DeepSeek R1 · Chip 对比

DeepSeek R1 — B200 vs GB300 NVL72

B200(NVIDIA Blackwell)与 GB300 NVL72(NVIDIA Blackwell)在 DeepSeek R1 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 70 tok/s/user 为目标在 DeepSeek R1 上运行时,B200 产出 1058 tok/s/Chip(每百万 token $0.46),GB300 NVL72 产出 3521($0.18)。GB300 NVL72 每 token 成本低 151%;GB300 NVL72 每 Chip 吞吐量高出 233%。

在 DeepSeek R1 上以 127 tok/s/user 交互性运行时,B200 吞吐量为 275 tok/s/Chip,每百万 token 成本 $1.75;GB300 NVL72 吞吐量为 848 tok/s/Chip,成本 $0.75。GB300 NVL72 每 token 成本低 133%;GB300 NVL72 每 Chip 吞吐量高出 208%。

B200 在 DeepSeek R1 上以 184 tok/s/user 运行时达到 157 tok/s/Chip(每百万 token $3.04);GB300 NVL72 达到 117 tok/s/Chip($5.58)。B200 每 token 成本低 83%;B200 每 Chip 吞吐量高出 34%。 (数据反映此 URL 的默认 1k/1k · fp8 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/chip)
B200:1058.4GB300 NVL72:3521.4
B200:274.8GB300 NVL72:847.5
B200:156.8GB300 NVL72:117.0
Cost ($/M tok)
B200:$0.458GB300 NVL72:$0.183
B200:$1.751GB300 NVL72:$0.750
B200:$3.042GB300 NVL72:$5.578
tok/s/MW
B200:618928GB300 NVL72:1661035
B200:160719GB300 NVL72:399769
B200:91701GB300 NVL72:55169
Concurrency
B200:~62GB300 NVL72:~977
B200:~8GB300 NVL72:~188
B200:~25GB300 NVL72:~20

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:
部署模式:
投机解码: