The family lineup of RTX 40 series graphics cards is becoming more and more neat, and it is time to look forward to the RTX 50 series.
In fact, as early as December last year, there were rumors that NVIDIA was verifying the RTX 50 series prototype card, and the GPU chip code name was Blackwell.
A few days ago, the breaking news god RedGamingTech took the lead in sharing some details of the RTX 50 series graphics card or Blackwell GPU.
He pointed out that the new generation of game graphics cards (such as GB102) is still designed with a single GPU chip, while single server/data center products will use MCM multi-chip interconnection.
In terms of specifications,RTX 5090 based on GB102 contains 144 groups of SM units, that is, 18432 CUDA (assuming each group of SM is still 128 CUDA), 12.5% more than RTX 4090, 96MB L2 cache, matching GDDR7 memory (384bit bit width), support PCIe 5.0 x16.
Thanks to TSMC’s 3nm process specially customized for NVIDIA, the overall energy efficiency of the RTX 50 series has been further improved, with a core frequency exceeding 3GHz, and its performance is expected to reach 2~2.6 times that of the RTX 40 series.
