NVIDIA's RTX 50 "Blackwell" GPUs: 2-2.6x Faster Than RTX 40, 3nm Monolithic, and Overhauled Streaming Multiprocessors

Rumors suggest that NVIDIA's next-gen GeForce RTX 50 series graphics cards will utilize the new Blackwell GPU architecture, which will retain a monolithic die approach and may be manufactured on TSMC's 3nm process node. The Blackwell GPUs are expected to feature massive overhaul to CUDA architecture, support for GDDR7 memory, and a PCIe Gen 5 interface with clock speeds of 3 GHz+. The flagship GB102 GPU is said to rock 144 SMs and 96 MB of L2 cache, and the performance is rumored to be another 2x leap over Ada GPUs.