← All hardware
The memory-upgraded Hopper GPU that remains the AI workhorse.
Pros
- 76% more memory than H100 at similar power
- Broad availability across every cloud
- Strong price/performance for inference
Cons
- Hopper generation now superseded by Blackwell
- Lower FP4 capability than Blackwell
- Still expensive ($30k-$40k) per GPU
✓ Where it shines / best for
- Memory-bound LLM inference and serving
- Large-context generative AI workloads
- Enterprises upgrading from H100 for inference efficiency
✕ Not the best fit for
- Frontier-scale training better served by Blackwell racks
- Edge or on-device deployment
- Small-budget projects
Features
- ✓ AI inference
- ✓ Data-center scale
- ✓ HBM3E
- ✓ NVLink
- ✓ LLM Training
- ✓ FP8
- ✓ Transformer Engine
- ✓ Mig
Pricing
| Plan | Price | Billing | Notes |
|---|---|---|---|
| H200 141GB (street price) | $30,000-$45,000 | one-time | Per-GPU street price (SXM/NVL) via OEM systems and resellers; not retail |
| DGX H200 system (8x H200) | $350,000-$500,000 | one-time | 8-GPU appliance, 1,128GB HBM3e total; via NVIDIA partners |
| Cloud rental (per H200) | ~$3.72-$10.60 | per hour | On-demand neocloud rate; lower with reserved commitments |
Pricing verified from the official source. Prices change often — confirm on the vendor's site before buying.
Specifications
| power | 700W TDP |
| memory | 141 GB HBM3e, 4.8 TB/s |
| nvlink | 4.0, 900 GB/s per GPU |
| cuda_cores | 16,896 |
| architecture | Hopper (TSMC 4N) |
| tensor_cores | 528 (4th-gen) |
| fp8_performance | 3,958 TFLOPS |
Sponsored
A full review is being generated for this product and will appear here shortly.