NVIDIA Blackwell Ultra datasheet, first-party copy (GB300 NVL72 / HGX B300)

VERBATIM VERIFICATION 2026-10-04, reached from NVIDIA's own resources.nvidia.com landing page for the Blackwell Ultra datasheet. Resolves the reseller-host audit flag on nv-blackwell-ultra-datasheet. Technical Specifications table, two per-GPU columns GB300 NVL72 | HGX B300, verbatim: 'Blackwell Ultra GPUs | Grace CPUs | 72 | 36 | 8 | 0' 'CPU Cores | 2,592 Arm Neoverse V2 Cores | -' 'Total FP4 Tensor Core1 | 1,440 PFLOPS | 1,080 PFLOPS | 144 PFLOPS | 108 PFLOPS' 'Total FP8/FP6 Tensor Core2 | 720 PFLOPS | 72 PFLOPS' 'Total Fast Memory | 37 TB | 2.1 TB' 'Total Memory Bandwidth | 576 TB/s | 62 TB/s' 'Total NVLink Switch Bandwidth | 130 TB/s' / '14.4 TB/s' 'FP4 Tensor Core1 | 20 PFLOPS | 15 PFLOPS | 18 PFLOPS | 14 PFLOPS' 'FP8/FP6 Tensor Core2 | 10 PFLOPS | 9 PFLOPS' 'INT8 Tensor Core2 | 330 TOPS | 307 TOPS' 'FP16/BF16 Tensor Core2 | 5 PFLOPS | 4.5 PLFOPS' [sic: 'PLFOPS' typo is NVIDIA's, in the printed sheet] 'TF32 Tensor Core2 | 2.5 PFLOPS | 2.2 PFLOPS' 'FP32 | 80 TFLOPS | 75 TFLOPS' 'FP64/FP64 Tensor Core | 1.3 TFLOPS | 1.2 TFLOPS' 'GPU Memory | Bandwidth | 279 GB HBM3E | 8 TB/s | 270 GB HBM3E | 7.7 TB/s' 'Max Thermal Design Power (TDP) | Configurable up to 1,400 W | Configurable up to 1,100 W' 'Interconnect | Fifth-Generation NVLink: 1.8 TB/s PCIe Gen6: 256 GB/s' FOOTNOTES VERBATIM: '1. Specification in Sparse | Dense 2. Specification in sparse. Dense is 1/2 sparse spec shown. 3. Supported formats provide these speed-ups over H100 GPUs: 2x H.264, 1.25x HEVC, 1.25x VP9. AV1' THIS SETTLES THE FP4 QUESTION. Footnote 1 on the FP4 row reads 'Specification in Sparse | Dense' and the row prints '20 PFLOPS | 15 PFLOPS' for GB300 NVL72 and '18 PFLOPS | 14 PFLOPS' for HGX B300, i.e. left-to-right SPARSE then DENSE. Dense FP4 is therefore 15 PFLOPS (GB300 column) and 14 PFLOPS (HGX column) - the stored nvidia-b300 fp4 15000 TFLOPS dense:true is CORRECT and the earlier 9000 value was wrong. Every other tensor row carries footnote 2 ('Dense is 1/2 sparse spec shown'), so those dense values are one-half of the printed figure. The INT8 row's '330 TOPS | 307 TOPS' remains an NVIDIA transcription artifact against a 10 PFLOPS FP8 row, confirming the existing record's 'contested' handling. Note the sheet's own 'Total Memory Bandwidth 576 TB/s' is the SYSTEM figure and is identical on the GB200 sheet while per-GPU memory differs (279 vs 186 GB) - do not divide it by 72 to get a per-GPU bandwidth.