NVIDIA H200 NVL, original version, part number 900-21010-0040-000.
Built for LLM inference and fine-tuning, generative AI, HPC simulation and memory-bound data center workloads.
Technical specifications
Compute
- GPU model
- NVIDIA H200 NVL
- Architecture
- NVIDIA Hopper
- FP64
- 30 TFLOPS
- FP64 Tensor Core
- 60 TFLOPS
- FP32
- 60 TFLOPS
- TF32 Tensor Core
- 835 TFLOPS*
- BF16 / FP16 Tensor Core
- 1,671 TFLOPS*
- FP8 / INT8 Tensor Core
- 3,341 TFLOPS*
Memory
- GPU memory
- 141GB HBM3e
- Memory bandwidth
- Up to 4.8 TB/s
System
- Interface
- PCIe Gen5 x16 (128 GB/s)
- NVLink
- Up to 900 GB/s with NVLink bridge
- Multi-Instance GPU
- Up to 7 MIG instances
- Power
- Up to 600W, configurable TDP
- Cooling
- Passive — server airflow required
- Part number
- 900-21010-0040-000 (original version)
Notes
- Passively cooled data center card — requires a server chassis with front-to-back airflow, power delivery and thermal support for 600W accelerators.
- Request a quote for current pricing and quantity.
Compliance & Important Information