DeepSeek V4 Pro 1.6T · GPU 对比

DeepSeek V4 Pro 1.6T — GB200 NVL72 vs GB300 NVL72

GB200 NVL72NVIDIA Blackwell)与 GB300 NVL72NVIDIA Blackwell)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。

在 DeepSeek V4 Pro 1.6T 上以 62 tok/s/user 交互性运行时,GB200 NVL72 吞吐量为 9247 tok/s/GPU,每百万 token 成本 $0.07;GB300 NVL72 吞吐量为 9940 tok/s/GPU,成本 $0.07。GB200 NVL72 每 token 成本低 12%;GB300 NVL72 每 GPU 吞吐量高出 7%。

GB200 NVL72 在 DeepSeek V4 Pro 1.6T 上以 109 tok/s/user 运行时达到 6270 tok/s/GPU(每百万 token $0.10);GB300 NVL72 达到 6207 tok/s/GPU($0.12)。GB200 NVL72 每 token 成本低 22%;GB200 NVL72 每 GPU 吞吐量高出 1%。

DeepSeek V4 Pro 1.6T 在 157 tok/s/user 交互性下的吞吐量:GB200 NVL72 为 523 tok/s/GPU,GB300 NVL72 为 965。每百万 token 成本分别为 $1.18 和 $0.76。GB300 NVL72 每 token 成本低 54%;GB300 NVL72 每 GPU 吞吐量高出 85%。 (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 NVL72:9246.6GB300 NVL72:9939.9
GB200 NVL72:6270.4GB300 NVL72:6207.4
GB200 NVL72:522.7GB300 NVL72:964.6
Cost ($/M tok)
GB200 NVL72:$0.066GB300 NVL72:$0.074
GB200 NVL72:$0.098GB300 NVL72:$0.119
GB200 NVL72:$1.176GB300 NVL72:$0.764
tok/s/MW
GB200 NVL72:4944713GB300 NVL72:4688640
GB200 NVL72:3353137GB300 NVL72:2928005
GB200 NVL72:279500GB300 NVL72:455004
Concurrency
GB200 NVL72:~12312GB300 NVL72:~1043
GB200 NVL72:~1934GB300 NVL72:~613
GB200 NVL72:~41GB300 NVL72:~31

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: