TensorPool

TensorPool is a model finder tool. TensorPool provides on-demand multi-node GPU clusters controlled through a Git-style CLI, with shared high-throughput storage and job submission built for distributed training. The site lists no published pricing or trial terms.

TensorPool is GPU infrastructure for teams training foundation models, delivered through a command line interface rather than a web console. The pitch is narrow and deliberate: the product handles scheduling, networking and storage, while the user stays focused on the model itself. Clusters are requested on demand at a chosen size, for example eight B200 GPUs across four nodes, and the site states that Infiniband is always 3.2Tb/s on those multi-node setups. The workflow follows a Git-style pattern. Shared storage volumes are created with a single command and mounted to clusters, quoted at 10GB/s reads and 5GB/s writes, which the site claims is faster than local NVMe. Training jobs are then submitted with a push command pointing at a TOML config file, so experiment management resembles pushing commits instead of hand-managing orchestration. Storage is engineered for distributed training throughput and low latency, with aggregate figures of up to 300 GB/s read, 150 GB/s write, 1.5M read IOPS and 750k write IOPS. Normal SSH access is also offered for teams that want direct shell control alongside the CLI. The capacity model matters more than the feature list. TensorPool does not run its own data centres; it schedules and binpacks users onto clusters that TensorPool operates in partnership with compute providers. That is how it offers on-demand multi-node hardware without asking customers to reserve long term. It suits ML engineers and research groups who already script their training runs and find the console-driven experience of larger clouds slow to navigate. Compared with hyperscaler GPU offerings, TensorPool trades breadth of services for a thin, fast path to a cluster; compared with bare rental marketplaces, it adds shared high-speed storage and a job submission layer. The homepage does not publish prices, trial terms, a support policy or a list of third party integrations, so those points need to be confirmed directly before committing a training budget.