NVIDIA Blackwell Architecture product page

VERBATIM 2026-10-04, curl -L -A 'Mozilla/5.0 (Windows NT 10.0; Win64; x64)'. No per-GPU FLOPS table here, but three figures used by these records, verbatim: 'NVIDIA Blackwell-architecture GPUs pack 208 billion transistors and are manufactured using a custom-built TSMC 4NP process. All NVIDIA Blackwell products feature two reticle-limited dies connected by a 10 terabytes per second (TB/s) chip-to-chip interconnect in a unified single GPU.' 'The NVIDIA NVLink Switch Chip enables 130TB/s of GPU bandwidth in one 72-GPU NVLink domain (NVL72) and delivers 4X bandwidth efficiency with NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol (SHARP) FP8 support. The NVIDIA NVLink Switch Chip supports clusters beyond a single server at the same impressive 1.8TB/s interconnect. Multi-server clusters with NVLink scale GPU communications in balance with the increased computing, so NVL72 can support 9X the GPU throughput than a single eight-GPU system.' 'NVIDIA Blackwell Ultra Tensor Cores are supercharged with 2X the attention-layer acceleration and 1.5X more AI compute FLOPS compared to NVIDIA Blackwell GPUs.' Corroborates nvidia-gb200-nvl72's interconnect entries (nvlink-5 + nvswitch-5) and its 130 TB/s / 1.8 TB/s per-GPU figures on a first-party page that is not a datasheet. Also corroborates the 4nm process and 208B transistor count. Note the 1.5x Blackwell-Ultra-over-Blackwell AI compute claim is a general statement and does not by itself fix any FP4 number; the datasheet's explicit Sparse|Dense pair governs.