SDXL Turbo · 22 GPUs measured first-party · text-to-image · Updated October 2026
What GPU Do You Need for SDXL Turbo?
We ran SDXL Turbo at 512px on 22 GPUs, from budget cards to datacenter parts, and measured image speed, power draw and peak VRAM on every one. The fastest was NVIDIA GeForce RTX 5090 at 902.0 images/min.
Benchmarked weights: stabilityai/sdxl-turbo
NVIDIA GeForce RTX 5090
902.0 images/min on SDXL Turbo, the ceiling. Measured on our bench. 32GB of VRAM, $1,999 at launch.
- 902.0 images/min on SDXL Turbo
- 32GB, clears the SDXL Turbo floor
- 575W board rating
NVIDIA GeForce RTX 4090
652.1 images/min on SDXL Turbo, fastest card you can buy at retail. Measured on our bench. 24GB of VRAM, $1,599 at launch.
- 652.1 images/min on SDXL Turbo
- 24GB, clears the SDXL Turbo floor
- 450W board rating
GeForce RTX 4060
182.0 images/min on SDXL Turbo, lowest launch price that still fits. Measured on our bench. 8GB of VRAM, $299 at launch.
- 182.0 images/min on SDXL Turbo
- 8GB, clears the SDXL Turbo floor
- 115W board rating
GeForce RTX 5070
383.4 images/min on SDXL Turbo, most speed per dollar. Measured on our bench. 12GB of VRAM, $549 at launch. That is 698.3 images/min per $1,000 of launch price.
- 383.4 images/min on SDXL Turbo
- 12GB, clears the SDXL Turbo floor
- 250W board rating
What GPU Do You Need for SDXL Turbo?, images/min by GPU
Top 15 shown; 7 more cards in the full table below.
Efficiency: images/min per 100W drawn
Top 15 shown; 6 more cards in the full table below.
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
Top 15 shown; 7 more cards in the full table below.
Launch price, not street price, so it ages. A speed leaderboard always crowns the most expensive card; this is the counterweight.
SDXL Turbo. Measured image speed by GPU
| GPU | Images/min | s per image | img/W·min | Avg power |
|---|---|---|---|---|
| NVIDIA GeForce RTX 5090 | 902 | 0.067 | 27.088 | 33.3 W |
| NVIDIA GeForce RTX 4090 | 652.1 | 0.092 | 7.23 | 90.2 W |
| NVIDIA H100 80GB HBM3 | 618.3 | 0.1 | 5.135 | 120.4 W |
| NVIDIA GeForce RTX 4070 Ti Super | 532.3 | 0.113 | 8.201 | 64.9 W |
| GeForce RTX 5080 | 512.8 | 0.117 | 6.17 | 83.1 W |
| NVIDIA GeForce RTX 4080 | 484.8 | 0.124 | 6.224 | 77.9 W |
| NVIDIA GeForce RTX 3090 Ti | 456.1 | 0.132 | 4.394 | 103.8 W |
| NVIDIA GeForce RTX 3080 Ti | 440.9 | 0.136 | 2.931 | 150.4 W |
| GeForce RTX 5070 Ti | 431.4 | 0.139 | 8.033 | 53.7 W |
| NVIDIA L40S | 423.5 | 0.14 | 4.598 | 92.1 W |
| GeForce RTX 4080 Super | 410.1 | 0.146 | 6.111 | 67.1 W |
| GeForce RTX 5070 | 383.4 | 0.157 | 8.209 | 46.7 W |
| NVIDIA GeForce RTX 3090 | 333 | 0.18 | 2.308 | 144.3 W |
| NVIDIA GeForce RTX 4070 | 286 | 0.21 | 5.779 | 49.5 W |
| GeForce RTX 5060 Ti | 260.8 | 0.23 | 5.434 | 48.0 W |
| NVIDIA GeForce RTX 4060 Ti 16GB | 255.5 | 0.235 | 3.967 | 64.4 W |
| NVIDIA GeForce RTX 3070 Ti | 250.1 | 0.24 | 1.633 | 153.2 W |
| NVIDIA L4 | 229.9 | 0.26 | 5.008 | 45.9 W |
| NVIDIA GeForce RTX 3060 | 216.6 | 0.277 | 2.622 | 82.6 W |
| NVIDIA GeForce RTX 3070 Founders Edition | 206.4 | 0.291 | 2.345 | 88.0 W |
| NVIDIA GeForce RTX 3060 Ti | 201.9 | 0.297 | 2.524 | 80.0 W |
| GeForce RTX 4060 | 182 | 0.33 | — | — |
What the numbers show. Across 20 GPUs measured on our own bench, RTX 5090 is fastest at 902 images/min. Per dollar of launch price, RTX 5060 Ti gives the most (792.8 images/min per $1,000). The fastest card with 16GB or less is RTX 4070 Ti Super at 532 images/min. The measured peak was ~8GB, so it runs on 8GB cards and up. That floor is a property of the model, so it applies to every GPU, measured or not.
