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 37 tok/s/user 交互性下的吞吐量:GB200 NVL72 为 9606 tok/s/GPU,MI355X 为 4265。每百万 token 成本分别为 $0.06 和 $0.10。GB200 NVL72 每 token 成本低 51%;GB200 NVL72 每 GPU 吞吐量高出 125%。

GB200 NVL72 / MI355X 在 DeepSeek V4 Pro 1.6T 上以 68 tok/s/user 运行:8901 / 1724 tok/s/GPU,$0.07 / $0.24 每百万 token。GB200 NVL72 每 token 成本低 246%;GB200 NVL72 每 GPU 吞吐量高出 416%。

在 6–130 tok/s/user 交互性区间的高端,即 DeepSeek V4 Pro 1.6T 上以 100 tok/s/user 运行时:GB200 NVL72 达到 6912 tok/s/GPU($0.09/百万 token),MI355X 达到 841($0.49/百万)。GB200 NVL72 每 token 成本低 450%;GB200 NVL72 每 GPU 吞吐量高出 722%。 (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 NVL72:9605.6MI355X:4264.7
GB200 NVL72:8900.9MI355X:1724.0
GB200 NVL72:6911.8MI355X:840.5
Cost ($/M tok)
GB200 NVL72:$0.064MI355X:$0.096
GB200 NVL72:$0.069MI355X:$0.239
GB200 NVL72:$0.089MI355X:$0.489
tok/s/MW
GB200 NVL72:5136680MI355X:2040546
GB200 NVL72:4759857MI355X:824894
GB200 NVL72:3696171MI355X:402167
Concurrency
GB200 NVL72:~13489MI355X:~512
GB200 NVL72:~7562MI355X:~32
GB200 NVL72:~2340MI355X:~8

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:
GB200 NVL72 vs MI355X:DeepSeek V4 Pro 推理基准测试 | InferenceX