NVIDIAAI GPU · SXM (HGX)
B300
NVIDIA B300 is a Blackwell Ultra data-center accelerator with 270 GB HBM3e and 7.7 TB/s of memory bandwidth.
Specifications
Dense throughput, no sparsity| Architecture | Blackwell Ultra |
|---|---|
| Form factor | SXM (HGX) |
| Process | TSMC 4NP |
| FP16 / BF16 tensor (dense) | 2,250 TFLOPS |
| FP8 tensor (dense) | 4,500 TFLOPS |
| FP4 tensor (dense) | 14,000 TFLOPS |
| Memory | 270 GB HBM3e |
| Memory bandwidth | 7.7 TB/s |
| Power | 1100 W |
| Launch | Aug 2025 |
AI index
Tensor × bandwidth100 × √((2,250 ÷ 989.4) × (7.7 ÷ 3.35)) = 228.6
The AI index is the geometric mean of dense 16-bit tensor throughput and memory bandwidth, each relative to the H100 SXM. It is a spec-sheet comparison, not a measured benchmark.