Compute
NVMe VMs on AMD EPYC, billed by the hour.
Two families today. t1a is burstable and cheap for dev boxes and runners; m1a is dedicated for web apps, APIs, and anything steady-state. 50 Gbps networking either way. 2 to 64 vCPUs.
$ exc compute create --name web-1 --instance_type t1a.micro --wait
Burstable — t1a
Cheap boxes for dev, CI, and small APIs.
All t1a instances run on burstable AMD EPYC CPUs with 50 Gbps of networking. Pick one when the workload is intermittent.
| Instance | vCPU | Memory | Disk | Rate |
|---|---|---|---|---|
| t1a.micro | 2 | 1 GiB | NVMe EBS | ₹0.236 /hr |
| t1a.small | 2 | 2 GiB | NVMe EBS | ₹0.472 /hr |
| t1a.medium | 2 | 4 GiB | NVMe EBS | ₹0.945 /hr |
Dedicated — m1a
Dedicated cores for steady-state workloads.
All m1a instances run on dedicated AMD EPYC CPUs with 50 Gbps networking. Each step doubles vCPU and RAM.
| Instance | vCPU | Memory | Disk | Rate |
|---|---|---|---|---|
| m1a.large | 2 | 8 GiB | NVMe EBS | ₹1.889 /hr |
| m1a.xlarge | 4 | 16 GiB | NVMe EBS | ₹3.778 /hr |
| m1a.2xlarge | 8 | 32 GiB | NVMe EBS | ₹7.556 /hr |
| m1a.4xlarge | 16 | 64 GiB | NVMe EBS | ₹15.112 /hr |
| m1a.8xlarge | 32 | 128 GiB | NVMe EBS | ₹30.224 /hr |
| m1a.16xlarge | 64 | 256 GiB | NVMe EBS | ₹60.448 /hr |
The CLI
From install to a running VM in three commands.
Every operation on the console is also a subcommand of exc compute. List what's available, create a VM, fetch its public IP — and you have a server.
- Resize in place. exc compute resize --vm_id ... --instance_type ...
- Stop the meter. exc compute stop — the bill stops with it.
- Serial console. Read the console when SSH isn't an option.
$ exc compute instancetype list NAME vCPU RAM RATE t1a.micro 2 1 GiB ₹0.236/hr t1a.small 2 2 GiB ₹0.472/hr m1a.large 2 8 GiB ₹1.889/hr m1a.16xlarge 64 256 GiB ₹60.448/hr $ exc compute create \ --name web-1 \ --instance_type m1a.large \ --image_id 1 --subnet_id 1 --wait ✓ instance web-1 running web-1 · 2 vCPU · 8 GiB · mum-1a
One honest limitation
One region today.
Compute runs from mum / mum-1a only. Cross-zone HA is on the roadmap; for now, most production setups put a small HAProxy or Caddy VM behind a reserved public IPv4 and fail over by re-associating the reservation.
One more thing
Stopped VMs don't bill.
Stopping a VM stops the meter. We don't charge a "parked instance" fee the way some clouds do. The full SLA — including a refund trigger after six minutes of downtime in any hour — is published in writing.