Catalog
Pods across consumer and datacenter GPUs, plus serverless. We list on-demand hours first. Community and Secure are different rows in our heads even when the GPU name matches.
Provider review / verified 2026-09-05
RunPodGPU review
RunPod sells GPU pods and serverless workers. Community Cloud is other people’s machines with a UI. Secure Cloud is RunPod-operated datacenter capacity. Mixing the two in your head is how you get surprised.
Models listed10
From /GPU/hr$0.39
Regions in sample1
Business modelManaged GPU cloud
Our reading
Who this provider is for
If I could keep one GPU cloud for week-to-week LLM work, it would be RunPod. Not because it is the cheapest, Vast.ai often wins the screenshot, but because I can go from a 4090 pod to an H100 Secure pod to a serverless worker without changing religion. Community is still a marketplace. Secure is the product I trust with a job I do not want to explain to a founder. Serverless is the reason I stop paying for idle on bursty inference.
Rent it if Solo builders and small teams who want SSH or a pod template today, and maybe serverless tomorrow, without a hyperscaler account manager.
Skip it if Skip Community if you need an SLA. Skip the whole brand if you need InfiniBand clusters and a reserved H100 fleet with a contract, that conversation is Lambda, CoreWeave, or a hyperscaler.
How we compare hoursStrengths
- Clear split once you read it: Community vs Secure vs Serverless
- Strong on-demand catalog from RTX 4090 up through H100, with per-second billing
- Serverless workers that actually scale to zero, rare among cheap GPU clouds
- Templates and a UI that do not punish people who are not SREs
Limits
- Community Cloud is not Secure Cloud. Host risk still exists on the cheap side
- H100 Secure is not trying to beat Vast.ai’s floor. You pay for the operator
- Networking and multi-node training are not why this company exists
- Support quality tracks the tier. Do not expect enterprise paper on a $0.34 4090
Same standard, every cloud
What we measured on RunPod
Four checks we apply to every provider. This is not a score, and it is not a paid ranking.
Price signal
Community undercuts Secure and often sits near Vast.ai. Secure H100 lives in managed-cloud territory. Serverless is a different unit, compare per-second active time, not the pod hour.
Product shape
A pod is a VM with a GPU. Serverless is a worker. FlashBoot and network volumes are how you stop treating every start as a blank disk. Use them.
Still verify
Check which cloud the listing sits on. Confirm stock. If you came for serverless, confirm the worker GPU and cold-start, not the pod price.
- Model
- Managed GPU cloud
- Headquarters
- Dover, US
- Setup
- Pods, SSH, templates, serverless workers
- Billing grain
- Per second
- Tiers
- Community · Secure · Serverless
- On-demand
- Yes
- Scale to zero
- Serverless workers
- Best known for
- Pods + cheap H100 without a quote
Catalog listings
RunPod catalog sample
Published on-demand hours first. Spot and waitlists are labeled. Confirm the live rate before you provision.
GPUsRegionBilling typeInterconnect/GPU/hr
Community is not Secure. I will keep saying it.
RunPod’s marketing puts them in one gradient. Operationally they are two products. Community is closer to Vast.ai with a better coat of paint. Secure is RunPod taking the host risk. If your Discord horror story starts with a pod that vanished, ask which cloud it was on before you write off the brand.
I develop on Community when I am iterating. I move the training that matters to Secure. The GPU name can stay H100. The product does not.
Serverless is the actual differentiator
Most cheap GPU clouds sell you a machine that idles at full price while your endpoint waits. RunPod’s workers can sit at zero. For spiky inference, a demo, a copilot, a batch of images, that is the hour that matters, not the H100 poster.
Cold starts are real. If your P99 cannot eat them, keep a min-worker warm and stop pretending you are at zero. Still cheaper than a forgotten pod.
RunPod vs Vast.ai vs Lambda
Vast.ai if I can checkpoint and I want the floor. RunPod if I want to ship this week. Lambda if I am training for days on a cluster I do not want to think about, or I already live in their stack.
If you are comparing a Community H100 to a Lambda on-demand H100, you are comparing a market to a cloud. Do it on purpose or do not do it.
What I wish they told you on the pricing page
Storage and network egress are where surprise lives after the GPU looks cheap. Network volumes are not free. A pod you leave up over the weekend is still a pod. Serverless that never scales down because you set min workers to one is a pod with extra YAML.
Read the GPU page, then the cloud column, then the storage line. Then rent.
How I actually rent it
- 01
Choose the cloud, then the GPU
Community for experiments. Secure for jobs you cannot shrug off. Serverless for bursty inference.
- 02
Use a template you trust
CUDA and PyTorch versions are how weekends die. Pin them.
- 03
Turn idle off
Stop the pod. Or use workers. An on-demand hour you forgot is the most expensive hour.
FAQ
RunPod questions
Is RunPod good for H100?+
Yes, especially Secure when you want on-demand without a quote. Community H100 can be cheaper and flakier. Compare both as separate products.
RunPod vs Vast.ai, which should I pick?+
Vast.ai for the lowest hour if you checkpoint. RunPod if you want a product (pods, Secure, serverless) instead of a host ad. I use both.
Does RunPod offer serverless GPUs?+
Yes. Workers scale down when idle. Confirm GPU type, flashboot, and min workers. It is not the same SKU as a pod.
Is RunPod Community Cloud safe?+
Safer than a random unverified Vast.ai host in my experience, still not an SLA. Do not put payroll on it.
Who should skip RunPod?+
Teams buying InfiniBand islands and reserved Blackwell clusters with legal paper. That is CoreWeave, Lambda, or a hyperscaler conversation.
Sources
Where these numbers come from
Prices and SKUs are from the GPUBeacon catalog sample. Ranking rules are on the methodology page. Outbound links may be affiliate. They never change the hour we display.
GPUs
Keep comparing
Other providers to compare
Same fields, different product. Compare the GPU hour before you lock a brand.
