MI300X vs H100 SXM inference cost per token (SemiAnalysis, May 2025): MI300X ahead on Llama 3 405B and at ultra-low latency on Llama 3 70B; behind H200 in most scenarios
wins perf/$ vs H100 on Llama 3 405B and ultra-low-latency Llama 3 70B; loses to H200 in most scenariosobserved
| Value kind | observed — Reported measurements, counts, and specifications keep the precision and scope stated by their source; an exact specification is not treated as a range. |
|---|---|
| Caveat | Previous-generation parts; software-stack results change month to month. The article's Key Insights also names DeepSeek V3 (671B) as an H100 win, but its DeepSeek section did not test H100, so that part is not carried here. |
| As of | 2025-05 |
| Source | SemiAnalysis inference benchmark (cost/Mtoken) · Key Insights; Llama 3 405B and Llama 3 70B results; DeepSeek V3 section (H100 not tested: does not fit DeepSeek V3 in one node). |
| Review | checking…review by 2027-03-31 · slow cadence |
| Recorded changes | last 2026-10-19 · 2 revisions tracked |
| Claim id | mi300x-cost-per-token-advantage-vs-h100-sxm-in |
Where the guide uses it
Not quoted in a chapter yet; it is kept in the curated register.
← Full numbers register — every date-stamped figure in the guide, with revision history.