# DediGPU

> DediGPU rents dedicated, bare-metal GPU servers — 13 models from the NVIDIA B300 to the RTX 4090, plus the AMD MI300X — billed monthly at a published cost sheet with zero margin on the server. Prices run from $139 to $1,762 per card per month, in 4 regions (Reykjavík, Helsinki, Amsterdam, Singapore), with payment in cryptocurrency (Bitcoin, Monero, Ethereum, Tether, Litecoin, TRON, Solana) and no identity check.

Key facts (checked 2 September 2026):

- The monthly price of every server is the sum of six published cost lines (hardware amortised over 36 months, power at $0.052/kWh with PUE 1.15, rack and cooling, network and IPv4, 2%/year spares reserve, operations), rounded up to the next dollar. The margin lives in optional extras, never in the server.
- Servers hold 1, 2, 4 or 8 cards; the price is linear (n cards = n × the per-card price). Prepaying 3, 6 or 12 months costs 3%, 6% or 10% less per month.
- $0 setup fee, $0 egress (unmetered bandwidth), one public IPv4 included, root SSH and IPMI, deploy in under 5 minutes, 99.9% uptime SLA credited ×10 automatically.
- Sign-up needs an email address and a password only: no KYC, no card, no phone. Balance is prepaid in crypto (minimum top-up $40) and unused balance is refundable to the sending address.
- Every page on this site has a Markdown mirror for agents: append `.md` to its URL (the homepage is [/index.md](https://dedigpu.com/index.md)).

## GPU servers (per card, per month, at cost)

- [Compare all 13 GPU servers](https://dedigpu.com/gpu.md): specs and monthly at-cost prices side by side
- [NVIDIA B300 SXM6 dedicated server](https://dedigpu.com/gpu/b300.md): 288 GB HBM3e, 8 TB/s, $1,762/mo (cheapest we found elsewhere: $5,840)
- [NVIDIA B200 SXM6 dedicated server](https://dedigpu.com/gpu/b200.md): 180 GB HBM3e, 8 TB/s, $1,408/mo (cheapest we found elsewhere: $3,723)
- [NVIDIA H200 SXM5 dedicated server](https://dedigpu.com/gpu/h200.md): 141 GB HBM3e, 4.8 TB/s, $1,252/mo (cheapest we found elsewhere: $2,037)
- [NVIDIA H100 SXM5 80 GB dedicated server](https://dedigpu.com/gpu/h100-sxm.md): 80 GB HBM3, 3.35 TB/s, $1,134/mo (cheapest we found elsewhere: $1,230)
- [NVIDIA H100 PCIe 80 GB dedicated server](https://dedigpu.com/gpu/h100-pcie.md): 80 GB HBM2e, 2 TB/s, $810/mo (cheapest we found elsewhere: $1,278)
- [AMD Instinct MI300X dedicated server](https://dedigpu.com/gpu/mi300x.md): 192 GB HBM3, 5.3 TB/s, $782/mo (cheapest we found elsewhere: $1,891)
- [NVIDIA A100 SXM4 80 GB dedicated server](https://dedigpu.com/gpu/a100-80.md): 80 GB HBM2e, 2.04 TB/s, $654/mo (cheapest we found elsewhere: $694)
- [NVIDIA RTX PRO 6000 Blackwell dedicated server](https://dedigpu.com/gpu/rtx-pro-6000.md): 96 GB GDDR7 ECC, 1.79 TB/s, $493/mo (cheapest we found elsewhere: $949)
- [NVIDIA L40S dedicated server](https://dedigpu.com/gpu/l40s.md): 48 GB GDDR6 ECC, 864 GB/s, $356/mo (cheapest we found elsewhere: $511)
- [NVIDIA RTX 6000 Ada dedicated server](https://dedigpu.com/gpu/rtx-6000-ada.md): 48 GB GDDR6 ECC, 960 GB/s, $331/mo (cheapest we found elsewhere: $613)
- [NVIDIA RTX A6000 dedicated server](https://dedigpu.com/gpu/a6000.md): 48 GB GDDR6 ECC, 768 GB/s, $208/mo (cheapest we found elsewhere: $256)
- [NVIDIA GeForce RTX 5090 dedicated server](https://dedigpu.com/gpu/rtx-5090.md): 32 GB GDDR7, 1.79 TB/s, $163/mo (cheapest we found elsewhere: $402)
- [NVIDIA GeForce RTX 4090 dedicated server](https://dedigpu.com/gpu/rtx-4090.md): 24 GB GDDR6X, 1.01 TB/s, $139/mo (cheapest we found elsewhere: $274)

