For Wan 2.2's large video models, rent an 80 GB H100. In Wan's own tests it made a 720p clip in about 17 minutes, against about 46 minutes on an A100.
On a 24 GB RTX 4090, Wan's smaller TI2V-5B model made a 720p clip in about 9 minutes. LTX and HunyuanVideo publish memory figures but no minutes per clip.
The times come from the table in Wan 2.2's README, and the memory figures from each maker's page, all read on 27 October 2026. Clips an hour and the ratios are our calculation.
Who this is for
- you want to make AI video with Wan, LTX or HunyuanVideo on a rented card
- you are choosing between a 24 GB card and an 80 GB one
- you want to know how many clips an hour of GPU time buys
Not for you if you only need to know whether a model fits: see our GPU memory guide for image and video models.
Key numbers
- 17 minutes for a 720p clip from Wan 2.2 T2V-A14B on one H100, in Wan's own test Official page,
- 9 minutes for a 720p clip from Wan 2.2 TI2V-5B on one RTX 4090, in the same table Official page,
- 2.6 times how much longer the A100 took than the H100 for the 720p clip Our calculation,
Minutes per clip on each card
Wan tested its models on several cards and published the total time and peak memory for each. On one card it kept parts of the model in the computer's main memory, which saves GPU memory but adds time.
The H100 made 720p clips from the 14B model about 2.6 times as fast as the A100.
| Card | Model | Resolution | Minutes per clip | Peak memory (GB) | Clips an hour |
|---|---|---|---|---|---|
| RTX 4090 (24 GB) | TI2V-5B, text to video | 720p | 8.9 | 22.9 | 6.7 |
| RTX 4090 (24 GB) | TI2V-5B, image to video | 720p | 8.7 | 22.8 | 6.9 |
| H100 (80 GB) | T2V-A14B, text to video | 480p | 5.4 | 41.3 | 11.0 |
| H100 (80 GB) | T2V-A14B, text to video | 720p | 17.4 | 59.8 | 3.5 |
| A100 (80 GB) | T2V-A14B, text to video | 480p | 13.1 | 41.3 | 4.6 |
| A100 (80 GB) | T2V-A14B, text to video | 720p | 45.6 | 59.8 | 1.3 |
Wan's table groups the H100 with the H800, and the A100 with the A800. Its image-to-video model, I2V-A14B, took almost exactly the same time as the text-to-video one.
The table does not state the clip length. Wan's default settings make clips of about five seconds. That is 81 frames at 16 a second for the 14B models, and 121 at 24 for the 5B.
Cost per clip
Divide the card's hourly price by the clips it makes in an hour. The price boxes below show today's hourly and monthly prices.
Speed decides more than the hourly price. For 720p clips, the H100 costs less per clip than the A100 unless its hourly price is more than 2.6 times higher (our calculation). At 480p the ratio is 2.4.
Count the clips you keep, not the clips you make. If you usually need three tries for one good clip, each kept clip costs three times the GPU time.
Several GPUs at once
Wan also tested four and eight cards working together. Eight H100s made the 720p clip in 155 seconds, about 2.6 minutes, instead of 17 minutes on one.
That saves waiting, not money. Eight cards for 155 seconds is about a fifth more GPU time than one card for the whole clip, by our calculation.
What LTX and HunyuanVideo need
By the files and figures their makers publish, collected in our GPU memory guide for image and video models:
- LTX-2.5 from Lightricks, which makes video with sound: its files take 70.1 GB at 16-bit, so it needs an 80 GB card (our reading). With its 8-bit files for ComfyUI, 38.7 GB, a 48 GB card works.
- HunyuanVideo from Tencent: peaks of 45 GB at 544 x 960 and 60 GB at 720 x 1,280, for 129 frames.
- HunyuanVideo 1.5: at least 14 GB with offloading, by Tencent.
Lightricks' model card for LTX-2.5 needs a Hugging Face login and comes under its own community licence. Read the terms before commercial use, as with every model here.
The hidden first hour: downloading the model
On a fresh server, the model has to arrive first. Wan 2.2 T2V-A14B's files take 69.0 GB at 16-bit, and TI2V-5B's 22.8 GB, by our sums in the memory guide.
Even at a full 1 Gbps, 69 GB takes about 9 minutes, by our calculation, and real downloads are often slower. Keep the files on storage that survives between sessions, or on a monthly server.
Which card for which job
- Short 720p clips on a budget: a 24 GB card with Wan 2.2 TI2V-5B, as Wan's README says it runs on an RTX 4090.
- Wan 2.2's 14B models: an 80 GB card. The README suggests dropping the offloading options on 80 GB to run faster.
- LTX-2.5 at 8-bit or HunyuanVideo at 544 x 960: a 48 GB card such as the L40S.
