NVIDIA B200 · Blackwell

BF16, dense

2,200 TFLOPS FP16/BF16 Tensor Core Dense/Sparse 2.2/4.5 petaFLOPS 2.2/4.5 petaFLOPS NVIDIA Blackwell Architecture Technical Brief (V2.1) (2.2 PFLOPS on the page; read from a table with a column for each product, whose head reads “Table 3. System Specifications for HGX B300 and HGX B200 HGX B300 HGX B200”)

FP8, dense

4,500 TFLOPS FP8/FP6 Tensor Core Dense/Sparse 4.5/9 petaFLOPS 4.5/9 petaFLOPS NVIDIA Blackwell Architecture Technical Brief (V2.1) (4.5 PFLOPS on the page; read from a table with a column for each product, whose head reads “Table 3. System Specifications for HGX B300 and HGX B200 HGX B300 HGX B200”)

FP4, dense

9,000 TFLOPS FP4 Tensor Core Dense/Sparse 14/18 petaFLOPS 9/18 petaFLOPS NVIDIA Blackwell Architecture Technical Brief (V2.1) (9 PFLOPS on the page; read from a table with a column for each product, whose head reads “Table 3. System Specifications for HGX B300 and HGX B200 HGX B300 HGX B200”)

Memory

180 GB 186 GB HBM3E | 8 TB/s 186 GB HBM3E | 8 TB/s 180 GB HBM3E | 7.7 TB/s NVIDIA Blackwell datasheet (Oct 2025) (read from a table with a column for each product, whose head reads “Technical Specifications1 GB200 NVL72 GB200 NVL4 HGX B200”)

Memory bandwidth

7.7 TB/s 186 GB HBM3E | 8 TB/s 186 GB HBM3E | 8 TB/s 180 GB HBM3E | 7.7 TB/s NVIDIA Blackwell datasheet (Oct 2025) (read from a table with a column for each product, whose head reads “Technical Specifications1 GB200 NVL72 GB200 NVL4 HGX B200”)

Maximum power

1,000 W Max thermal design power (TDP) 1100 W 1000W NVIDIA Blackwell Architecture Technical Brief (V2.1) (read from a table with a column for each product, whose head reads “Table 3. System Specifications for HGX B300 and HGX B200 HGX B300 HGX B200”)

Compute per watt

2,200 GFLOPS/Wdense BF16 over maximum power; 4,500 GFLOPS/W at FP8

Per-GPU figures are for the HGX B200 GPU (second column of each table). The technical brief says 'Up to 192 GB'; the Blackwell datasheet gives 180 GB, which matches the DGX B200 total of 1,440 GB for 8 GPUs, so 180 GB is used. NVIDIA first said 'The Blackwell platform is in full production' on 2024-05-22, but its first statement of sales is the 2025-02-26 release used here.

1. 18 Mar 2024 Promised Blackwell-based products available from partners starting later this year in 2024. Blackwell-based products will be available from partners starting later this year. NVIDIA Blackwell Platform Arrives to Power a New Era of Computing
2. 26 Feb 2025 Shipped: Blackwell in massive-scale production, billions of dollars in sales in its first quarter. We've successfully ramped up the massive-scale production of Blackwell AI supercomputers, achieving billions of dollars in sales in its first quarter. NVIDIA Announces Financial Results for Fourth Quarter and Fiscal 2025

Epoch AI’s names for this chip: B200.

• Tesla V100 SXM2 417 gigaflops of dense FP16 compute per watt, at 300 W
• A100 SXM 80GB 780 gigaflops of dense BF16 compute per watt, at 400 W
• H100 SXM 1,414 gigaflops of dense BF16 compute per watt, at 700 W
• H200 SXM 1,414 gigaflops of dense BF16 compute per watt, at 700 W
• GH200 Grace Hopper Superchip 990 gigaflops of dense BF16 compute per watt, at 1,000 W
• GB200 Grace Blackwell Superchip 5,000 teraflops of dense BF16 compute a chip; NVIDIA publishes no power figure
• B300 (Blackwell Ultra) 2,000 gigaflops of dense BF16 compute per watt, at 1,100 W
• Rubin GPU 4,000 teraflops of dense BF16 compute a chip; NVIDIA publishes no power figure