We have a strict process for getting approved. We will only approve hosts that have a solid setup.
Once you're approved, we'll send you a script to install our Libvirt-based hypervisor software. This will automatically add your server to our marketplace.
Last step: set your prices! Once this is done, marketplace users will be able to deploy virtual machines on your hardware! Now that customers are using your server, keep it running reliably — downtime will be penalized!
Becoming a TensorDock host requires a good setup and good hardware.
On our marketplace, customers will be actively using your server. As such, excessive downtime will be penalized. You should be able to keep your servers running 24/7 and investigate issues when they arise.
Before you join, you will be required to sign a data protection policy, which affirms that you will not attempt to look at our customers' data. Customers are here to do more for less, not get packetsniffed. You will also be forbidden from contacting customers directly.
Each location should have...
You need at least a 1 gigabit internet connection for each location that you are hosting servers in. This internet connection can be shared amongst multiple servers.
Customers need to be able to SSH into their machines, so we require a static IP address for each location. Note that we can handle a DDNS setup (contact us), but we would prefer to use an IP address instead.
If you do not have a reliable power supply, you should get UPSes and backup generators. Customers should not have to wait for a machine to get back online.
Each server should have...
When customers rent a GPU, they should be able to utilize it fully. PCIE 3.0 x8 (4.0 x4) simply isn't enough for modern workloads. That's why were're requiring PCIE 3.0 x16 or 4.0 x8 for each GPU.
We recommend at least 16 GB of DDR4 RAM per GPU. For GPUs with larger amounts of memory (e.g. A100 80GB), we'd recommend having slightly more than the VRAM. E.g. for an A6000 48GB, we'd recommend at least 64 GB of RAM.
Each GPU should have access to at least 8 CPU threads (4 physical cores when hyperthreading is enabled) that can boost to a clock speed of at least 3.0 GHz.
Our customers use a lot of storage!
We require 100 ports to be forwarded to each physical server, or a dedicated IP address for each server. We can lend out /24 to /22 ARIN IPv4 address subnets if you would prefer to use our IP addresses instead of yours.
Your server needs these graphics so that the NVIDIA GPU can be fully passed through to the virtual machines that will host customers.
You set your own server's rate. You can charge as much as you think customers will be willing to pay to maximize profits, leading to potentially high earnings
Make money when a GPU is rented, and make additional money when a large amount of CPU/RAM/disk is used as well! Each adds up.
We'll charge a 25% fee, which will go to credit card processing fees, marketing, software development, etc. — all essential to generating adequate demand to continually rent out your servers.
|Graphics Card||Typical Monthly Earnings|
|A6000 / A40||$300|
|Quadro RTX 8000||$200|
|A5000 / A10||$125|
|GeForce 3080 Ti||$100|
|All other NVIDIA 30-series cards equivalent to or better than a GeForce 3060||$50|
|All other NVIDIA RTX GPUs with at least 8 GB VRAM||$30|
|All other NVIDIA GTX GPUs with at least 8 GB VRAM||$20|
Mixed systems work! You can have different NVIDIA GPU types on the same server
Yes, transitioning from being a miner to a TensorDock Marketplace host will likely require some capital investments to upgrade your setup. However, remember that you'll be earning much more as well.
Customers are deployed in isolated virtual machines. They cannot access the bare metal server.
Hosts with good uptime and no data loss will eventually be given "Top Host" badges so that they'll be
prominently. New hosts will start without badge, whereby prices will likely need to be set below the
prices of similar
machines at "Top Hosts" to attract customers. Once hosts demonstrate operational excellence, they'll
Hosts will be able to schedule downtime, which will give us the ability to move the customer to the next available machine. Scheduled downtime won't be penalized.
As of July 2022
As companies scale, marginal costs rise. We're here to unite the independent hosts of the world — the ones who have the lowest costs by sourcing hardware locally and being lean — and place them together on a single, intuitive marketplace to provide the compute that rivals the big clouds but at a fraction of the cost.
We're always receptive to feedback, just contact us!