CLUSTERCluster · Hopper deployment · partial profile
- Operator / site
- xAI
- Country
- United States
- Profile scope
- October 2024 deployment snapshot
- Data checked
- 14 September 2026
Colossus
xAI
CPU Accelerator Interconnect
i
Reading this map
Colossus node architecture
Colossus is operated by xAI in United States. This profile covers October 2024 deployment snapshot. Specifications are listed below; anything the operator has not published is itemised in the documented-fields list.
Hopper deployment · partial profile
- System / node model
- Supermicro 4U Universal GPU system (liquid-cooled)
- Accelerator model
- NVIDIA Hopper
- Physical accelerators per node
- 8
- Interface model
- NVIDIA BlueField-3 SuperNIC
- Interface speed
- 400 GbE per link · about 3.6 Tb/s per compute server
- Interface-to-accelerator topology
- A dedicated network adapter per accelerator
- Management / storage networking
- A separate 400 GbE connection per CPU, on a different switch fabric from the accelerator network
- Form factor
- Eight 4U servers per rack (64 GPUs per rack) with a coolant distribution unit; racks arranged in groups of eight (512 GPUs); the initial cluster comprised four 25,000-GPU data halls
Scope and uncertainty
xAI states Colossus reached 200,000 H100 GPUs in a single interconnected cluster. The per-rack and per-node layout comes from a case study hosted by Supermicro but authored by a third party and sponsored by Supermicro, so it is a weaker claim than an operator statement: eight 4U servers of eight H100 per rack, with a dedicated network adapter per GPU. No CPU model is published by xAI or any vendor.
Documented fields: 8 / 25
Every field this catalog checks. Fields listed above have a value; the remainder are shown here. A field marked absent is not publicly available in the sources we could verify, which does not mean the hardware is missing. Fields that cannot apply, such as benchmark results for an unranked system, are excluded from the total.
Cluster — 0 / 3
- Manufacturer
- Ranked core count
- Theoretical peak (PFlop/s)
- Measured power (kW)
- Energy efficiency (GFlop/s per watt)
- Operating system
- Network topology
Hopper deployment · partial profile — 8 / 22
- System / node model
- Compute nodes of this type
- CPU model
- CPU sockets per node
- Cores per socket
- Physical CPU cores
- CPU note
- Accelerator model
- Physical accelerators per node
- Host / unified memory
- Accelerator memory
- Accelerator ↔ accelerator link
- On-node fabric
- CPU ↔ accelerator link
- Network interfaces per node
- Interface model
- Interface speed
- Interface-to-accelerator topology
- Aggregate scale-out per node
- GPUDirect RDMA
- Management / storage networking
- Local storage
- NVMe data disks
- Max remote data disks
- Uncached disk IOPS
- Uncached disk throughput (MBps)
- Form factor