GB300 NVL72
AI·HPC 고성능 GPU 클러스터
AI 추론 성능 및 효율성을 위한 설계

AI 추론 성능 및 효율성을 위한 설계
| Configuration | 72 NVIDIA Blackwell Ultra GPU, 36 NVIDIA Grace CPU |
|---|---|
| NVLink bandwidth | 130TB/s |
| NVLink memory bandwidth | Up to 40 TB |
| GPU memory bandwidth | 20 TB | Up to 576 TB/s |
| CPU memory bandwidth | 17 TB LPDDR5X | 14 TB/s |
| CPU Core | 2,592 ArmNeoverse V2 Core |
| FP4 Tensor Core | 1,400 | 1,100² FLOPS |
| FP8/FP6 Tensor Core | 720 FLOPS |
| INT8 Tensor Core | 23 FLOPS |
| FP16/BF16 Tensor Core | 360 FLOPS |
| TF32 Tensor Core | 180 FLOPS |
| FP32 | 6 FLOPS |
| FP64/FP64 Tensor Core | 2100 Tera FLOPS |