## Pricing and platform

- [Pricing method and full grid](https://dedigpu.com/pricing.md): the six cost lines, the constants, every price from 1 to 8 cards, term discounts, options
- [Clusters](https://dedigpu.com/clusters.md): 16 to 512 GPUs on one InfiniBand fabric at the same per-card price
- [Regions](https://dedigpu.com/regions.md): Reykjavík, Helsinki, Amsterdam, Singapore — power, cooling and latency per region
- [Templates](https://dedigpu.com/templates.md): OS images and one-click stacks (PyTorch, vLLM, SGLang, TGI, Ollama, ComfyUI, Slurm, k3s…), ports and first commands
- [API reference](https://dedigpu.com/api.md): the customer API (servers, GPUs, account) with token auth

## Documentation

- [Getting started](https://dedigpu.com/docs/getting-started.md): From an empty account to a root shell on a dedicated GPU server in four steps: sign up, top up in crypto, order, connect.
- [What a server includes](https://dedigpu.com/docs/what-you-get.md): Exactly what comes with a DediGPU server: the hardware, the network, the access, the software, and what is deliberately not included.
- [Account and security](https://dedigpu.com/docs/account-security.md): How a DediGPU account is protected: password rules, sessions, anti-robot checks and API tokens.
- [Balance and top-ups](https://dedigpu.com/docs/balance-and-top-ups.md): How the prepaid balance works: coins and networks accepted, minimum and maximum, confirmations, what happens when a payment is short or late, and refunds.
- [Terms and renewals](https://dedigpu.com/docs/terms-and-renewals.md): How term billing works: 1, 3, 6 or 12 months prepaid, the discount per term, automatic renewal at a locked price, grace, suspension, extending and cancelling.
- [Ordering a server](https://dedigpu.com/docs/ordering-a-server.md): The six steps of the order form and the rules it enforces: card and count, region, OS and template, pre-loaded model, storage and network options, term and access.
- [SSH and root access](https://dedigpu.com/docs/ssh-and-root-access.md): Connecting to a DediGPU server: SSH keys, the generated root password, hostnames, first-login checks and the hardening we recommend.
- [IPMI console](https://dedigpu.com/docs/ipmi-console.md): Out-of-band access to a DediGPU server: the IPMI console, power control and virtual media, reached through the region jump host.
- [Managing a running server](https://dedigpu.com/docs/server-actions.md): What the server page lets you do: reboot, rename, extend the term, switch auto-renewal, resume a suspended server, cancel, and what each action costs.
- [Operating systems and drivers](https://dedigpu.com/docs/operating-systems.md): The five operating systems offered on DediGPU servers, the driver and toolkit each ships with, which cards they run on, and how long they are supported.
- [Templates and ports](https://dedigpu.com/docs/templates-and-ports.md): What each of the fourteen DediGPU templates installs, which port it listens on, the first command to check it, and how to change or remove it.
- [Pre-loading a model](https://dedigpu.com/docs/preloading-a-model.md): How serving templates pull a model before first boot: Hugging Face and Ollama names, precision, the memory-fit rule the order form enforces, changing the model later.
- [Windows Server](https://dedigpu.com/docs/windows-server.md): Windows Server 2022 on DediGPU: which cards, what the licence costs, RDP access, drivers, and the templates available on Windows.
- [Networking](https://dedigpu.com/docs/networking.md): The network side of a DediGPU server: port speeds, the public IPv4 and reverse DNS, extra addresses, private VLANs, 25 Gbit/s uplinks, free egress, and what is filtered.
- [Storage and backups](https://dedigpu.com/docs/storage-and-backups.md): Local NVMe on DediGPU servers: capacities per card, extra NVMe, RAID-1, what happens to data on suspension and cancellation, and why you need your own backups.
- [Limits and fair use](https://dedigpu.com/docs/limits-and-fair-use.md): The few limits on a DediGPU server: no throttling of GPU load, unmetered ports, port 25 closed by default, eight extra IPv4, five SSH keys, the acceptable-use policy in short.