About SDXL Turbo. SDXL Turbo: from stabilityai, 2.6B parameters, on Hugging Face since November 2023. 934,040 downloads in the last 30 days and 1 community quantizations.
How it compares. L40S: SDXL Turbo 423.5 images/min, Playground v2.5 10.85 (3B), Stable Diffusion 3.5 Medium 11.58 (2B), FLUX.2 klein 4B 50.29 (4B), Stable Diffusion 1.5 58.54. SDXL Turbo beats all 4 here.
Cost on a rented GPU. 1,000 images of SDXL Turbo: $0.003 on a RTX 3060 ($0.036/hr, 5 min), $0.007 on a RTX 5090 ($0.39/hr, 1 min, 2.6x the cost).
SDXL Turbo: cost per 1,000 images on rented GPUs
| GPU | Cheapest rate | Speed (images/min) | Cost per 1,000 images |
|---|---|---|---|
| NVIDIA GeForce RTX 3060 | $0.036/hr | 216.6 | $0.003 |
| NVIDIA GeForce RTX 4070 Ti Super | $0.16/hr | 532.3 | $0.005 |
| GeForce RTX 5070 Ti | $0.15/hr | 431.4 | $0.006 |
| NVIDIA GeForce RTX 3070 Founders Edition | $0.075/hr | 206.4 | $0.006 |
| NVIDIA GeForce RTX 3090 | $0.12/hr | 333 | $0.006 |
| NVIDIA GeForce RTX 3080 Ti | $0.18/hr | 440.9 | $0.007 |
| GeForce RTX 5080 | $0.21/hr | 512.8 | $0.007 |
| NVIDIA GeForce RTX 4080 | $0.20/hr | 484.8 | $0.007 |
| NVIDIA GeForce RTX 4070 | $0.12/hr | 286 | $0.007 |
| NVIDIA GeForce RTX 5090 | $0.39/hr | 902 | $0.007 |
| GeForce RTX 5070 | $0.19/hr | 383.4 | $0.008 |
| NVIDIA GeForce RTX 4090 | $0.34/hr | 652.1 | $0.009 |
| GeForce RTX 5060 Ti | $0.14/hr | 260.8 | $0.009 |
| GeForce RTX 4080 Super | $0.22/hr | 410.1 | $0.009 |
| NVIDIA GeForce RTX 3090 Ti | $0.27/hr | 456.1 | $0.010 |
| NVIDIA L40S | $0.79/hr | 423.5 | $0.031 |
| NVIDIA L4 | $0.44/hr | 229.9 | $0.032 |
| NVIDIA H100 80GB HBM3 | $2.14/hr | 618.3 | $0.058 |
Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.
Speed tiers for SDXL Turbo. 30+ images/min: 22 (RTX 5090, RTX 4090, RTX 4070 Ti Super). 30 images/min means two seconds or less per picture.
Time per image. SDXL Turbo at 512px and 1 steps: 0.1s per image on the RTX 5090, 0.3s on the RTX 4060. A batch of 100 takes 0 min on the fastest card and 1 min on the slowest.
VRAM for SDXL Turbo. Measured peak 7.6GB, so 8GB is the smallest common card size; smallest card it ran on: RTX 3070 Ti (8GB).
Power on SDXL Turbo. Most efficient: RTX 4070 Ti Super, 65W, 2.0 Wh per 1,000 images. Hungriest: RTX 3070 Ti, 153W, 10.2 Wh.
Our verdict
Fastest on SDXL Turbo: NVIDIA GeForce RTX 5090, 902.0 images/min. Cheapest consumer card that ran it: GeForce RTX 4060 ($299, 182.0 images/min). Cheapest to rent per job: NVIDIA GeForce RTX 3060, $0.003 per 1,000 images.
FAQ
- What GPU do I need to run SDXL Turbo?
- About 8GB. Cheapest consumer card that ran it: GeForce RTX 4060 (8GB, 182.0 images/min).
- How fast is SDXL Turbo on the NVIDIA GeForce RTX 5090?
- 902.0 images/min, faster than every datacenter card we ran it on.
- How much does it cost to run SDXL Turbo in the cloud?
- $0.003 per 1,000 images on a NVIDIA GeForce RTX 3060 at $0.036/hr, cheapest of 18 rentable cards we measured.
- Can I run SDXL Turbo on a 12GB, 16GB or 24GB card?
- It used 7.6GB at the precision we tested. 12GB: yes; 16GB: yes; 24GB: yes.
- Is the RTX 4090 or the RTX 3090 faster for SDXL Turbo?
- The RTX 4090: 652.1 vs 333.0 images/min, 96% faster on our bench.
- Should I buy or rent a RTX 5090 for SDXL Turbo?
- Its $1,999 launch price buys 5,139 rented hours at $0.39/hr, enough for about 278,125x 1,000 images of SDXL Turbo. Buy only if you'll run more than that.
How we test
SDXL Turbo at 512px 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.