AWS: cheapest 4 vCPU / 32GB RAM and 32GB+ VRAM EC2 picks (us-east-1, Nov 2026)

No GPU (4 vCPU/32GB RAM): r-family xlarge. On-demand x86 r6a.xlarge ~$0.23/hr; ARM r6g.xlarge ~$0.20/hr. Spot was ~60-70% cheaper: r6g.xlarge $0.06-0.09, r5a.xlarge ~$0.09. Ubuntu 24.04 shows ~30Gi in free -h. Root gp3 on Ubuntu AMIs is /dev/sda1 in block-device-mappings; 500GB gp3 is ~$40/mo. AMI via SSM: /aws/service/canonical/ubuntu/server/24.04/stable/current/amd64/hvm/ebs-gp3/ami-id.

GPU with >=32GB on one card: g6e.xlarge (L40S 48GB, 4 vCPU) ~$1.86/hr on-demand, and spot was only ~10% cheaper ($1.61-1.79). A10G/L4 are 24GB, too small. New accounts have 0 for both G/VT quotas (L-DB2E81BA on-demand, L-3819A6DF spot); request via service-quotas before launching. Pending quota requests cannot be cancelled via API.