## Guides

- [How much VRAM do you need to run an LLM? Every size, every precision](https://dedigpu.com/guides/how-much-vram-to-run-an-llm.md): A single rule, a table for eleven popular models from 8B to 671B at BF16, FP8, INT8 and INT4, and the cheapest dedicated server that fits each one.
- [H100 vs H200 vs B200 vs B300: which one to rent in 2026](https://dedigpu.com/guides/h100-vs-h200-vs-b200-vs-b300.md): Memory, bandwidth, NVLink and price per month of the four NVIDIA data-centre generations on the catalogue, and a plain answer for training, fine-tuning and inference.
- [RTX 4090 vs RTX 5090 for Stable Diffusion, Flux and video generation](https://dedigpu.com/guides/rtx-4090-vs-rtx-5090-for-diffusion.md): Memory, FP8 and FP4, real workflow fit for SDXL, Flux.1, Wan and HunyuanVideo, and the monthly price of each card on a dedicated server.
- [Renting vs buying an H100: the 36-month math, line by line](https://dedigpu.com/guides/renting-vs-buying-an-h100.md): What an H100 costs to own and run over three years, from the same cost sheet that prices our servers, and the break-even against renting at cost.
- [Serve Llama 3.3 70B with vLLM on a dedicated GPU server, step by step](https://dedigpu.com/guides/serve-llama-70b-with-vllm.md): Pick the server that fits, pre-load the model at order time, check the OpenAI-compatible API, tune context and parallelism, and put a token in front of it.
- [Fine-tune an 8B model with QLoRA on a single RTX 4090](https://dedigpu.com/guides/qlora-fine-tune-on-one-rtx-4090.md): Everything that fits in 24 GB: 4-bit base weights, LoRA adapters, a real dataset, the training script, memory numbers, hours, and what it costs on a server billed at cost.
- [Dedicated GPU server vs GPU cloud: what the hourly price hides](https://dedigpu.com/guides/dedicated-gpu-server-vs-gpu-cloud.md): The lines that do not appear on an hourly GPU price: storage that keeps billing, egress, idle hours, pre-emption, shared hosts. And when the cloud is still the right call.
- [How to pay for a GPU server in crypto, without KYC, and get a refund if you need one](https://dedigpu.com/guides/pay-for-a-gpu-server-with-crypto.md): The prepaid balance explained: which coin to use and why, fees and confirmation times, what to do when a payment is short, how refunds go back to the address they came from.
- [Multi-GPU training: NVLink, InfiniBand, and when one 8× node beats two 4× servers](https://dedigpu.com/guides/multi-gpu-nvlink-infiniband.md): How the interconnect decides training speed: PCIe vs NVLink bandwidth, tensor and pipeline parallelism, InfiniBand between nodes, and how to size a cluster.
- [Run ComfyUI on a rented GPU server: setup, checkpoints, secure access](https://dedigpu.com/guides/run-comfyui-on-a-rented-gpu-server.md): Order a server with the ComfyUI template, reach it safely through an SSH tunnel, bring your checkpoints, add custom nodes, and keep it running for months.

## Company and policies

- [About DediGPU](https://dedigpu.com/about.md): who runs it, what we own, why at cost
- [Trust and security](https://dedigpu.com/security.md): platform, network and account security measures
- [Contact](https://dedigpu.com/contact.md): support is a ticket from the console, answered around the clock
- [Terms of service](https://dedigpu.com/legal/terms.md): The terms under which DediGPU rents dedicated GPU servers: prepaid crypto balance, monthly terms, automatic renewal, grace, refunds, acceptable use, liability.
- [Privacy policy](https://dedigpu.com/legal/privacy.md): What DediGPU stores about you (an email, hashes, invoice references, a login IP), what it never asks for, how long it is kept, and who can see it.
- [Acceptable use policy](https://dedigpu.com/legal/aup.md): What may and may not run on a DediGPU server: attacks, malware, spam and illegal content are out; training, serving, rendering and full-power workloads are welcome.
- [Service level agreement](https://dedigpu.com/legal/sla.md): The DediGPU SLA: 99.9% monthly availability per server, how downtime is measured, what is excluded, and credits of ten times the downtime, paid automatically.

## Optional

- [llms-full.txt](https://dedigpu.com/llms-full.txt): the whole site — catalogue, cost sheets, docs, guides, FAQ, legal — in one Markdown document
- [Live service status](https://dedigpu.com/status): current availability and incident history (also [JSON](https://dedigpu.com/status.json))
- [Sitemap](https://dedigpu.com/sitemap.xml)