Hi Hunters,
I’m Nicholas, co-founder of GPUDeploy.
GPUDeploy is like DoorDash or Airbnb for GPUs. You can launch an instance of on-demand A100s right now at $1.28/GPU/hour.
The previous iteration of this startup was a robotics company (drone delivery), which did not work out. However, training our RL models, we were renting GPUs and realized that what's out there is counterintuitive to use and overpriced - plus, the GPUs you want are always booked out.
Meanwhile, there are hundreds of unutilized, high-quality, cheap Nvidia GPU clusters out there. We have written software to make them available on-demand. We have GPUs online from data centers in Europe and the US and are offering them on-demand for wholesale prices that you would normally have to negotiate.
Any session you launch comes with ssh and Jupyter notebook access and is ready to be used for your machine learning needs immediately. I recently trained Andrej Karpathy's nanoGPT to plagiarize Shakespeare for <$0.20 on an RTX 4090.
We have H100s at $3.45, A100s at $1.28 and RTX 4090s at $0.49 available right now.
If you own any of these GPUs (AI company/hobbyist/data center), it is also super easy to take them online on our site and earn money. You can get in contact with us at [email protected] for custom onboardings or onboard yourself at www.gpudeploy.com/control. It takes 5 minutes and we are very flexible: you can take your node online and schedule offline time at any time. We are however opinionated on a high degree of reliability, while your node is online.
Compared to what's out there:
- we make it super simple to launch GPUs at wholesale prices now, no reservations, sales calls or the likes required.
- we have a very clean, to-the-point user experience.
- we don't put huge markups on the GPU prices.
Try it out!
