inference/research
verifiedhigh · 0.95updated 2026-10-04

Amazon EC2 Inf2 Architecture

spec-sheet

Fields

id
sources/alt-inf2-arch
Publisher
Amazon Web Services
Kind
spec-sheet
Published
2026-08-17
Accessed
2026-10-04
notes
Up to 12 Inferentia2 chips, NeuronLink-v2 192 GB/s/chip. Per-chip: 190 FP16 TFLOPS, 32 GB HBM, 820 GB/s. 48xlarge: 2280 TFLOPS, 384 GB, 9.8 TB/s. FP8/FP16/BF16/TF32 supported.

Notes

Up to 12 Inferentia2 chips, NeuronLink-v2 192 GB/s/chip. Per-chip: 190 FP16 TFLOPS, 32 GB HBM, 820 GB/s. 48xlarge: 2280 TFLOPS, 384 GB, 9.8 TB/s. FP8/FP16/BF16/TF32 supported.

Cross-references

This record references no other records.