DeepSeek V4 Pro 1.6T · GPU 对比

DeepSeek V4 Pro 1.6T — GB200 NVL72 vs MI355X

GB200 NVL72NVIDIA Blackwell)与 MI355XAMD CDNA 4)在 DeepSeek V4 Pro 1.6T 上的正面 AI 推理基准测试对比。涵盖各类 LLM 工作负载的延迟、吞吐量与成本。使用下方图表控件切换序列、精度和指标——交互方式与主推理图表相同。

DeepSeek V4 Pro 1.6T 在 34 tok/s/user 交互性下的吞吐量:GB200 NVL72 为 9617 tok/s/GPU,MI355X 为 2683。每百万 token 成本分别为 $0.06 和 $0.16。GB200 NVL72 每 token 成本低 145%;GB200 NVL72 每 GPU 吞吐量高出 258%。

GB200 NVL72 / MI355X 在 DeepSeek V4 Pro 1.6T 上以 63 tok/s/user 运行:9177 / 1049 tok/s/GPU,$0.07 / $0.39 每百万 token。GB200 NVL72 每 token 成本低 486%;GB200 NVL72 每 GPU 吞吐量高出 775%。

在 6–120 tok/s/user 交互性区间的高端,即 DeepSeek V4 Pro 1.6T 上以 92 tok/s/user 运行时:GB200 NVL72 达到 7154 tok/s/GPU($0.09/百万 token),MI355X 达到 586($0.72/百万)。GB200 NVL72 每 token 成本低 741%;GB200 NVL72 每 GPU 吞吐量高出 1121%。 (数据反映此 URL 的默认 8k/1k · fp4 选择——如果您在控件中更改序列、精度或模型,下方表格和图表会自动更新。)

查看每美元性能对比 →

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/gpu)
GB200 NVL72:9616.5MI355X:2683.2
GB200 NVL72:9177.0MI355X:1049.4
GB200 NVL72:7154.0MI355X:585.8
Cost ($/M tok)
GB200 NVL72:$0.064MI355X:$0.157
GB200 NVL72:$0.067MI355X:$0.392
GB200 NVL72:$0.086MI355X:$0.720
tok/s/MW
GB200 NVL72:5142524MI355X:1283814
GB200 NVL72:4907501MI355X:502088
GB200 NVL72:3825657MI355X:280264
Concurrency
GB200 NVL72:~12680MI355X:~79
GB200 NVL72:~11085MI355X:~16
GB200 NVL72:~2628MI355X:~7

推理性能

不同模型、硬件配置和服务参数下的推理性能指标。

厂商:
聚合模式:
投机解码: