DeepSeek V4 Pro 1.6T · GPU 对比

DeepSeek V4 Pro 1.6T — GB300 NVL72 vs MI355X

GB300 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。

以 44 tok/s/user 为目标在 DeepSeek V4 Pro 1.6T 上运行时,GB300 NVL72 产出 11642 tok/s/GPU(每百万 token $0.06),MI355X 产出 3048($0.14)。GB300 NVL72 每 token 成本低 115%;GB300 NVL72 每 GPU 吞吐量高出 282%。

在 DeepSeek V4 Pro 1.6T 上以 73 tok/s/user 交互性运行时,GB300 NVL72 吞吐量为 9256 tok/s/GPU,每百万 token 成本 $0.08;MI355X 吞吐量为 1369 tok/s/GPU,成本 $0.30。GB300 NVL72 每 token 成本低 277%;GB300 NVL72 每 GPU 吞吐量高出 576%。

GB300 NVL72 在 DeepSeek V4 Pro 1.6T 上以 102 tok/s/user 运行时达到 6567 tok/s/GPU(每百万 token $0.11);MI355X 达到 815 tok/s/GPU($0.50)。GB300 NVL72 每 token 成本低 347%;GB300 NVL72 每 GPU 吞吐量高出 706%。 (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)
GB300 NVL72:11642.1MI355X:3047.8
GB300 NVL72:9255.8MI355X:1369.2
GB300 NVL72:6567.0MI355X:814.8
Cost ($/M tok)
GB300 NVL72:$0.063MI355X:$0.136
GB300 NVL72:$0.080MI355X:$0.300
GB300 NVL72:$0.113MI355X:$0.504
tok/s/MW
GB300 NVL72:5491567MI355X:1458279
GB300 NVL72:4365947MI355X:655106
GB300 NVL72:3097648MI355X:389857
Concurrency
GB300 NVL72:~4152MI355X:~121
GB300 NVL72:~1206MI355X:~20
GB300 NVL72:~581MI355X:~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:
GB300 NVL72 vs MI355X:DeepSeek V4 Pro 推理基准测试 | InferenceX