Where to rent out your GPU: hosting programmes compared
Vast.ai is one of the few programmes still open to a gaming PC, alongside Salad and Clore.ai; RunPod Community Cloud no longer accepts new hosts. A DGX Spark or another GB10 machine can go on Vast.ai or on GPUwerk Platform, which pays the owner $0.55 per rented hour. Datacenter cards fit TensorDock, which wants new hardware in a data center, or RunPod's partner programme for data centers. I run GPUwerk, which operates GPUwerk Platform, so read my own row with that in mind: it takes only GB10 machines, and the table says where the others are better.
The programmes side by side
Every cell below comes from the platform's own page or docs, checked on 11 October 2026. "Not stated" means the page I read did not say. "Not confirmed" means I could not confirm the point from a first-party page, so I left the figure out. "Inference" marks my own reading of a published rule.
| Platform | Accepts GB10? | Accepts consumer GPUs? | Fee / share (as published) | Payout method and frequency | Who runs the OS | Network requirements | Verification / uptime rules | Source and date |
|---|---|---|---|---|---|---|---|---|
| Vast.ai | Yes: ARM and GB10 support announced on the NVIDIA developer forum in June 2026, and Vast.ai has a GB10 rental page | Yes | No hosting fee since June 2024, replaced by an undisclosed internal surcharge. Vast.ai says the prices hosts set are the prices they earn. | Wise, PayPal or Stripe; weekly invoices on Fridays; first payout can take up to 2 weeks | The host: Ubuntu, drivers and ports. Vast.ai offers no support for getting the machine working. | Host opens ports on the router; renters connect directly for most jobs | Automated checks of reliability, configuration, DLPerf score and demand; hosts should plan for 100% uptime during a rental | Payment docs, hosting overview, verification, June 2024 update, GB10 thread, GB10 page. Checked 11 Oct 2026. |
| RunPod | Not stated | No new hosts: Community Cloud no longer accepts them | Not published; partner compensation "varies by partner track" | Not published | Not stated | Not stated | Secure Cloud runs in T3/T4 data centers; new capacity comes through a partner programme for data centers, cloud providers and GPU operators, reviewed on application | Choose a Pod, Partner Program. Checked 11 Oct 2026. |
| TensorDock | Not stated; the hardware rules do not fit a GB10 (inference) | New hardware must be in a datacenter, so a home PC does not fit (inference) | Hosts set prices | Stripe, whenever you want | TensorDock's hypervisor on Ubuntu 22.04 | 1 Gbit, static IP, 100 forwarded ports | Excessive downtime is penalised; maintenance with two weeks' notice is not. Per GPU: 8+ vCPUs, 16 GB+ RAM, 1 TB NVMe, PCIe 4.0 x8 or 3.0 x16 | TensorDock host page. Checked 11 Oct 2026. |
| Salad | Not stated | Yes; favours NVIDIA 30 and 40-series cards with 12 GB or more | Not published as a percentage | Earnings are Salad Balance, spent on rewards in Salad's storefront | The host: Salad runs as an app on your own PC | Not stated | Not stated on the pages checked | Earnings article, withdrawal cap. Checked 11 Oct 2026. |
| Akash | Not stated | Not stated | Providers set their own prices; a fee is not stated on the page checked | Leases settle in ACT, Akash's compute credit; the provider also needs AKT for bid escrow and fees | The provider: Linux and Kubernetes skills needed | Static IP or dynamic DNS; 100+ Mbps | Not stated on the page checked | Akash provider docs. Checked 11 Oct 2026. |
| io.net | Not stated | Not confirmed | Not stated on the page checked | In IO Coin, claimed to a Solana or Aptos wallet; block rewards in $IO on top | Not confirmed | Not confirmed | Not confirmed | io.net rewards and wallets. Checked 11 Oct 2026. |
| Clore.ai | Not stated; it asks for x86_64 Ubuntu or HiveOS, which a GB10 (ARM) does not fit (inference) | Yes, NVIDIA only; it has an RTX 5090 hosting page | Not confirmed: the fee page could not be loaded | Accrues per rented minute in BTC, CLORE, USDT or USDC | The host: Ubuntu 22.04 or 24.04 on x86_64, or HiveOS, with a working NVIDIA driver | Outbound HTTPS to Docker Hub, NVIDIA, GitHub, GitLab and Clore.ai | Not confirmed | Clore.ai RTX 5090 hosting. Checked 11 Oct 2026. |
| GPUwerk Platform | Yes, GB10 machines only | No | 70% of what renters are charged; $0.55 per rented hour at the current $0.79 rate | Monthly self-billing statement; EUR by SEPA (converted at the ECB rate) or CZK to a Czech bank account, within 30 days of month end; or Wise, PayPal or Stripe | GPUwerk: you install its locked system from a USB stick, which erases the machine; no local login or shell while listed | At least 300 Mbit/s down and 100 Mbit/s up; outbound HTTPS and UDP only, no port forwarding | At least 1 hour of burn-in with a restart, then a GPUwerk review; an approved Spark can be waitlisted. Unlist any time; a rental cut short on your side forfeits its last 24 hours. | Requirements, earnings page, safety page. Checked 11 Oct 2026. |
Checked 11 October 2026. Vast.ai: payment, hosting overview, verification, June 2024 product update, GB10. RunPod: Secure and Community Cloud, Partner Program. TensorDock: host page. Salad: earnings, withdrawal cap. Akash: should I run a provider. io.net: rewards and wallets. Clore.ai: RTX 5090 hosting. GPUwerk figures come from the requirements page, earnings page and safety page. Terms change, so check them before you commit a machine.
