Docker Docker Execution

Any container.
Any GPU.

Full environment control. Per-second billing. Your containers, running on the exact hardware you need.

Terminal
$ lyceum docker run myregistry/train:v2 \
--machine gpu.h100.8x \
--volume ./data:/data
Pulling myregistry/train:v2...
Image size: 4.2 GB
Provisioning 8x H100 cluster...
Mounting volumes...
Container running on 8x H100
[rank 0] Initializing distributed training...
[rank 0] Model loaded: 70B parameters
[rank 0] Starting epoch 1/100...

Full control. Here's why.

Any Docker image

If it runs in Docker, it runs on Lyceum. No modifications needed.

Per-second billing

Pay only for what you use. A 4-minute job costs 4 minutes.

Instant availability

GPUs provision in seconds. No waiting in queues.

Docker
train:v2
4.2 GB image
Your container active
CUDA 12.1
8× H100 640 GB
Running
$24.80/hr

Available GPUs

T4 to B200. Multi-GPU and multi-node available.

GPU VRAM Price/hr
NVIDIA B200
192 GB $5.89
NVIDIA H200
141 GB $3.69
NVIDIA H100 Popular
80 GB $3.29
NVIDIA A100
80 GB $1.99
NVIDIA L40S
48 GB $1.49

Want something simpler?

For rapid prototyping with just Python scripts, try Python execution. Zero config, automatic dependency detection, instant iteration.

Explore Python execution

Run your first GPU job.

No credit card required. Request access to get started.