Ready for pilot deployment

ClusterMarshal

ClusterMarshal helps teams choose where, when and how each GPU job runs.

Book a meeting See how it works

See it at SC26, booth 5617, Nov 17–19

From job to placement

  1. 01
    Describe your job

    Training or inference

  2. 02
    Compare 2–4 options

    GPU · Time · Cost

  3. 03
    Choose an option

    Your team makes the choice

  4. 04
    Submit to your scheduler

    Kubernetes first · Slurm to follow

How it works

A clear choice for every GPU job.

Compare the trade-offs before a job runs. ClusterMarshal works with the scheduler your cluster already uses.

  1. 01

    Describe the job

    Provide a training script, or an inference model with a request rate and latency target.

  2. 02

    Compare the options

    Review two to four ways to run it, each with GPUs, start time, run time and cost.

  3. 03

    Choose one

    Select the option that fits your job's requirements and your team's priorities.

  4. 04

    Submit and run

    Once submission is enabled, ClusterMarshal submits the chosen option as you to the scheduler the cluster already runs. If ClusterMarshal does not respond in time, the scheduler places the job as usual.

Pilot deployment

Start in shadow mode.

Compare recommendations with real results before enabling submission.

Observe first

ClusterMarshal recommends alongside your scheduler without affecting placement. Compare predicted run times and costs with what actually ran.

Enable at your pace

Once your operators trust the results, enable submission for a queue or an entire cluster.

Kubernetes first, Slurm to follow

Pilots launch with Kubernetes support. Slurm support follows.

Meet XPerf

See ClusterMarshal at SC26.

Booth 5617 · Nov 17–19, 2026 · Chicago

Meet the team at our booth, shared with Infraeo, and discuss a pilot for your cluster.

Book a meeting

Alex Carter
alex.carter@xperf.ai