Qwen-Image 2.1 · 4 GPUs measured first-party · text-to-image · Updated October 2026
What GPU Do You Need for Qwen-Image 2.1?
Qwen-Image 2.1 on 4 GPUs, measured first-party: NVIDIA H100 80GB HBM3 leads at 4.59 images/min, L40S trails at 1.78 images/min, and it peaked at 41GB of VRAM.
Benchmarked weights: Qwen/Qwen-Image-2.1
NVIDIA H100 80GB HBM3
4.59 images/min on Qwen-Image 2.1, the ceiling. Measured on our bench. 80GB of VRAM, $30,000 at launch.
- 4.59 images/min on Qwen-Image 2.1
- 80GB, clears the Qwen-Image 2.1 floor
- Rentable by the hour rather than bought
- 700W board rating
- Datacenter or workstation hardware, not a retail purchase
NVIDIA RTX PRO 6000 Blackwell Workstation Edition
3.21 images/min on Qwen-Image 2.1, lowest launch price that still fits. Measured on our bench. 96GB of VRAM, $8,565 at launch.
- 3.21 images/min on Qwen-Image 2.1
- 96GB, clears the Qwen-Image 2.1 floor
- Rentable by the hour rather than bought
- 600W board rating
- Datacenter or workstation hardware, not a retail purchase
What GPU Do You Need for Qwen-Image 2.1?, 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.
Qwen-Image 2.1. Measured image generation speed by GPU
| GPU | Images/min | s per image | img/W·min | Avg power |
|---|---|---|---|---|
| NVIDIA H100 80GB HBM3 | 4.59 | 13.08 | 0.007 | 689.6 W |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 3.21 | 18.67 | 0.005 | 591.6 W |
| NVIDIA A100 80GB SXM4 | 2.06 | 29.19 | 0.005 | 395.1 W |
| NVIDIA L40S | 1.78 | 33.77 | 0.005 | 347.3 W |
What the numbers show. