RunPod is the easiest on-ramp to renting a cloud GPU: a clean dashboard, one-click templates for Ollama / ComfyUI / Jupyter, and per-second billing so you only pay while the pod runs. This guide walks through your first pod end to end, and shows when RunPod is the right call versus buying a card or using Vast.ai.
When (and why) to use RunPod
Rent on RunPod instead of buying when:
- You need more VRAM than you own — a 70B model or a fine-tune won't fit your 8–16 GB card, but an A100 80 GB does.
- Your usage is bursty — a few hours of fine-tuning a month, not 8 hours a day. (At sustained daily use, buying wins — check the exact break-even on the local vs cloud cost calculator.)
- You want to try before you buy — run Llama 3.3 70B on a rented RTX 4090 for ~$0.50 before spending $1,600 on one. See what your current card can already handle with the GPU & VRAM checker.
- You want reliability without marketplace lottery — RunPod's Secure Cloud sits between Vast.ai's cheapest-but-variable hosts and Lambda's datacenter pricing.
Need more VRAM than you own? Spin up a cloud GPU big enough for any model in minutes — pay for the hours you use instead of buying a card.
Affiliate links — we may earn a commission if you sign up, at no extra cost to you.
Marketplace pricing — the cheapest per-hour rates for spot/interruptible GPUs.
Rent GPUs on Vast.ai →On-demand pods with a simple UI — good for a quick one-off inference or fine-tune.
Rent GPUs on RunPod →Step 1 — Create an account and add credits
- Sign up at RunPod and add a payment method (credit card; $10 minimum deposit).
- Decide between Community Cloud (cheaper, hosts vary) and Secure Cloud (pricier, datacenter SLA). For a first run, Community Cloud is fine.
Step 2 — Deploy a pod from a template
RunPod's templates save you the container setup:
- Ollama — fastest way to pull and chat with a model.
- text-generation-webui / vLLM — an OpenAI-compatible API endpoint.
- ComfyUI — image/video models, not just LLMs.
Pick a GPU that fits your model's VRAM (the VRAM Requirements Guide has the numbers). A rough map:
| Model | GPU to rent | Ballpark price |
|---|---|---|
| Llama 3.1 8B (Q4) | RTX 4060/4070 | ~$0.15–0.25/hr |
| Llama 3.3 70B (Q4) | RTX 4090 24 GB | ~$0.35–0.70/hr |
| Fine-tune 8B (LoRA) | RTX 3090/4090 | ~$0.40/hr |
| Fine-tune 70B (QLoRA) | A100 80 GB | ~$1.89–2.49/hr |
Step 3 — Connect and run a model
# SSH into your pod (RunPod shows the exact command in the dashboard)
ssh root@<pod-ip> -p <port> -i ~/.ssh/id_rsa
# With the Ollama template, just pull and run:
ollama run llama3.3:70b
# The API is exposed at:
# https://<pod-id>-11434.proxy.runpod.netStep 4 — Serverless for infrequent inference
If you only need a model to answer occasional requests, RunPod Serverless bills per second of compute instead of per hour — no idle cost between calls:
import requests
r = requests.post(
"https://api.runpod.io/v2/<endpoint-id>/runsync",
headers={"Authorization": "Bearer <your-api-key>"},
json={"input": {"prompt": "Explain quantization in one paragraph", "max_tokens": 300}},
)
print(r.json()["output"])Step 5 — Save money (and don't leak data)
- Terminate the pod when you're done — an idle pod still bills. Download results first.
- Use a network volume for data you want to survive between pods (~$0.07/GB/month) instead of re-uploading each time.
- Keep secrets in environment variables, never hardcoded.
- The fine-tune-on-cloud, infer-at-home pattern is the cheapest of all: train a LoRA adapter on a rented A100 (it's small, 100–500 MB), download it, and run inference on your own GPU. Full walkthrough in Fine-Tuning: SFT, LoRA & DPO.
RunPod vs the alternatives
- Cheaper, willing to tolerate variability? Use Vast.ai — a marketplace that's often 30–50% cheaper.
- Need guaranteed datacenter uptime? Lambda and Hyperstack (see the Cloud GPU Rental Guide for the full comparison).
- Using it sustainably every day? Buying probably wins — price a machine with the Build Guide and confirm on the cost calculator.
FAQ
Do I need a credit card? Yes — RunPod requires one plus a $10 minimum deposit.
Is my data private on RunPod? Secure Cloud runs in vetted datacenters; Community Cloud runs on third-party hosts. Don't put sensitive data on Community Cloud — for that, prefer Secure Cloud, Lambda, or running locally.
What's the cheapest way to run a 70B model on RunPod? A Community Cloud RTX 4090 spot pod (~$0.35–0.50/hr). For occasional use that beats buying a $1,600 card outright.
Related Guides
To size a specific model, use the VRAM calculator: the memory it needs at every quantization and context length.