expected to exceed 15,000, respectively, than the RTX 3080 / RTX 3090 Ti flagship lead 68% / 33% or so.
However, RTX 4090 still may be higher than RTX 4080 by nearly 27%, such a huge gap, far more than the previous generation RTX 3090 / 3080 between 13%.
In addition, RTX 4070 is said to have a TSE score of around 10,000 points, an improvement of nearly 50% over RTX 3070 and settings close to the non-Ti version of RTX 3090.
If the RTX 4070 is eventually priced at $499, it will obviously be a huge draw for gamers, but there may be a huge performance gap of up to 50% between it and the RTX 4080.
Even if Nvidia subsequently releases the RTX 4070 Ti with a 10-15% performance increase, there will still be a 35% performance gap between these two segments.
For reference, the performance gap between the previous generation RTX 3080 / 3070 is 36%, RTX 3070 Ti and the former’s gap is reduced to 23%!
In the end, it looks like only the Ada Lovelace AD102 GPU will be able to meet the 2x performance target with even more insane power, while the rest of the product line will see performance gains of around 50-70%.
However, in previous tweets, the tipster mentioned these “conservative estimates”, so the final GPU performance parameters may be more “reasonable”.