Day 2 ready, fixed-cost infrastructure. Full root access. Powered by OpenStack and Ceph.
Enterprise servers. Supports virtualization, big data, blockchain, and more use cases.
Built for AI training, inference, and HPC, with transparent monthly pricing and no metered hours.
High performance object, block, and file storage. Simple prices, fair egress. Powered by Ceph.
Prefill Wants Compute, Decode Wants Bandwidth: The Case for Two Inference Pools
Prefill is compute-bound, decode is memory-bandwidth-bound. Why splitting inference into two purpose-fit GPU pools beats one uniform fleet.
Running Llama 3.3 70B on an OpenMetal H200
Yes, Llama 3.3 70B runs on a single OpenMetal H200 at FP8 with full 128K context. See the VRAM fit math, KV-cache budget, and vLLM setup.
