Kolors · 3 GPUs measured first-party · text-to-image · Updated October 2026
What GPU Do You Need for Kolors?
Kolors on 3 GPUs, measured first-party: NVIDIA GeForce RTX 5090 leads at 13.53 images/min, RTX 3090 trails at 4.69 images/min, and it peaked at 23GB of VRAM.
Benchmarked weights: Kwai-Kolors/Kolors-diffusers
NVIDIA GeForce RTX 5090
13.53 images/min on Kolors, the ceiling. Measured on our bench. 32GB of VRAM, $1,999 at launch.
- 13.53 images/min on Kolors
- 32GB, clears the Kolors floor
- 575W board rating
NVIDIA GeForce RTX 4090
9.88 images/min on Kolors, fastest card you can buy at retail. Measured on our bench. 24GB of VRAM, $1,599 at launch.
- 9.88 images/min on Kolors
- 24GB, clears the Kolors floor
- 450W board rating
NVIDIA GeForce RTX 3090
4.69 images/min on Kolors, lowest launch price that still fits. Measured on our bench. 24GB of VRAM, $1,499 at launch.
- 4.69 images/min on Kolors
- 24GB, clears the Kolors floor
- 350W board rating
What GPU Do You Need for Kolors?, images/min by GPU
Efficiency: images/min per 100W drawn
Power is the average pulled during the run, sampled at 1Hz. The fastest card is often not the one here, and for anything left running this is the number that shows up on the bill.
Value: images/min per $1,000 of MSRP
Launch price, not street price, so it ages. A speed leaderboard always crowns the most expensive card; this is the counterweight.
Kolors. Measured image generation speed by GPU
| GPU | Images/min | s per image | img/W·min | Avg power |
|---|---|---|---|---|
| NVIDIA GeForce RTX 5090 | 13.53 | 4.435 | 0.023 | 593.1 W |
| NVIDIA GeForce RTX 4090 | 9.88 | 6.071 | 0.024 | 417.4 W |
| NVIDIA GeForce RTX 3090 | 4.69 | 12.789 | 0.013 | 347.7 W |
What the numbers show. Across 3 GPUs measured on our own bench, RTX 5090 is fastest at 13.5 images/min. The slowest, RTX 3090, manages 4.69, so the spread is 2.9x from top to bottom. RTX 4090 is the most efficient, 9.88 images/min at 417W.
How it compares. RTX 4090: Kolors 9.88 images/min, Playground v2.5 9.96 (3B), Z-Image Turbo 7.24, Stable Diffusion 3 Medium 13.95 (2B), SSD-1B 15.77. 3 of 4 beat Kolors here.
Cost on a rented GPU. 1,000 images of Kolors: $0.43 on a RTX 3090 ($0.12/hr, 3.6 hours), $0.48 on a RTX 5090 ($0.39/hr, 74 min, 1.1x the cost).
Kolors: cost per 1,000 images on rented GPUs
| GPU | Cheapest rate | Speed (images/min) | Cost per 1,000 images |
|---|---|---|---|
| NVIDIA GeForce RTX 3090 | $0.12/hr | 4.69 | $0.43 |
| NVIDIA GeForce RTX 5090 | $0.39/hr | 13.53 | $0.48 |
| NVIDIA GeForce RTX 4090 | $0.34/hr | 9.88 | $0.57 |
Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.
Speed tiers for Kolors. 6-30 images/min: 2 (RTX 5090, RTX 4090); under 6 images/min: 1 (RTX 3090). 30 images/min means two seconds or less per picture.
Time per image. Kolors at 1024px and 50 steps: 4.4s per image on the RTX 5090, 12.8s on the RTX 3090. A batch of 100 takes 7 min on the fastest card and 21 min on the slowest.
VRAM for Kolors. Measured peak 21.6GB, so 24GB is the smallest common card size; smallest card it ran on: RTX 4090 (24GB).
Power on Kolors. Most efficient: RTX 4090, 417W, 0.70 kWh per 1,000 images. Hungriest: RTX 5090, 593W, 0.73 kWh. At $0.15/kWh: $0.11 per 1,000 images.
Our verdict
Fastest on Kolors: NVIDIA GeForce RTX 5090, 13.53 images/min. Cheapest consumer card that ran it: NVIDIA GeForce RTX 3090 ($1,499, 4.69 images/min). Cheapest to rent per job: NVIDIA GeForce RTX 3090, $0.43 per 1,000 images.
FAQ
- What GPU do I need to run Kolors?
- About 23GB. Cheapest consumer card that ran it: NVIDIA GeForce RTX 3090 (24GB, 4.69 images/min).
- How fast is Kolors on the NVIDIA GeForce RTX 5090?
- 13.53 images/min at 593W, faster than every datacenter card we ran it on.
- How much does it cost to run Kolors in the cloud?
- $0.43 per 1,000 images on a NVIDIA GeForce RTX 3090 at $0.12/hr, cheapest of 3 rentable cards we measured.
- Can I run Kolors on a 12GB, 16GB or 24GB card?
- It used 21.6GB at the precision we tested. 12GB: no; 16GB: no; 24GB: yes.
- Is the RTX 4090 or the RTX 3090 faster for Kolors?
- The RTX 4090: 9.88 vs 4.69 images/min, 111% faster on our bench.
- Should I buy or rent a RTX 5090 for Kolors?
- Its $1,999 launch price buys 5,139 rented hours at $0.39/hr, enough for about 4,172x 1,000 images of Kolors. Buy only if you'll run more than that.
How we test
Kolors at 1024px in diffusers, bf16, native precision with no offload, timed over three generations after a warmup, with power and VRAM sampled throughout. Image speed tracks tensor compute and architecture generation more than memory bandwidth, so the order here differs from our LLM boards.