- Many clips against a deadline: several H100 cards together, as in Wan's eight-GPU test.
| Plan | Price | New-customer offer | Link |
|---|---|---|---|
| GPU Mart Enterprise Dedicated GPU Server - RTX 4090 24 GB GPU | $409/month monthly rate of a 24-month term, paid in advance · VAT not stated checked 25 Oct 2026 | A free 24-hour trial of a VPS, dedicated or GPU server for new clients, on request checked 29 Oct 2026 | Visit |
| Unihost AZM-04, NVIDIA RTX 4090 24 GB GPU | €359.90/month monthly, paid in advance; minimum one month checked 25 Oct 2026 | No offer | Visit |
| Liquid Web GPU server, NVIDIA L4 24GB 24 GB GPU | $584.25/month monthly; 25% off the first month, then $779 · VAT not stated checked 25 Oct 2026 | No offer | Visit |
| UpCloud GPU Server, 1 x NVIDIA L4, 8 cores, 64 GB 24 GB GPU | €0.58/hour (about €390/month; no charge after 28 days of use)30-day month: the last 2 days are free.31-day month: the last 3 days are free.February: no free days (one in a leap year).Started mid-month: it may never reach the cap that month. on demand, billed by the starting hour; monthly cost capped at 28 days of use checked 25 Oct 2026 | No offer | Visit |
Monthly costs for hourly prices assume the server runs all month (730 hours), or less where a provider caps the monthly bill.
Monthly plans suit daily work; hourly ones suit a few clips.
| Plan | Price | New-customer offer | Link |
|---|---|---|---|
| DigitalOcean GPU Droplet, 1 x NVIDIA L40S 48 GB GPU | $1.57/hour (about $1,146/month) on demand, billed per second · VAT not stated checked 25 Oct 2026 | New accounts get a $5 credit that lasts 90 days and covers every product except SaaS add-ons checked 29 Oct 2026 | Visit |
| UpCloud GPU Server, 1 x NVIDIA L40S, 8 cores, 64 GB 48 GB GPU | €1.11/hour (about €746/month; no charge after 28 days of use)30-day month: the last 2 days are free.31-day month: the last 3 days are free.February: no free days (one in a leap year).Started mid-month: it may never reach the cap that month. on demand, billed by the starting hour; monthly cost capped at 28 days of use checked 25 Oct 2026 | No offer | Visit |
| Liquid Web GPU server, NVIDIA L40S 48GB 48 GB GPU | $1,049.25/month monthly; 25% off the first month, then $1,399 · VAT not stated checked 25 Oct 2026 | No offer | Visit |
Per-second and hourly billing: delete the server when the clips are done.
| Plan | Price | New-customer offer | Link |
|---|---|---|---|
| DigitalOcean GPU Droplet, 1 x NVIDIA H100 80 GB GPU | $4.41/hour (about $3,219/month) on demand, billed per second · VAT not stated checked 25 Oct 2026 | New accounts get a $5 credit that lasts 90 days and covers every product except SaaS add-ons checked 29 Oct 2026 | Visit |
| UpCloud GPU Server, 1 x NVIDIA H100, 12 cores, 240 GB 80 GB GPU | €2.17/hour (about €1,458/month; no charge after 28 days of use)30-day month: the last 2 days are free.31-day month: the last 3 days are free.February: no free days (one in a leap year).Started mid-month: it may never reach the cap that month. on demand, billed by the starting hour; monthly cost capped at 28 days of use checked 25 Oct 2026 | No offer | Visit |
| GPU Mart Enterprise Dedicated GPU Server - H100 80 GB GPU | $2,099/month monthly rate of a 24-month term, paid in advance · VAT not stated checked 25 Oct 2026 | A free 24-hour trial of a VPS, dedicated or GPU server for new clients, on request checked 29 Oct 2026 | Visit |
| GPU Mart Enterprise Dedicated GPU Server - A100(80GB) 80 GB GPU | $1,559/month monthly rate of a 24-month term, paid in advance · VAT not stated checked 25 Oct 2026 | A free 24-hour trial of a VPS, dedicated or GPU server for new clients, on request checked 29 Oct 2026 | Visit |
Compare the cost per clip, not only the hourly price.
To run these models through a visual editor, see our guide to running ComfyUI in the cloud.
How we checked
We read Wan 2.2's README, its efficiency table and its configuration files, and the LTX and HunyuanVideo pages, on 27 October 2026. Minutes, clips an hour and ratios are our calculation from Wan's seconds.
Our own measurements will be added here, with their dates, when our test bench runs them.
Mistakes to avoid
Renting by the cheapest hourly price
Instead: A slower card can cost more per clip; divide by clips an hour.
Forgetting the download time
Instead: 69 GB of model files can take a large part of the first hour.
Counting every clip as usable
Instead: Divide the cost by the clips you keep.
Running the 14B model on a 24 GB card
Instead: Wan measured peaks of 41.3 GB at 480p and 59.8 GB at 720p.
Other questions
What is the best GPU for Wan 2.2?
For the 14B models, an 80 GB H100: about 17 minutes per 720p clip in Wan's test, against 46 on an A100. For TI2V-5B, a 24 GB RTX 4090 is enough.
Can Wan 2.2 run on an RTX 4090?
The 5B model can: Wan measured about 9 minutes per 720p clip with a peak of 22.9 GB. The 14B models need about 41 GB even at 480p.
How long does Wan 2.2 take to make a clip?
In Wan's own tests, a clip of about five seconds took from about 5 minutes (480p, H100) to about 46 minutes (720p, A100).
Sources and test notes
- Wan 2.2 README on GitHub (computational efficiency table), read
- Wan 2.2 efficiency table (image in the README), read
- Wan 2.2 default settings: 14B models, read
- Wan 2.2 default settings: TI2V-5B, read
- LTX-2.5 on Hugging Face (Lightricks), read
- HunyuanVideo README on GitHub (Tencent), read
- HunyuanVideo 1.5 README on GitHub (Tencent), read
Every page read on 27 October 2026. The times and peaks are Wan's own; minutes, clips an hour and ratios are our calculation.