← Clastiq solution brief
Solution brief · NVIDIA HGX H200 · 8 GPU
Clastiq H200
Several production models, a training pool and dozens of notebooks at once. 141 GB per GPU serves 70B-class models at 128k context with headroom for fine-tuning.
Enterprise LLM gatewayRAG at scaleLoRA / QLoRA fine-tuningMulti-team AI platform
8× H200
SXM with NVLink
1.1 TB
HBM3e total
30 TB
NVMe model store
- GPU
- 8 × NVIDIA H200 SXM, 141 GB each, NVLink
- GPU memory
- 1.1 TB total HBM3e
- Serves
- 70B-class models at 128k context, plus fine-tuning
- CPU · Memory
- Dual-socket, 192 cores · 2 TB DDR5 ECC
- Storage
- 30 TB NVMe model and dataset store
- Network
- 2 × 200 GbE, 1 × management
- Software
- Clastiq Enterprise pre-installed, unlimited users and projects
- Support
- 3 years hardware and platform, next-business-day parts