Overview
High-performance GPU instances and servers for model training, inference and rendering.
GPU Compute Server is backed by our team 24/7 and integrates seamlessly with our other services, giving you more stability and security at lower cost.
Request a quotePricing depends on configuration, region and term — your account manager will tailor the best offer.Get a quote
Key features
Multi-GPU interconnect
High-speed interconnect for multi-GPU training.
AI-ready images
Pre-installed CUDA, PyTorch and other frameworks.
Flexible billing
On-demand, monthly or long-term rental.
Specifications
| GPU | Mainstream data-centre GPUs (subject to stock) |
|---|---|
| GPUs per server | 1, 2, 4 or 8 |
| Network | High-speed fabric; RDMA on supported models |
| Environment | CUDA / cuDNN / PyTorch / vLLM |
* Final specifications are confirmed in your quote.
Use cases
LLM training & fine-tuning
Online inference
3D rendering & video processing
How to get started
Tell us your needsOnline or by phone
Proposal & quoteTailored by an architect
Sign & provisionFast delivery
Ongoing support24/7 operations
FAQ
Yes. CPU, memory and bandwidth can be upgraded online or after a restart, and data disks can be expanded online.
On-demand, monthly and annual billing are available; see your quote for details.
Related products
Get started with GPU Compute Server
A dedicated architect will tailor a plan for you