Guides, benchmarks, and opinions on renting NVIDIA GPUs in the cloud — DGX Spark, H100, H200, B300 — and building AI on greener, cheaper compute.
Your Enverge instance only exposes SSH. Share its web app without opening a port: SSH forwarding for yourself, Cloudflare Tunnel for others, or Tailscale Funnel — with auth and teardown for each.
Inference is not just chat: embeddings, agents, speech, vision, and batch scoring each behave differently. Map the sub-types and see which fit a 128 GB DGX Spark — and which need cloud or a rack.
Bare metal vs VM vs container on a DGX Spark: real GPU overhead numbers, why 128 GB unified memory and no BMC change the answer, and a decision guide.
A visual timeline of every NVIDIA GPU that mattered for AI — GTX 580, K80, P100, V100, T4, A100, H100, H200, B200, GB300, DGX Spark, and Rubin R200 — with specs and what each unlocked.