Where each one is better
- Vast.ai takes consumer and datacenter cards and now GB10 machines, invoices weekly, and publishes earnings ranges by card. In exchange you are the sysadmin, and its surcharge is undisclosed.
- Salad asks the least of a gaming PC owner: it runs as an app, with no server to administer. You are paid in Salad Balance for its storefront, and it publishes no GB10 or ARM support.
- Clore.ai takes consumer NVIDIA cards on x86_64 Ubuntu or HiveOS and pays in crypto, including USDT and USDC.
- TensorDock suits an operator with data center space, a static IP and spare Ubuntu servers.
- RunPod only matters to you if you run data center capacity. Community Cloud no longer accepts new hosts.
- Akash settles leases in ACT and expects Kubernetes skills.
- GPUwerk Platform has the shortest setup for a GB10 owner: no operating system to run, no open ports, EUR payouts by SEPA or CZK to a Czech bank, or Wise, PayPal or Stripe. It takes nothing else, and renters are offered a partner Spark only when GPUwerk's own are all in use, so hours depend on GPUwerk's demand.
io.net pays workers in IO Coin. I could not confirm its fees, or Clore.ai's, from a first-party page, so I make no claim about them, and none about token prices.
What it really earns
GPUwerk Platform pays $0.55 per rented hour for a GB10, which is 70% of the current $0.79 renter rate. The owner pays electricity and internet, and the hardware stays the owner's risk.
I am not claiming that GPUwerk pays more than any other platform. That figure does not count the hours nobody rents: you earn only on rented hours, and no platform here guarantees any. On GPUwerk Platform a Spark is offered only when GPUwerk's own Sparks are all in use.
Your electricity is the cost to set against it. The power and electricity cost guide has the kWh table for a DGX Spark; a gaming card has its own draw, so use your card's rated board power and your tariff.
Costs every host pays
- Electricity. The machine runs whether it is rented or not. Multiply its average draw by your price per kWh.
- Internet. Renter traffic and model downloads use your line. GPUwerk Platform asks for 300/100 Mbit/s, Akash for 100+ Mbps, TensorDock for 1 Gbit with a static IP.
- Wear and hardware risk. Sustained load ages fans, power supplies and the card. The hardware stays your property and your risk on every platform here.
- Tax. Rental income is normally taxable where you live. I am not a tax adviser, so ask one before you start, and keep the payout statements.
If you own a DGX Spark
Read the GPUwerk Platform overview and the earnings page for the arithmetic, and GPUwerk Platform against Vast.ai for a closer comparison of the two programmes that accept a GB10. The safety page explains how renters are kept away from your network.
FAQ
Can I rent out my gaming GPU?
Yes, on Vast.ai, Salad or Clore.ai. Salad runs as an app on an ordinary PC and says NVIDIA 30 and 40-series cards with 12 GB or more of VRAM get more jobs. Clore.ai has an RTX 5090 hosting page and wants x86_64 Ubuntu or HiveOS. RunPod Community Cloud no longer accepts new hosts, TensorDock wants new hardware in a datacenter, and GPUwerk Platform takes only GB10 machines such as the DGX Spark.
Is RunPod still accepting new hosts?
Not for Community Cloud. RunPod's documentation says it is no longer accepting new hosts for Community Cloud, and existing Community Cloud resources remain available. Secure Cloud runs in T3/T4 data centers, and RunPod's partner programme is aimed at data centers, cloud providers and GPU operators, so a home or office machine has no route in (my inference). I checked RunPod's docs and partner page on 11 October 2026.
How much can I make renting out a GPU?
Nobody can say for your machine. GPUwerk Platform pays $0.55 per rented hour for a GB10 at the current $0.79 renter rate, and that figure does not include your electricity, internet or hardware wear. Hours rented depend on demand, and no platform here guarantees earnings.
Do I need Linux skills to host?
On Vast.ai, TensorDock, Akash and Clore.ai, yes. A Vast.ai host installs Ubuntu and drivers and opens ports, and Vast.ai says it does not offer support for getting your machine working. TensorDock requires Ubuntu 22.04 with its hypervisor. Akash asks for Linux and Kubernetes knowledge. Clore.ai needs Ubuntu or HiveOS with a working NVIDIA driver. Salad is an app you install on your PC, and GPUwerk Platform runs a locked image from a USB stick, so the owner has no operating system to administer.
Which platforms accept a DGX Spark?
Of the programmes here, Vast.ai and GPUwerk Platform. Vast.ai announced ARM and GB10 support in June 2026 and has a GB10 rental page. GPUwerk Platform accepts only GB10 machines. RunPod, TensorDock, Salad, Akash, io.net and Clore.ai do not state GB10 support on the pages I checked. TensorDock's data center rule and Clore.ai's x86_64 requirement do not fit a GB10 on their face, which is my inference, not a statement from either.
Is renting out a GPU safe for my home network?
It depends on how the platform connects to your machine. Vast.ai and TensorDock require open or forwarded ports, which puts a service on your network edge that you must secure yourself. GPUwerk Platform uses an outbound-only tunnel with no port forwarding, its firewall drops connections from other devices on your network, and renter containers are refused your local network. Renter traffic still counts against your bandwidth, and hardware risk stays with you on every platform.