CLUSTERCluster · Trn2 instance · representative building block
- Operator / site
- AWS / Anthropic
- Country
- United States
- Profile scope
- Multi-site cluster · representative Trn2 instance
- Data checked
- 14 September 2026
Project Rainier
AWS / Anthropic
CPU Accelerator Interconnect
i
Reading this map
Project Rainier node architecture
Project Rainier is operated by AWS / Anthropic in United States. This profile covers Multi-site cluster · representative Trn2 instance. Specifications are listed below; anything the operator has not published is itemised in the documented-fields list.
Trn2 instance · representative building block
- CPU model
- Intel Xeon Sapphire Rapids · 192 vCPUs per instance
- Accelerator model
- AWS Trainium2
- Physical accelerators per node
- 16
- Host / unified memory
- 2 TB host memory (instance specification)
- Accelerator memory
- 1.5 TB HBM3 across 16 chips
- Accelerator ↔ accelerator link
- NeuronLink
- Local storage
- 4 × 1.92 TB NVMe (instance specification)
Scope and uncertainty
Amazon states Project Rainier features nearly half a million Trainium2 chips across multiple United States data centres, and that an UltraServer combines four Trn2 instances into a 64-chip domain. It publishes no instance or UltraServer count, so none is recorded. A separate 900 MW Crusoe campus announced for Abilene supports Microsoft and is not part of this system; its 2.1 GW and 336 MW-per-building figures belong to that project.
Documented fields: 7 / 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
Trn2 instance · representative building block — 7 / 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