Skip to content
Launch imminent

Official date to be announced shortlyGPU offers available at launch; figures shown are indicative.

Priority access: sign up

What is InfiniBand? Networking for GPU clusters

InfiniBand is the low-latency network that links GPU servers in training clusters. Learn how it compares with NVLink inside a server.

Glossary3 sections1 min read

What is InfiniBand?

InfiniBand is a high-bandwidth, low-latency network fabric used to connect servers in high-performance computing and multi-node GPU training clusters. It supports RDMA, so network cards can move data directly between machines' memory, including GPU memory with GPUDirect RDMA.

Why it matters when renting a GPU

Inside one server, GPUs talk over NVLink; between servers, they talk over the network. Multi-node training needs a fast fabric such as InfiniBand.

For single-node jobs, the network between servers matters little; for multi-node training, ask about the fabric and its speed per GPU.

Worked example

A 400 Gb/s InfiniBand port carries 400 ÷ 8 = 50 GB/s per direction, while the H100 SXM lists NVLink 4 (900 GB/s) between GPUs inside a server: that gap is why models are split to keep the heaviest traffic within a node.

Frequently asked questions

Do I need InfiniBand for one GPU?
No. InfiniBand matters when a job spans several servers.

Be ready on launch day

Create your account now: you can top up your balance and launch a GPU as soon as the market opens.