Run Your First Job on a European GPU.

Four short steps, about five minutes of your time. In our runs the GPU was ready in under two minutes. You'll need an SSH key, and Python 3.11 or later for pip install, on macOS or Linux.

~5 minread & runPython 3.11+macOS or LinuxSSHauthSDK v0.5

Register Your SSH Key

Mimiry uses an SSH key to sign you in. There's no password and no API token to paste. If you don't already have a key for Mimiry, generate one:

BASH

ssh-keygen -t ed25519 -f ~/.ssh/mimiry -C "mimiry"

Platform managerplatform.mimiry.com · Add SSH Key Manually

Install and Sign In

BASH

pip install mimiry

This brings in the mimiry command and a mimiry Python module. It needs Python 3.11 or later, on macOS or Linux. Then sign in once:

BASH

mimiry auth --host platform --key ~/.ssh/mimiry

↳ expected output

OUTPUT

Signed in to https://platform.mimiry.com as you@example.com
Key SHA256:…
Token valid until 2026-11-09 14:51:46

The command remembers the host and the key, so nothing after this needs them. Check your balance:

BASH

mimiry balance

↳ expected output

OUTPUT

Account          user 25f60592-…
Balance          20.00 EUR
Available        20.00 EUR (0.00 reserved)
Active sessions  0

Then see which GPUs are free right now:

BASH

mimiry availability

↳ expected output

OUTPUT

MODEL          VRAM    FROM        GPUS  VCPUS  RAM
H100_80G_SXM   80 GB   … EUR/h     1     16     200 GB
H200_141G_SXM  141 GB  … EUR/h     1     16     200 GB
L40S_48G_PCIe  48 GB   … EUR/h     1     16     96 GB

Each row is a machine you can start, with its hourly rate (left out here: rates move with the catalog). Stock changes through the day, sometimes within minutes, so check just before you start one.

Run Your First GPU Job

One command starts a machine, runs your command in a container, and shuts the machine down when the command ends. Use a MODEL from the availability list:

BASH

mimiry session create docker.io/nvidia/cuda:12.4.1-base-ubuntu22.04 \
  --name hello-gpu --gpu H100_80G_SXM \
  --command "nvidia-smi --query-gpu=name,memory.total --format=csv,noheader" \
  --wait

--wait reports progress while the machine comes up and returns when the job is done. In our runs that took under two minutes, most of it creating the machine:

OUTPUT

Session 4954a694-… submitted.

submitted (provisioning)                 5s
provisioned (provisioning)               57s
setting_up (starting__setting_up)        1m30s
pulling_image (starting__pulling_image)  1m35s

Your command's output is in the session's log:

BASH

mimiry session logs 4954a694-…

↳ expected output

OUTPUT

… [session]    pulling_image
… [container]  container_exited
… [container]  NVIDIA H100 80GB HBM3, 81559 MiB
… [session]    exited
… [session]    terminated

mimiry session status 4954a694-… shows the exit code and what the job cost. You pay for the time the machine existed, not for the whole hour.

Get a GPU You Can SSH Into

For interactive work, start a machine that stays up. sleep infinity keeps the container running, and --auto-terminate never keeps the machine:

BASH

mimiry session create docker.io/nvidia/cuda:12.4.1-base-ubuntu22.04 \
  --name my-gpu --gpu H100_80G_SXM \
  --command "sleep infinity" --auto-terminate never --wait

↳ expected output

OUTPUT

ID       4954a694-…
Name     my-gpu
State    running
GPU      H100_80G_SXM × 1
SSH      ssh -p 22 root@203.0.113.10

Use the SSH line it prints, with your key:

BASH

ssh -i ~/.ssh/mimiry -p 22 root@203.0.113.10 nvidia-smi

This machine runs, and costs, until you end it. When you are done: mimiry session terminate 4954a694-…

Success

That's It. Welcome to Mimiry.

That was a European GPU, from your terminal, in one command.

Where to go next

Three small things to try once your first job works.

Pick a different GPU

mimiry availability --min-vram 80 narrows the list. Pass any MODEL it shows to --gpu; a comma-separated list means “any of these, in this order”.

Script it from Python

The same install gives you a Python module that does what the command does: c.sessions.create(…), wait_ready, logs, terminate.

See what you spent

mimiry transactions lists every charge, one line per session, with your balance after each.

The same from Python

PYTHONhello.py

import mimiry

c = mimiry.Client()          # uses the host and key `mimiry auth` saved
print(c.account.balance().available)
for s in c.sessions.list(all=True):
    print(s.name, s.state)