LumaDock is a London-based hosting provider running VPS, GPU and dedicated servers on its own hardware across the EU, UK, and US. Servers deploy in seconds. Support is 24/7 and in-house. No setup fees, no contracts and no surprises on the invoice.
Rent a dedicated NVIDIA Blackwell GPU by the month
Every plan gets a physical NVIDIA card passed through to a single server, so it answers to one machine only and there's no vGPU slicing. The range runs from RTX PRO 2000 Blackwell at 16 GB up to RTX PRO 6000 Blackwell at 96 GB of GDDR7 ECC, each paired with AMD EPYC cores and NVMe storage. Deploy in seconds, keep full root, and serve Llama or Qwen through vLLM or Ollama on hardware we own and run ourselves. Every GPU plan carries a 7-day money-back guarantee.
I m part of the team at LumaDock (a developer-focused cloud platform built by infra people who actually own their hardware). We ve been quietly working on a new line of Ryzen VPS hosting that s about to go live, and I d love to share what we ve built and get your thoughts before launch.
Hi Product Hunt, Andrei here, CMO at LumaDock.
We started renting many GPU servers with a single Tesla T4 tier, and for a while that covered what people asked us for. Then the requests changed. Someone wants Llama 70B served at 8-bit for an internal tool, someone else needs 48 GB for a LoRA run, and a T4 gets neither of them there.
So we went full-Blackwell :) The range now goes from RTX PRO 2000 with 16 GB up to RTX PRO 6000 with 96 GB of GDDR7 ECC, and every card is passed straight through to one server. You get root, you install whichever CUDA stack you prefer or take our template, and the card behaves like one sitting under your desk.
Two things worth knowing before you click through. Cards sell out, so some tiers show a waitlist instead of a buy button, and we'd rather you hear that here than find it at checkout. If your favorite is out of stock, use the register interest button or reach out to us through a ticket. Each plan is also tied to a physical card, so moving up a tier means deploying on the new plan and bringing your data across.
Every GPU plan comes with 7 days money back, which is enough time to benchmark your own model instead of trusting a spec sheet.
Im curious what everyone here is running. If you're serving a model today, what card size did you settle on, and did you pick it for VRAM headroom or for throughput?
Report
No reviews yetBe the first to leave a review for LumaDock