DeepSeek V4 Pro 1.6T · Chip 对比

DeepSeek V4 Pro 1.6T — GB300 NVL72 vs MI355X

GB300 NVL72(NVIDIA Blackwell)与 MI355X(AMD 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 43 tok/s/user 为目标在 DeepSeek V4 Pro 1.6T 上运行时,GB300 NVL72 产出 11707 tok/s/Chip(每百万 token $0.05),MI355X 产出 3219($0.13)。GB300 NVL72 每 token 成本低 138%;GB300 NVL72 每 Chip 吞吐量高出 264%。

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/Chip,每百万 token 成本 $0.07;MI355X 吞吐量为 1664 tok/s/Chip,成本 $0.25。GB300 NVL72 每 token 成本低 260%;GB300 NVL72 每 Chip 吞吐量高出 456%。

GB300 NVL72 在 DeepSeek V4 Pro 1.6T 上以 103 tok/s/user 运行时达到 6534 tok/s/Chip(每百万 token $0.10);MI355X 达到 793 tok/s/Chip($0.52)。GB300 NVL72 每 token 成本低 429%;GB300 NVL72 每 Chip 吞吐量高出 724%。 (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/chip)
GB300 NVL72:11706.7MI355X:3219.4
GB300 NVL72:9255.8MI355X:1664.3
GB300 NVL72:6533.9MI355X:793.0
Cost ($/M tok)
GB300 NVL72:$0.055MI355X:$0.130
GB300 NVL72:$0.069MI355X:$0.249
GB300 NVL72:$0.099MI355X:$0.523
tok/s/MW
GB300 NVL72:5522011MI355X:1540404
GB300 NVL72:4365947MI355X:796312
GB300 NVL72:3082049MI355X:379414
Concurrency
GB300 NVL72:~4613MI355X:~149
GB300 NVL72:~1206MI355X:~29
GB300 NVL72:~587MI355X:~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:
部署模式:
投机解码:
GB300 NVL72 vs MI355X:DeepSeek V4 Pro 推理基准测试 | InferenceX