trn1.2xlargeAWS Pricing, Specs & Benchmarks

The trn1 family is a generation-1 AWS Trainium ML training line running on Intel Xeon (x86_64). trn1.2xlarge provides 8 vCPUs and 32GB of RAM (a 1:4 vCPU-to-memory ratio), priced at $1.3438/hr On-Demand — about $967.50 per month at 24×7 utilization. In Sysbench CPU benchmarks it scores 2981 single-core and 12530 multi-core.

AWS Trainium ML training instances are sized for large-scale ML model training and fine-tuning using the Neuron SDK as a cheaper alternative to NVIDIA training. If your workload doesn't match that profile — for example, if you're running a memory-bound app on a compute-optimized instance, or a CPU-pegged batch job on a burstable t-series — you'll likely pay for capacity you can't use or run into bottlenecks you didn't anticipate. The pages for sibling families (linked further down) cover those alternatives in detail.

In the AWS instance-naming convention, the digit after the series letter is the generation — higher is newer. This appears to be one of AWS's newer generations in the series, so for most fresh workloads it's a reasonable default within its category.

Pricing varies by region (sometimes meaningfully — the gap between the cheapest and most expensive AWS region for the same SKU is often 15–30%). The regional pricing table further down shows where trn1.2xlarge is cheapest today, both On-Demand and Spot. Spot pricing for trn1.2xlarge is currently $0.3384/hr — roughly 75% off On-Demand — which makes the instance viable for fault-tolerant batch jobs, CI fleets, and any workload that can tolerate interruption with short notice.