MI355X: FP8 vs BF16 Precision Comparison
How FP8 and BF16 precision affect Qwen 3.5 397B-A17B inference on MI355X (AMD CDNA 4). Throughput, latency, and cost across LLM workloads. Use the chart controls below to switch sequences and metrics — same interactions as the main inference chart.
At 64 tok/s/user on Qwen 3.5 397B-A17B (MI355X), FP8 delivers 2131 tok/s/chip at $0.20 per million tokens; BF16 delivers 879 tok/s/chip at $0.47. FP8 is 143% cheaper per token; FP8 delivers 143% more tok/s/chip. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact.
FP8 posts 1244 tok/s/chip for $0.33 per million tokens at 99 tok/s/user on Qwen 3.5 397B-A17B (MI355X); BF16 posts 597 tok/s/chip for $0.70. FP8 is 108% cheaper per token; FP8 delivers 108% more tok/s/chip. Quantization-level accuracy differences are tracked on the evaluation tab.
Throughput at 133 tok/s/user on Qwen 3.5 397B-A17B (MI355X): FP8 hits 913 tok/s/chip, BF16 hits 279. Per-million costs land at $0.46 and $1.49 respectively. FP8 is 227% cheaper per token; FP8 delivers 227% more tok/s/chip. The cost-throughput tradeoff from lower precision is only part of the picture — see the evaluation page for accuracy data. (Numbers reflect the default 1k/1k selection for this URL — table and chart below update if you change sequence or model in the controls. Each side uses the best available serving configuration for that precision, which may include speculative decoding such as MTP where recipes exist — the same convention as the other comparison pages.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | FP8:2131.3BF16:878.7 | FP8:1244.0BF16:596.7 | FP8:913.0BF16:279.3 |
| Cost ($/M tok) | FP8:$0.195BF16:$0.474 | FP8:$0.335BF16:$0.698 | FP8:$0.456BF16:$1.492 |
| tok/s/MW | FP8:1019777BF16:420444 | FP8:595231BF16:285490 | FP8:436843BF16:133623 |
| Concurrency | FP8:~68BF16:~128 | FP8:~26BF16:~24 | FP8:~14BF16:~9 |
Inference Performance
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.