Use every core, GPU and server you have

Spread loops, parameter sweeps and simulation studies across your hardware. Run on one machine, or scale up to remote servers and clusters with the same code.

Download RunMat
LaptopRemote serverLARGE CLUSTER

Run more at once

Every core on your laptop

A parfor loop runs on all your cores with nothing to install or configure.

Same answer on any number of workers

A loop gives the same result on 1 worker or 64, including loops that use random numbers.

A cluster without a rewrite

Add your own servers and the code you wrote on your laptop runs there as it is.

Simulations too

Parameter sweeps and meshing spread across workers the same way loops do.

Run MATLAB parfor loops across every core

Change for to parfor and the iterations run across your cores. Results come back in iteration order, so the rest of your script doesn't change.

  • If a loop can't run in parallel, RunMat tells you before it starts, with the line and the reason
  • parfor (k = 1:n, 0) runs the same loop in your own session, so you can step through it while debugging
  • An error on a worker comes back with its message and the line it came from
sweep.m
speeds = linspace(10, 60, 200);
peak = zeros(size(speeds));
parfor k = 1:numel(speeds)
peak(k) = max(abs(rideResponse(speeds(k))));
end
plot(speeds, peak)

Run parameter sweeps and simulation studies in parallel

Write a sweep once and its cases run at the same time, on your cores or on a cluster. Results come back in the order you wrote the cases, ready to compare. Meshing a large part spreads across workers on its own.

  • If one case fails, the others still finish and the failed case is listed with its error
  • No manual splitting of geometry for large meshes

Run a parameter sweepExplore simulation

bracket.fea
sweep load_envelope over study.bracket_response {
axis service_load {
half = 0.5 kN
nominal = 1.0 kN
proof = 1.5 kN
}
output peaks = collect(case.output.peak_stress.value)
}
terminal
$ runmat run bracket.fea --sweep load_envelope

Scale to remote servers and clusters

Join workstations and servers you own to a cluster, then run scripts, tests and studies on them. RunMat encrypts your code, data and results for the execution machines.

  • Join with runmat cluster join engineering, or add --service to start the node at boot
  • Servers can run Linux, macOS or Windows
  • A worker that uses a server's GPU has it to itself until its task finishes

Remote execution

your computer
$ runmat cluster create --name engineering
cls_4c1e… engineering active
server
$ runmat login
$ runmat cluster join engineering --service
installed and started RunMat execution-node service
your computer
$ runmat estimate_pi.m --cluster engineering
Job run_2f6d (queued)
run_2f6d succeeded engineering 2026-10-16T18:02:41Z
piEstimate = 3.1414

FAQ

01How many workers does it use?

By default, one per core. parpool(4) starts four, and parfor (k = 1:n, M) limits a single loop to M workers.

02Do results depend on the number of workers?

No. Each iteration gets its own random numbers and results are combined in a fixed order, so a loop gives the same answer on any number of workers.

03Do I need to change my code to run it on a cluster?

No. Join the servers once, then run the same script with runmat analysis.m --cluster engineering.

04Can I run RunMat under Slurm, PBS or LSF?

Yes. Run it as an ordinary job and start a pool on each node you're given. To use several servers as one pool, enroll them as a RunMat cluster.

05Does each worker use the GPU?

Yes. Array math on a worker runs on the GPU the same way it does in a normal session. On a cluster, each worker gets a GPU to itself.

06Does parfor work in the browser?

Yes. In the browser, the workers run as web workers and give the same results.

Run it on every core

Download RunMat
← Back to the runtime