NVIDIA GB300 NVL72 product page spec table

VERBATIM 2026-10-04 from NVIDIA's markdown rendering of the GB300 NVL72 page (curl -L -A 'Mozilla/5.0 ...'). Spec table headed 'NVIDIA GB300 NVL72' with footnote markers: 'Configuration | 72 NVIDIA Blackwell Ultra GPUs, 36 NVIDIA Grace CPUs' 'NVLink Bandwidth | 130 TB/s' 'Fast Memory | 37 TB' 'GPU Memory | Bandwidth | 20 TB | Up to 576 TB/s' 'CPU Memory | Bandwidth | 17 TB LPDDR5X | 14 TB/s' 'CPU Core Count | 2,592 Arm Neoverse V2 cores' 'FP4 Tensor Core | 1440 | 1080 PFLOPS' [OCR-rendered by NVIDIA's own markdown as '1440 | 10802 PFLOPS'] 'FP8/FP6 Tensor Core | 720 PFLOPS' 'INT8 Tensor Core | 24 POPS' [sic - NVIDIA's page] 'FP16/BF16 Tensor Core | 360 PFLOPS' 'TF32 Tensor Core | 180 PFLOPS' 'FP32 | 6 PFLOPS' [sic - should read TFLOPS] 'FP64 / FP64 Tensor Core | 100 TFLOPS' FOOTNOTES VERBATIM: '1. All Tensor Core specifications are with sparsity unless otherwise noted. 2. Without sparsity.' This is a DIFFERENT sparsity convention from the datasheets: the page says every Tensor Core number is sparse 'unless otherwise noted'. Dividing its system figures by 72: NVFP4 1440/72 = 20 PFLOPS sparse, FP8 720/72 = 10, FP16/BF16 360/72 = 5, TF32 180/72 = 2.5 - all sparse, and exactly the left-hand column of the Blackwell Ultra datasheet's per-GPU rows. Dense FP4 = 10, which CONTRADICTS the datasheet's 15 PFLOPS dense. The datasheet and the product page disagree on dense FP4 for the same silicon (15 vs 10) under their two different footnotes; the datasheet's explicit 'Sparse | Dense' pair is the better evidence. The page's FP64 100 TFLOPS system figure and INT8 '24 POPS' are internally inconsistent with the datasheet (FP64/FP64 per-GPU 1.3 TFLOPS x 72 = 93.6, near 100; INT8 330 TOPS x 72 = 23,760 TOPS, not 24 POPS) and should not be used.