ultraData Center
H200 NVL GPU
The H200 NVL is the PCIe version of the Hopper H200: a dual-slot, air-cooled card with the same 141 GB of HBM3e at 4.8 TB/s as the SXM module, a power limit of up to 600 W, and 2- or 4-way NVLink bridges in place of the SXM board's NVLink switch.
VRAM 141GB
CUDA Cores 16,896
Tensor Cores 528
TDP 600W
Process 4nm
From
$3.29/hr
across 2 providers
Cloud Pricing
Cheapest on gpu.ai — 37% below avg| Provider | Price / hr |
|---|---|
$3.29/hr 1× | |
$5.39/hr 2×4× | |
$5.40/hr 7× | |
$5.40/hr 3× | |
$5.40/hr 1× | |
$6.53/hr 8× |
Prices updated daily. Last check: Oct 12, 2026
H200 NVL pricing by provider
Every configuration, price rank, and alternative for one provider at a time.
Performance
FP16
835.5 TFLOPS
FP32
60 TFLOPS
BF16
835.5 TFLOPS
FP8
1670.5 TFLOPS
INT8
1670.5 TOPS
Bandwidth
4800 GB/s
Common Use Cases
Inference and fine-tuning of models that need up to 141 GB per GPU in standard air-cooled PCIe servers, with up to four cards bridged over NVLink; MIG partitioning into up to seven instances
Full Specifications
Hardware
- Manufacturer
- NVIDIA
- Architecture
- Hopper
- CUDA Cores
- 16,896
- Tensor Cores
- 528
- Process Node
- 4nm
- TDP
- 600W
Memory & Performance
- VRAM
- 141GB
- Memory Bandwidth
- 4800 GB/s
- FP32
- 60 TFLOPS
- FP16
- 835.5 TFLOPS
- BF16
- 835.5 TFLOPS
- FP8
- 1670.5 TFLOPS
- FP64
- 30 TFLOPS
- INT8
- 1670.5 TOPS
- Release
- 2024