GPU guide
GPU for Stable Diffusion and ComfyUI: generate images on a rented server
Updated on
Short answer
GPU Cloud installs ComfyUI in one click on an NVIDIA RTX A6000 with 48 GB, at CA$0.67/h, taxes extra. With that memory, SDXL, Stable Diffusion 3.5 and FLUX.1 [dev] run in 16-bit, without lowering the model’s precision to make it fit.
Why rent a GPU to generate images
An 8 or 12 GB gaming card is enough for Stable Diffusion 1.5, but it quickly limits batch size, resolution and recent models. FLUX.1 [dev] alone weighs 24 GB in 16-bit: on a consumer card, you have to compress it or offload part of it, which slows every image down.
A server rented by the hour gives you a data center GPU for the length of a project: a poster, a series of variations for a client, a test of a video model. You pay for the hours you use, then you delete the server.
The memory common models take
| Model | Parameters | Weights in 16-bit | Fits in 48 GB |
|---|---|---|---|
| Stable Diffusion 1.5 | 1 billion | 2 GB | yes |
| SDXL | 3.5 billion | 7 GB | yes |
| Stable Diffusion 3.5 Large | 8.1 billion | 16.2 GB | yes |
| FLUX.1 [dev] | 12 billion | 24 GB | yes |
| Wan 2.1 14B | 14 billion | 28 GB | yes |
Parameters published by the authors of each model; weights = parameters × 2 bytes. Text encoders (CLIP, T5) add up to about ten gigabytes in 16-bit, and ComfyUI can unload them from memory once the text is encoded.
Price and availability of the GPUs for ComfyUI
| GPU | GPUs per server | Server GPU memory | On demand, per hour | Availability |
|---|---|---|---|---|
| NVIDIA RTX A6000 | 1 | 48 GB | $0.67 | |
| NVIDIA L40 | 1 | 48 GB | $1.34 | |
| NVIDIA RTX PRO 6000 | 1 | 96 GB | $2.48 | |
| NVIDIA H100 PCIe | 1 | 80 GB | $3.35 |
Prices in Canadian dollars, taxes extra, read from the catalog when the page is built. Availability is read live; a machine out of stock shows the closest equivalent machine in stock.
- The RTX A6000 is the cheapest choice: its 48 GB are enough for every model of the table.
- The L40 has the same memory on a newer architecture (Ada Lovelace, with FP8), for CA$1.34/h.
- The RTX PRO 6000, with 96 GB, leaves room for video and large batches, at CA$2.48/h.
Launch ComfyUI
- Deploy the ComfyUI one-click template from the console. The install takes 5 to 15 minutes.
- ComfyUI opens over HTTPS in your browser, protected by a username and a password generated for you.
- Over SSH, copy your models (checkpoints, LoRA, VAE) into the ComfyUI folders and install the extensions you want.
- Generated images stay on the server’s disk until you download them or delete the server.
Costs and licenses
- A 2-hour work session on an RTX A6000: $1.34.
- An 8-hour night of batch generation: $5.36, or $4.32 on Spot if an interruption does not bother you.
- The license of some models limits their commercial use. Read the license of the model you pick before a client project.
Good to know before you launch
- Billing is hourly, at the server’s hourly rate, from prepaid credit. There is no per-minute billing, and the first hour is charged when you deploy.
- A stopped server is still billed at the full hourly rate. To pay less, hibernate it ($0.02 per hour (about $14.60 per month), disk kept) or delete it. A Spot machine cannot be hibernated: delete it when you are done.
- The minimum top-up is CA$25.00, paid by credit card.
- Servers and their disks stay in Canada (region canada-montreal, in Montreal). Account information (identity, billing, e-mails) may be processed outside Quebec, as the privacy policy explains.
Frequently asked questions
How much does a GPU for Stable Diffusion cost?
From CA$0.67/h for an NVIDIA RTX A6000 with 48 GB, taxes extra, billed by the hour in Canadian dollars.
How much GPU memory for FLUX.1 [dev]?
About 24 GB for the model in 16-bit, plus the T5 text encoder. The 48 GB of an RTX A6000 are enough without compressing it.
Is ComfyUI already installed?
Yes, with the ComfyUI one-click template: it is installed for you and served over HTTPS behind a generated username and password.
Can I add my own models and LoRA?
Yes. Over SSH, you copy your files into the ComfyUI folders and install the extensions you want.
Can I generate video?
Yes, an open video model such as Wan 2.1 14B weighs 28 GB in 16-bit and fits in 48 GB. Video generation is slow: plan for several minutes per clip.
Need GPUs? We’ve got you.
Launch an NVIDIA GPU server by the hour in Canada, paid in Canadian dollars by credit card, or reserve a sold-out GPU: it launches as soon as one frees up.
Other guides
- Your AI says you need a GPU? Run your Python script on a GPU in minutes
- CUDA out of memory: causes, fixes, and when to move to a GPU with more VRAM
- torch.cuda.is_available() returns False: causes and fixes (Mac, PC without NVIDIA, Docker)
- How much does a GPU cost per hour in Canada, in Canadian dollars?
- OpenAI API too expensive? Host your LLM on a GPU by the hour, in Canada
- Rent a GPU by the hour in Canada: prices, steps and billing
- GPU cloud in Montreal, Canada: NVIDIA GPU servers hosted in the country
- Rent an H100 in Canada: PCIe, NVLink or SXM, by the hour
- Which GPU to fine-tune an LLM? Memory, machine and cost
- Cheap GPU cloud: how to pay as little as possible for a GPU
- Ollama on a cloud GPU: run an LLM online on your own server
- Rent an 8-GPU server: 8x H100 or 8x A100 NVLink, by the hour
- RunPod alternative in Canada: GPUs by the hour, in Canadian dollars
- GPU for university research in Quebec: rent a GPU by the hour
- Health data and AI: GPUs hosted in Canada