AMDAI GPU · PCIe
MI100
AMD Instinct MI100 is a CDNA data-center accelerator with 32 GB HBM2 and 1.229 TB/s of memory bandwidth.
Specifications
Dense throughput, no sparsity| Architecture | CDNA |
|---|---|
| Form factor | PCIe |
| Process | TSMC 7 nm |
| FP16 / BF16 tensor (dense) | 184.6 TFLOPS FP16 184.6 · BF16 92.3 |
| FP8 tensor (dense) | — |
| FP4 tensor (dense) | — |
| Memory | 32 GB HBM2 |
| Memory bandwidth | 1.229 TB/s |
| Power | 300 W |
| Launch | Nov 2020 |
AI index
Tensor × bandwidth100 × √((184.6 ÷ 989.4) × (1.229 ÷ 3.35)) = 26.2
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.