AWS Trainium2
AWS Trainium2 is AWS’s third-generation machine-learning chip, used in Amazon EC2 Trn2 instances.
- Vendor
- AWS
- Architecture
- NeuronCore-v3 (8 cores)
- Memory
- 96 GiB HBM
- Memory bandwidth (TB/s)
- 2.9 TB/s
- FP8 (TFLOPS)
- 1,299 TFLOPS
- Form factor
- Cloud instances (EC2 Trn2, Trn2 UltraServers)
- Status
- Shipping
Trainium2 has eight NeuronCore-v3 cores and 96 GiB of device memory at 2.9 TB/s. It powers Amazon EC2 Trn2 instances and Trn2 UltraServers, and NeuronLink-v3 provides 1.28 TB/s of chip-to-chip bandwidth with memory pooling across up to 64 chips. AWS rates it at 1,299 TFLOPS FP8 dense, 6.7 times the first Trainium.
- Memory (GB)
- 96 GB
- Precision note (sparse or dense)
- 1,299 TFLOPS FP8 is AWS's dense figure; AWS lists 2,563 TFLOPS for FP8/FP16/BF16/TF32 with sparsity.
- Vendor source
- awsdocs-neuron.readthedocs-hosted.com
- Checked
Questions
How much memory does AWS Trainium2 have?
Trainium2 has 96 GiB of device memory with 2.9 TB/s of bandwidth.
What is Trainium2’s FP8 performance?
AWS rates Trainium2 at 1,299 TFLOPS FP8 dense, and lists 2,563 TFLOPS for FP8/FP16/BF16/TF32 with sparsity.
How many Trainium2 chips can be connected?
NeuronLink-v3 provides 1.28 TB/s of chip-to-chip bandwidth with memory pooling across up to 64 chips.
Source
Every figure is taken from the vendor’s own datasheet, product page, technical documentation or announcement, linked on each chip with the date it was last checked. How we pick and check sources: editorial standards.