NVIDIA RTX PRO 6000 Blackwell Workstation Edition datasheet - verbatim spec table (vendor re-fetch 2026-10-04)

VERIFY-ONLY re-fetch of the primary datasheet PDF (doc code 3519208, 'Mar 25'). Every row below transcribed character-for-character from the Specifications table as rendered in the PDF text layer; no value was derived. SPEC TABLE VERBATIM: 'GPU architecture NVIDIA Blackwell' | 'NVIDIA CUDA Cores 24,064' | 'Tensor Cores 5th Generation' | 'Ray Tracing Cores 4th Generation' | 'AI TOPS (1,2) 4000 AI TOPS' | 'Single-precision performance 125 TFLOPS' footnote 1 | 'RT Core performance 380 TFLOPS' footnote 1 | 'GPU memory 96 GB GDDR7 with ECC' | 'Memory interface 512-bit' | 'Memory bandwidth 1792 GB/s' | 'System interface PCIe 5.0 x16' | 'Display connectors 4x DisplayPort 2.1b' | 'Video Engines 4x NVENC (9th Gen) 4x NVDEC (6th Gen)' | 'MIG Support Up to 4x 24 GB / Up to 2x 48 GB / Up to 1x 96 GB' | 'Power consumption Total board power: 600 W' | 'Power connector 1x PCIe CEM5 16-pin' | 'Thermal solution Double-flow-through' | 'Form factor 5.4in H x 12in L, dual slot, extended height'. FOOTNOTES VERBATIM: '1. Peak rates based on GPU Boost Clock.' '2. Effective FP4 TOPS WITH SPARSITY.' FLAG CHECK - this is the row that carries the repo's flag trap, and the footnote is decisive: the only row bearing footnote 2 is 'AI TOPS 4000'. So 4000 is the SPARSE FP4 number and any dense-flagged storage of 4000 would be wrong. The 'Single-precision performance 125 TFLOPS' row bears footnote 1 ONLY (boost clock) and NOT footnote 2, so 125 is a dense figure with no sparsity marker. The datasheet prints NO dense FP4 row anywhere, so the dense twin (~2000) is not a vendor-stated number and must not be recorded as one. Marketing copy in the same document repeats 'up to 4000 TOPS of AI processing power' and '1.8 TB/s of memory bandwidth' (rounded from the 1792 GB/s spec-table value) - the spec table is the precise figure and is what the accelerator record stores. NOT PUBLISHED by this datasheet: any FP16/BF16/TF32/FP8/INT8 TFLOPS row, any boost-clock value, any NVLink figure, and any dense/sparse pair. PCIe is stated only as 'PCIe 5.0 x16'; no GB/s host bandwidth figure is printed.