Accelerated Computing (F1)
- What it is
- Uses hardware accelerators (co-processors) to do work more efficiently than software on CPU.
- You pay for the accelerator, so only use it when the workload can actually exploit it.
- F1 specifically
- Field Programmable Gate Arrays (FPGAs) — hardware you reprogram for a custom algorithm.
- For genomics, financial analytics, real-time video processing, custom hardware acceleration.
- The other accelerated families
P/G— NVIDIA GPUs.Pfor ML training,Gfor graphics and inference.Trn— AWS Trainium, purpose-built for training large models.Inf— AWS Inferentia, purpose-built for cheap high-throughput inference.
- Not to be confused with
- Compute Optimized (C5) — more/faster general CPUs, no specialized silicon.
- Amazon SageMaker AI — a managed ML platform that runs on these instances for you.