According to the well-known NVIDIA hardware leaker kopite7Kimi on X, details about the final versions of NVIDIA's initial upcoming wave of GeForce RTX 50 series "Blackwell" graphics cards are now available. The leaked GPUs include the GeForce RTX 5090 and RTX 5080, with a noticeable increase in the gap between xx80 and xx90 SKUs. Starting with the high-end GeForce RTX 5090, NVIDIA has opted for the GB202-300-A1 die and has activated 21,760 FP32 CUDA cores on this top-tier model. Alongside the extensive 170 SM GPU configuration, the RTX 5090 boasts 32 GB of GDDR7 memory on a 512-bit bus, with each GDDR7 die operating at 28 Gbps, resulting in a memory bandwidth of 1,568 GB/s. All of this is contained within a 600 W TGP.
As for the GeForce RTX 5080, NVIDIA has chosen to further differentiate between its xx80 and xx90 SKUs. The RTX 5080 features 10,752 FP32 CUDA cores paired with 16 GB of GDDR7 memory on a 256-bit bus. With GDDR7 running at 28 Gbps, the memory bandwidth is halved at 784 GB/s. This SKU utilizes a GB203-400-A1 die, designed to operate within a 400 W TGP power limit. Comparing to the RTX 4080, the RTX 4090 has 68% more CUDA cores. The speculated RTX 5090 is rumored to have approximately 102% more CUDA cores than the speculated RTX 5080, indicating a further separation of NVIDIA's top-tier SKUs. It will be interesting to see the pricing strategy NVIDIA adopts for its upcoming GPUs, allowing for a comparison of generational improvements and the distinctions between xx80 and xx90 models with their expanded gaps.
