What are you running?

Start from a workload or set the minimums yourself. We pick the cheapest instance that meets every one.

Web or API server Stateless, scales out Database Memory and fast disk Cache Redis, Memcached CI and batch jobs Bursty, retry-safe Model inference Serve a 7–13B model Model training Multi-GPU, big VRAM

CPU at least 32 vCPU

1 2 4 8 16 32 64 96 192

1 2 4 8 16 32 64 128 256 512 1T

None 16 GB 24 GB 48 GB 80 GB+ 160 GB+

Any Arm (Graviton) x86 only

Pick x86 if you run Windows or x86-only binaries.

Network (EBS) is fine Fast local NVMe

Not sure A year or more Three years

19 instances meet your needs in N. Virginia, priced on-demand.

inf Inferentia

2 Generation 2

24xlarge 96 vCPU · 384 GiB

$4,738/mo

$6.4906 an hour

96 vCPU EPYC Milan

384 GiB 4 GiB per vCPU

EBS only Local disk

50 Gbps Network

Cheapest match. Meets every requirement, with 64 extra vCPU and 256 GiB extra memory at no cost. Full details