Built for nodes that come and go
Salad machines are other people's computers, so a node can disappear. A request that was running on it goes back to the queue and is retried on another worker.
Salad on Surfra
SaladCloud GPUs, with a queue and a fallback.
Salad runs containers on a huge network of consumer GPUs, which is why it is so cheap. Surfra puts a queue in front of it, scales your Salad container groups with demand, and sends work to another cloud when Salad has no room.
Connect
What you need from Salad.
Add your Salad account under Credentials. Surfra checks the key with Salad before saving it, then encrypts it with your team's key. It is never shown again.
| Detail | Salad |
|---|---|
| You provide | API key, Organization name, Project name |
| Who bills you | Salad, directly, at its own price |
| Surfra markup | None |
| CPU machines | Supported |
GPUs
Salad GPUs you can rank.
Every GPU class Salad lists, 59 in all, from GTX 1650s to RTX 5090s, including laptop cards and AMD Radeon. Salad charges by priority: batch is the lowest rung and high is the top. Prices are Salad's own, per GPU-hour, on 16 Oct 2026. Your Surfra row picks the class, and Salad picks the node.
| GPU class | VRAM | Batch to high, per hour |
|---|---|---|
| RTX 5090 | 32 GB | $0.25 – $0.50 |
| RTX 4090 | 24 GB | $0.16 – $0.33 |
| RTX 5080 | 16 GB | $0.15 – $0.26 |
| RTX A5000 | 24 GB | $0.09 – $0.25 |
| RTX 5090 Laptop | 24 GB | $0.10 – $0.235 |
| RTX 4080 | 16 GB | $0.11 – $0.23 |
| RTX 5070 Ti | 16 GB | $0.10 – $0.225 |
| RTX 4070 Ti Super | 16 GB | $0.09 – $0.21 |
| RTX 4070 Ti | 12 GB | $0.08 – $0.20 |
| AMD RX 9070 XT | 16 GB | $0.07 – $0.196 |
| RTX 3090 Ti | 24 GB | $0.10 – $0.19 |
| RTX 5080 Laptop | 16 GB | $0.08 – $0.175 |
| RTX 3090 | 24 GB | $0.09 – $0.17 |
| AMD RX 7900 XTX | 24 GB | $0.066 – $0.168 |
| RTX 4090 Laptop | 16 GB | $0.075 – $0.165 |
| RTX 5070 | 12 GB | $0.08 – $0.165 |
| AMD RX 7900 XT | 20 GB | $0.054 – $0.156 |
| RTX 3080 Ti | 12 GB | $0.08 – $0.155 |
| RTX 4070 | 12 GB | $0.07 – $0.15 |
| RTX 5060 Ti | 16 GB | $0.07 – $0.15 |
| RTX 3080 | 10 GB | $0.06 – $0.14 |
| RTX 4060 Ti | 16 GB | $0.08 – $0.14 |
| RTX 4080 Laptop | 12 GB | $0.065 – $0.14 |
| AMD RX 7800 XT | 16 GB | $0.042 – $0.132 |
| RTX 5060 Ti 8GB | 8 GB | $0.068 – $0.13 |
| RTX 5070 Ti Laptop | 12 GB | $0.06 – $0.13 |
| AMD RX 9060 XT | 16 GB | $0.049 – $0.126 |
| RTX 4060 Ti 8GB | 8 GB | $0.065 – $0.11 |
| RTX 5060 | 8 GB | $0.065 – $0.11 |
| RTX 5070 Laptop | 8 GB | $0.05 – $0.105 |
| RTX 2080 Ti | 11 GB | $0.06 – $0.10 |
| RTX 3070 | 8 GB | $0.04 – $0.10 |
| RTX 3070 Ti | 8 GB | $0.06 – $0.10 |
| RTX 3080 Ti Laptop | 16 GB | $0.045 – $0.10 |
| RTX 2080 | 8 GB | $0.05 – $0.08 |
| RTX 3060 | 12 GB | $0.04 – $0.08 |
| RTX 3060 8GB | 8 GB | $0.03 – $0.08 |
| RTX 3060 Ti | 8 GB | $0.03 – $0.08 |
| RTX 3080 Laptop | 8 GB | $0.04 – $0.08 |
| RTX 4060 | 8 GB | $0.05 – $0.08 |
| RTX 4070 Laptop | 8 GB | $0.05 – $0.08 |
| RTX 5060 Laptop | 8 GB | $0.036 – $0.08 |
| RTX 3050 | 8 GB | $0.03 – $0.07 |
| RTX 4060 Laptop | 8 GB | $0.035 – $0.07 |
| RTX 5050 | 8 GB | $0.032 – $0.07 |
| RTX 5050 Laptop | 8 GB | $0.029 – $0.065 |
| RTX 2070 | 8 GB | $0.02 – $0.06 |
| RTX 2060 | 6 GB | $0.02 – $0.05 |
| RTX 3060 Laptop | 6 GB | $0.023 – $0.05 |
| RTX 4050 Laptop | 6 GB | $0.02 – $0.045 |
| RTX 3050 Laptop 6GB | 6 GB | $0.02 – $0.042 |
| GTX 1070 / 1080 / 1080 Ti | 8 GB | $0.02 – $0.04 |
| GTX 1660 | 6 GB | $0.02 – $0.04 |
| GTX 1660 Super | 6 GB | $0.02 – $0.04 |
| RTX 3050 Ti Laptop | 4 GB | $0.016 – $0.035 |
| GTX 1060 | 6 GB | $0.02 – $0.03 |
| RTX 3050 Laptop | 4 GB | $0.016 – $0.03 |
| GTX 1050 Ti | 4 GB | $0.015 – $0.02 |
| GTX 1650 | 4 GB | $0.015 – $0.02 |
How it runs
What Surfra adds on top of Salad.
Cheap base, reliable burst
Keep a steady pool on Salad and give a second cloud a lower priority row. When Salad is short on a card, replicas start there instead.
The Hirebase setup
Hirebase moved its workload off AWS onto Salad through Surfra and got comparable performance for a fraction of the cost.
Other clouds
Pair it with a second cloud.
FAQ
Common questions
Do I need my own Salad account?
Yes. Surfra starts container groups in your Salad organization and project, and Salad bills you directly.
What happens when a Salad node goes away?
The replica is replaced, and any request that was running on it goes back to the queue and is retried.
Which Salad GPUs can Surfra use?
Every single-card class Salad lists: GeForce RTX 20, 30, 40, and 50 series, their laptop versions, GTX 10 and 16 series, RTX A5000, and AMD Radeon RX 7000 and 9000 cards. AMD cards need an image built for ROCm.
Autoscale Salad from your own account.
Surfra is invite only while we test it. Tell us what you want to run and we will send an invite.
Request access