TensorPool Logo

A CLI for GPUs

On-demand Multi-node GPU Clusters

TensorPool Logo

A CLI for GPUs

CLI COMMANDS
tp cluster create -i -t 8xH100 -n 4 ~/.ssh/id_rsa.pub
Create mult-node GPU clusters on demand. Always 3.2Tbs Infiniband.
tp nfs create -s 10000
Create shared NFS volumes for your clusters up to hundreds of TB large. 7GB/s reads, 2GB/s writes. Faster than local NVMe storage.
tp job run tp.train.toml
Kick off training jobs with a git-style interface. You never have to SSH into an instance again.

Why TensorPool?

FEATURES

High Performance
Storage

We've devoted a ton of effort into engineering fast storage solutions. For example, our NFS is 7x faster than AWS EBS and our NVMe is twice as fast! See more storage performance stats in our FAQ here!

Git style CLI

While we offer standard SSH access, we also have developed a new git-style training paradigm. With this offering, you can push jobs to a cluster and receive your weights when its done. This way, you never have to pay for idle time.

Reliability

Prior to delivering our GPUs, TensorPool conducts an extensive burn in process to ensure GPU reliability.

Capacity

TensorPool give you an on-demand experience for clusters. We do this by managing reservations ourselves, and binpacking people onto large clusters. This pushes the price advantages of long reservations onto our customers. We also partner with our cloud compute providers (Voltage Park, DataCrunch, Lambda Labs, GCP, etc.) to push our prices further down.

Frequently Asked
Questions