Straight, dated answers to the questions buyers ask about NVIDIA and Supermicro AI servers: where to buy in the UK, what they cost, H100 versus H200, and how many GPUs a 70B model needs.
SuperServer.ai supplies Supermicro NVIDIA HGX systems to the UK and Europe with a delivery target of 4 to 5 weeks from order, against an industry norm of two months or more, through a direct channel into Taiwan's manufacturing ecosystem.
An eight-GPU NVIDIA HGX H200 server from Supermicro costs in the low-to-mid six figures in pounds sterling excluding VAT, and SuperServer.ai delivers it to the UK in 4 to 5 weeks from order.
The H200 is the better inference GPU: the same Hopper compute as the H100 with 141GB of HBM3e at 4.8TB/s against the H100's 80GB at 3.35TB/s, which NVIDIA measures as up to 1.9× the throughput on Llama 2 70B.
To run a 70B model in 16-bit you need about 140GB of GPU memory for the weights plus room for the KV cache: two NVIDIA H100 80GB GPUs, or one H200 141GB at a small batch. At 8-bit it fits in roughly 70GB, and at 4-bit in about 35GB.