Northbit GPU A100

A single-socket box built around an NVIDIA A100 80 GB SXM accelerator, paired with a 24-core EPYC, 256 GB of DDR4 ECC, and 4 TB of fast NVMe. The everyday GPU SKU for training small models, fine-tuning open-weight LLMs, and inference at scale.

Provisioning takes around 52 seconds from API call to SSH-ready. Boots into Ubuntu 24.04 LTS by default with the NVIDIA driver stack pre-installed — nvidia-smi works on the first ssh. CUDA, cuDNN, and the major frameworks (PyTorch, TensorFlow, JAX) ship as opt-in images.

Hourly billing is the default. Reserve for one month, three months, or a year and the rate drops by 18%, 26%, and 38% respectively. We bill by the second, with no minimums. Stopped servers do not bill — only running ones do.

Available in four regions: eu-north-1, eu-west-2, us-east-1, apac-sg-1. Inter-region transfer between Northbit regions is free. Egress to the public internet is metered after the first 5 TB per month.