Key Architecture Highlights
- FP64 Compute performance: FP64: 30 teraFLOPs; FP64 Tensor Core: 60 teraFLOPs
- FP32 Compute performance: 60 teraFLOPs
- Tensor Compute performance: TF32 Tensor Core: 835 teraFLOPs; BFLOAT16 Tensor Core: 1,671 teraFLOPS; FP16 Tensor Core: 1,671 teraFLOPS; FP8 Tensor Core: 3,341 teraFLOPS
Applications
- AI training and inference
- High-performance computing
- Data center
Engineering Parameters
| FP64 Compute performance | FP64: 30 teraFLOPs FP64 Tensor Core: 60 teraFLOPs |
|---|---|
| FP32 Compute performance | 60 teraFLOPs |
| Tensor Compute performance | TF32 Tensor Core: 835 teraFLOPs BFLOAT16 Tensor Core: 1,671 teraFLOPS FP16 Tensor Core: 1,671 teraFLOPS FP8 Tensor Core: 3,341 teraFLOPS |
| Integer Compute performance | 3,341 TOPS |
| GPU memory ／ Memory | 94GB (per GPU, not double-cards) |
| GPU memory bandwidth | 3.9TB/s |
| Video decoding | 7 NVDEC 7 JPEG |
| Power consumption | 350-400W (configurable) |
| MIG Division | Up to 7 MIGS @ 12GB each |
| Form factor and installation | PCIe dual-slot air-cooled |
| Interconnect Interface | NVIDIA NVLink: 600GB/s PCIe Gen5: 128GB/s |
Warranty
Public manufacturer information does not specify the warranty term applicable to our supply channel. The supply region, channel and actual warranty service are confirmed in the supply contract.
Manufacturer information
Manufacturer informationVerified Documentation & Datasheets
Manufacturer product information
Download Document