GeForce RTX 3080
The GeForce RTX 3080 is the first product to be launched on the new architecture, and for some reason NVIDIA is referring to it as "the new flagship." The RTX 3080 "starts at" $699, and succeeds the RTX 2080 and RTX 2080 Super from the previous generation. Based on the GA102 silicon, the RTX 3080 is endowed with 8,704 CUDA cores across 68 SM (34 TPCs), 68 RT cores, 272 tensor cores, 272 TMUs, and 96 ROPs. The chip features a 320-bit wide GDDR6X memory interface that drives 10 GB of memory at 19 Gbps, which works out to 760 GB/s memory bandwidth. NVIDIA rates the board power of the RTX 3080 at 320 W. The target application is 4K UHD gaming.
NVIDIA claims massive near-double performance gains for the RTX 3080 over the previous generation RTX 2080 at 4K UHD resolution.
GeForce RTX 3090
The GeForce RTX 3090 will be the second product based on the new architecture and is being positioned in the "halo" segment usually occupied by the TITAN brand, which could also mean that there likely won't be a TITAN Ampere. It "starts" at $1,499, and in NVIDIA's marketing slides and presentations, it has extensively been compared to the TITAN RTX. The RTX 3090 uses more of the GA102 silicon, with 10,496 CUDA cores across 82 SM (41 TPCs), 82 RT cores, 328 tensor cores, 328 TMUs, and likely 112 ROPs. The chip maxes out the 384-bit wide GDDR6X memory interface, pulling 24 GB of memory at 19.5 Gbps, which works out to 940 GB/s of memory bandwidth. NVIDIA rates the board power at 350 W. Not only is this card designed to pull 4K at high refresh rates, but it is also 8K capable.
NVIDIA claims performance gains of 50%–70% for the RTX 3090 over the $2,500 TITAN RTX.
GeForce RTX 3070
The GeForce RTX 3070 is being planned for an October 2020 launch and should appeal to a wider audience base at its $499 starting price. It's still being targeted at 1440p high refresh rate gaming, although NVIDIA claims this card to be "faster than the RTX 2080 Ti", which is very much a 4K gaming card. At this point, we have no official information on what silicon the RTX 3070 is based, although we think it's GA104. NVIDIA revealed its CUDA core count to be 5,888—higher than with the TITAN RTX. These CUDA cores are spread across 46 SM (23 TPC) and feature 46 RT cores, 184 tensor cores, 184 TMUs, and likely 64 ROPs. It features a conventional 256-bit wide GDDR6 memory interface pulling 8 GB of memory, and whether NVIDIA runs it at 14 Gbps or 16 Gbps remains to be seen. Typical board power is rated at 220 W.
The RTX 3070 promises a nearly 60% performance "leap" over the RTX 2070, helping it not only rule the 1440p segment, but also be 4K UHD gaming capable, as the RTX 2070 + 60% is approaching RTX 2080 Ti performance levels.