Description:
- GPU memory: 40 GB HBM2
- Architecture: NVIDIA Ampere
- Manufacturing process: 7 nm NVIDIA Custom Process (TSMC)
- CUDA cores: 6,912
- Tensor Cores: 432 (3rd Generation)
- Streaming Multiprocessors: 108
- FP64 performance: 9.7 TFLOPS
- FP32 performance: 19.5 TFLOPS
- TF32 Tensor Core: 311.8 TFLOPS*
- INT8 Tensor Core: 1,247.4 TOPS*
- Memory interface: 5,120-bit
- Memory bandwidth: 1,555.2 GB/s
- NVLink bandwidth: 400 GB/s (bidirectional)
- Multi-Instance GPU: up to 7 instances
- System interface: PCI Express 4.0 x16
- Power consumption: 240 W
- Power connector: CEM5 16-pin
- Thermal solution: Active (blower)
- Form factor: Dual slot, low profile (4.4" × 10.5")
- Display output: None (companion GPU required)
* With structural sparsity enabled
