Where should I run my model or workflow?

On your own card if it fits, and the checkers tell you in seconds. For a ComfyUI workflow whose nodes Comfy Cloud has, Comfy Cloud bills only run time. For any custom node or a setup of your own, rent a GPU by the hour from RunPod or another cloud. For catalogue models such as FLUX or Kling, an API that prices each image or video is the simplest.

How do you make money?

From labelled referral links: when you sign up with a provider through one, the provider may pay us. A referral link is always marked as one, and it never changes a verdict. The cheapest or best option is shown whether or not it pays us.

Where do the numbers come from?

Three places, always labelled. Measured: we rent GPUs and run the jobs, and say when and on which card. Computed: memory from published file sizes and the model's own config, with the formula shown. Read: prices from each provider's own page, with the date we read them.

How accurate is the memory check?

For LLMs on the cards we measured, our estimate was within 1% of the real peak. For image and video workflows we add each stage of the workflow, and the memory a generation works in comes on top; we show how much it measured on the jobs we ran. A verdict is an estimate until you run the job.

Is my workflow uploaded?

No. The checkers run in your browser. A shared link carries the file names and any sizes you typed, never the workflow.

Can my AI agent use this?

Yes. The free MCP server at mcp.nodegrove.io answers whether a model fits a GPU, with the same maths as the site, and the open dataset and the cloud GPU price feed are free to use with credit.

Will it run? Where? For how much?

Drop a ComfyUI workflow or pick a template. Three answers, each from a measurement, a stated formula or a provider's own price.

Workflow

Or drop your own .json or PNG

Your GPU

The model files of ComfyUI's Wan 2.2 Image to video, A14B template: 4 files, LoRAs not counted

Will it run?

On your RTX 4090

Yes

Loads as it is.

Largest stage: 14.5 GB
Your card: 24.0 GB
Left for the work: 9.5 GB
Download: 35.6 GB

Stages: text encoders 6.7 GB · diffusion 14.5 GB. Working memory for generation comes on top and grows with resolution and frame count.

Computed 16 October 2026

Where?

Places that run it

Comfy CloudYes

ComfyUI's template for it uses built-in nodes only. The largest stage, 14.5 GB, fits its 96 GB RTX PRO 6000. Plans from $20 a month; it bills only while a workflow runs.

Read 16 October 2026

Go to Comfy Cloud

A GPU by the hourYes

Any node and your own setup. An RTX 3090 holds it, from $0.17 an hour on SaladCloud, or $0.50 on RunPod, which runs its own data centres. It bills while it idles.

Read 19 October 2026

13 providers compared

A per-clip APIModel only

RunPod sells Wan 2.2 by the clip, $0.30 for 5 seconds at 1280×720: its model and settings, not your workflow.

Read 9 October 2026

APIs compared

For how much?

One output, at each source’s own settings

Comfy Cloud11 credits for a 5-second clip at 640×640, Comfy's own figure
$0.05Read 16 October 2026
RunPod's APIA 5-second clip at 1280×720, on its model, not your workflow
$0.30Read 9 October 2026

Licence: Apache 2.0

Which file to download, in the full checker

Read in your browser: your workflow is never uploaded. File sizes from Hugging Face, read 16 October 2026; Comfy Cloud's node list of 10 July 2026; prices before tax, read 9 to 19 October 2026. NVIDIA cards. Updated 22 October 2026.

Running an LLM instead? Can I run it? checks any open model against your card. Buying a GPU? The build vs rent calculator finds the break-even month.

The dearer card was cheaper per output.

Three jobs we ran on rented cards in Finland on 20 October 2026. The RTX PRO 6000 costs 1.8 times as much an hour as the RTX 6000 Ada, and finished each job 2.7 to 3.5 times sooner, so each output cost 35 to 50% less.

All results, and how we measure

Z-Image Turbo, 4 images at 1024×1024

RTX PRO 6000 96 GB, $2.42/h: 6.5 s · $0.004
RTX 6000 Ada 48 GB, $1.37/h: 17.7 s · $0.007

2.7 times faster, and 35% less for 4 images.

Wan 2.2 5B, a 5-second 720p video

RTX PRO 6000 96 GB, $2.42/h: 74 s · $0.05
RTX 6000 Ada 48 GB, $1.37/h: 200 s · $0.076

2.7 times faster, and 35% less for one video.

A 70B LLM (Q4), 32 answers, 8,000 tokens out

RTX PRO 6000 96 GB, $2.42/h: 49 s · $0.033
RTX 6000 Ada 48 GB, $1.37/h: 172 s · $0.065

3.5 times faster, and 50% less for 32 answers.

Measured 20 October 2026Time from submitting the job to its last output. Cost is that time at each card's on-demand list price before VAT.

Four places to run it.

Where to run, compared

Your own GPU $0.09 an hour in power An RTX 4090 at its full 450 W, at $0.20 a kWh. No meter, and your files stay at home. Computed
Comfy Cloud $4.54 an hour Of generation only, on a 96 GB RTX PRO 6000. Plans from $20 a month; a curated set of nodes. Read 16 October 2026
A GPU by the hour $0.33 an hour An RTX 4090 on SaladCloud, $0.89 on RunPod, which runs its own data centres. Any node; it bills while it idles. Read 19 October 2026
A per-clip API $0.30 a clip Wan 2.2 at 1280×720, 5 seconds, on RunPod's model API: their model and settings, not your workflow. Read 9 October 2026

On-demand prices before tax, as each provider lists them. Power is the card's full board power, before the rest of the PC.

Models people check.

Questions people ask.

Where should I run my model or workflow?: On your own card if it fits, and the checkers tell you in seconds. For a ComfyUI workflow whose nodes Comfy Cloud has, Comfy Cloud bills only run time. For any custom node or a setup of your own, rent a GPU by the hour from RunPod or another cloud. For catalogue models such as FLUX or Kling, an API that prices each image or video is the simplest.
How do you make money?: From labelled referral links: when you sign up with a provider through one, the provider may pay us. A referral link is always marked as one, and it never changes a verdict. The cheapest or best option is shown whether or not it pays us.
Where do the numbers come from?: Three places, always labelled. Measured: we rent GPUs and run the jobs, and say when and on which card. Computed: memory from published file sizes and the model's own config, with the formula shown. Read: prices from each provider's own page, with the date we read them.
How accurate is the memory check?: For LLMs on the cards we measured, our estimate was within 1% of the real peak. For image and video workflows we add each stage of the workflow, and the memory a generation works in comes on top; we show how much it measured on the jobs we ran. A verdict is an estimate until you run the job.
Is my workflow uploaded?: No. The checkers run in your browser. A shared link carries the file names and any sizes you typed, never the workflow.
Can my AI agent use this?: Yes. The free MCP server at mcp.nodegrove.io answers whether a model fits a GPU, with the same maths as the site, and the open dataset and the cloud GPU price feed are free to use with credit.