Comparable performance, at a fraction of the AWS bill.
AWS is a great place to run almost everything except a big GPU bill. Surfra lets you move GPU and batch work onto cheaper clouds like Salad, RunPod, and Vast.ai, and keep AWS as a fallback if you want it.
Where each one fits.
| AWS on its own | Surfra on your clouds | |
|---|---|---|
| Where GPUs come from | AWS regions and instance types | Any cloud you connect: Salad, RunPod, Vast.ai, DigitalOcean, Yotta, your own machines, and AWS in beta |
| Who bills GPU time | AWS | Each cloud you use, directly. Surfra adds no markup |
| Scaling to zero | You build it with autoscaling groups or a queue service | Built in. Idle replicas stop after your timeout |
| When a GPU type is out of stock | You pick another region or type yourself | The next row in your priority list takes the replica |
| Logs for one request | You wire up logging across instances | Every line carries a timestamp and the request it belongs to |
| Best for | Workloads deep in the AWS ecosystem | GPU inference, scraping, and batch work where cost matters |
What Hirebase saw.
Hirebase reads company career pages to find jobs, and it needed to go from tens of millions of pages a month to over a billion. A couple of VMs could not keep up, and an in-house orchestrator kept getting stuck replicas with no way to see why.
They moved the workload off AWS onto Salad through Surfra and got comparable performance for far less. By their own estimate, running today's volume on AWS would cost tens of thousands of dollars a month.
You do not have to leave AWS.
Surfra's AWS connector is in beta, so AWS can stay in your list as a row. Put a GPU marketplace first for price and AWS last for the times you need it. Your app keeps one name and one queue either way.
Looking at other options?
Common questions
Is Salad really cheaper than AWS for GPUs?
For workloads that tolerate a node going away, it usually is, because Salad runs on consumer GPUs. Hirebase estimates its current volume would cost tens of thousands of dollars a month on AWS. Check current prices on each provider for your own GPU and region.
Do I lose reliability by leaving AWS?
You trade a single provider for several. Surfra retries a request when its worker fails and moves new replicas to the next cloud when one is full.
Can I keep using AWS with Surfra?
Yes. The AWS connector is in beta and AWS can sit in the same priority list as the cheaper clouds.
Move GPU work off the AWS bill.
Surfra is invite only while we test it. Tell us what you want to run and we will send an invite.
Request access