DeepSeek V4 Pro 0813 1.6T · Chip comparison

DeepSeek V4 Pro 0813 1.6T — GB200 NVL72 vs MI355X

Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and MI355X (AMD CDNA 4) on DeepSeek V4 Pro 0813 1.6T. 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.

AgentX replays real coding-agent sessions rather than fixed-length prompts, so context grows turn over turn and most of each request is served from cache instead of being recomputed. That turns the comparison into a systems question: KV transfer between nodes, prefix-aware routing, and cache capacity all move the curve alongside raw chip throughput. Fixed-sequence workloads stay the clean baseline for kernel and silicon performance, so the two scenarios answer different questions about the same hardware. Learn more about AgentX →

Near the low end of the 56–248 tok/s/user interactivity band, at 104 tok/s/user on DeepSeek V4 Pro 0813 1.6T: GB200 NVL72 runs 39,971 tok/s/chip at $0.013 per million tokens, MI355X runs 23,964 at $0.017. GB200 NVL72 is 26% cheaper per token; GB200 NVL72 delivers 67% more tok/s/chip.

Setting 152 tok/s/user as the target on DeepSeek V4 Pro 0813 1.6T, GB200 NVL72 produces 17,503 tok/s/chip ($0.030 per million tokens) and MI355X produces 18,320 ($0.023). MI355X is 23% cheaper per token; MI355X delivers 5% more tok/s/chip.

At 200 tok/s/user interactivity on DeepSeek V4 Pro 0813 1.6T, GB200 NVL72 delivers 7,437 tok/s/chip at $0.069 per million tokens; MI355X delivers 6,114 tok/s/chip at $0.068. MI355X is 2% cheaper per token; GB200 NVL72 delivers 22% more tok/s/chip at this point. (Numbers reflect the default agentic-traces · fp4 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)
GB200 NVL7239,971MI355X23,964
GB200 NVL7217,503MI355X18,320
GB200 NVL727,437MI355X6,114
Cost ($/M tok)
GB200 NVL72$0.013MI355X$0.017
GB200 NVL72$0.030MI355X$0.023
GB200 NVL72$0.069MI355X$0.068
tok/s/MW
GB200 NVL7221,374,840MI355X11,465,980
GB200 NVL729,359,712MI355X8,765,690
GB200 NVL723,977,117MI355X2,925,295
Concurrency
GB200 NVL72~129MI355X~48
GB200 NVL72~63MI355X~33
GB200 NVL72~18MI355X~10