Dedicated Training

Full control over your training infrastructure.

Root access, any framework, persistent storage. From single GPUs to thousands.

Request Access Contact for Clusters

training-node-01

$ nvidia-smi --query-gpu=name,memory.used --format=csv

name, memory.used [MiB]

NVIDIA H100 80GB, 72438 MiB

NVIDIA H100 80GB, 71892 MiB

NVIDIA H100 80GB, 73104 MiB

NVIDIA H100 80GB, 70256 MiB

$ torchrun --nproc_per_node=4 train.py

[INFO] Training started on 4x H100 GPUs...

$ ssh root@ gpu-node

Full root access

SSH in, install anything, configure your way. Complete control over your compute environment.

PyTorch

JAX

Any framework

PyTorch, JAX, TensorFlow, custom CUDA - all supported. Use the tools you already know.

ready

0:03

Launch in seconds

Spin up a VM from the CLI in seconds. No waiting, no tickets, no friction.

InfiniBand networking

High-speed interconnects for distributed training. 400Gb/s GPU-to-GPU communication.

Choose your scale

From interactive development to frontier model training, we have the infrastructure for every stage.

Virtual Machines

1–8 GPUs

For fine-tuning, research, and interactive development. Self-service provisioning, ready in seconds.

Self-service provisioning
Ready in seconds
Per-second billing

Launch VM

Documentation

GPU Clusters

8–8,000 GPUs

For pre-training, large-scale fine-tuning, and frontier research we also offer large scale clusters. Dedicated capacity with InfiniBand.

Dedicated infrastructure
InfiniBand networking
Engineering support

Talk to engineering

GPU options

Choose the right GPU for your training workload. Per-hour pricing, no long-term commitments.

GPU	VRAM	Price/hour
NVIDIA B200	192 GB	$4.29
NVIDIA H200	141 GB	$3.19
NVIDIA H100 Popular	80 GB	$2.49
NVIDIA A100	80 GB	$1.39
NVIDIA L40S	48 GB	$1.05
NVIDIA T4	16 GB	$0.39

View full pricing details →

Ready to start training?

Get dedicated GPU capacity for your training workloads.

Launch a VM