GLM 5/5.1 · Performance per Dollar
GLM 5/5.1 — H200 vs MI325X Performance per Dollar
Cost per million tokens of H200 (NVIDIA Hopper) versus MI325X (AMD CDNA 3) on GLM 5/5.1. Large-hyperscaler-volume ownership TCO normalized by total tokens — performance per dollar across LLM workloads. Pick the more cost-efficient SKU at every target interactivity level. Use the chart controls below to switch sequences, precisions, and metrics — same interactions as the main inference chart.
At 20 tok/s/user on GLM 5/5.1, H200 costs $0.40 per million tokens; MI325X costs $1.08. H200 delivers 172% more tokens per dollar at this operating point.
H200 edges MI325X at 23 tok/s/user on GLM 5/5.1 — $0.40 per million tokens versus $1.40, a 72% lower cost per token.
Push GLM 5/5.1 to 26 tok/s/user and H200 lands at $0.40 per million tokens against MI325X's $1.92 — H200 costs 79% less. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
Chip pricing (owning hyperscaler): H200 $1.22/chip/hr · MI325X $1.10/chip/hr. Source: SemiAnalysis Market July 2026 Pricing Surveys & AI Cloud TCO Model.
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Dollars per Million Tokens | H200$0.40MI325X$1.08 | H200$0.40MI325X$1.40 | H200$0.40MI325X$1.92 |
| Concurrency | H200~32MI325X~13 | H200~32MI325X~9 | H200~31MI325X~6 |
