DeepSeek R1 · Chip comparison

DeepSeek R1 — B300 vs MI325X

Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and MI325X (AMD CDNA 3) on DeepSeek R1. Latency, throughput, and cost across LLM workloads. Use the chart controls below to switch sequences, precisions, and metrics — same interactions as the main inference chart.

B300 / MI325X on DeepSeek R1 at 35 tok/s/user: 6,590 / 966 tok/s/chip, $0.095 / $0.32 per million tokens. B300 is 70% cheaper per token; B300 delivers 6.8× the tok/s/chip.

Around the middle of the 14–100 tok/s/user interactivity band, at 57 tok/s/user on DeepSeek R1: B300 runs 5,456 tok/s/chip at $0.12 per million tokens, MI325X runs 747 at $0.41. B300 is 72% cheaper per token; B300 delivers 7.3× the tok/s/chip.

Setting 78 tok/s/user as the target on DeepSeek R1, B300 produces 3,144 tok/s/chip ($0.20 per million tokens) and MI325X produces 607 ($0.50). B300 is 60% cheaper per token; B300 delivers 5.2× the tok/s/chip. (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.)

View performance per dollar →

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)
Throughput (tok/s/chip)
B3006,590MI325X966
B3005,456MI325X747
B3003,144MI325X607
Cost ($/M tok)
B300$0.095MI325X$0.32
B300$0.12MI325X$0.41
B300$0.20MI325X$0.50
tok/s/MW
B3003,468,197MI325X571,610
B3002,871,707MI325X442,003
B3001,654,798MI325X358,995
Concurrency
B300~599MI325X~24
B300~219MI325X~11
B300~69MI325X~7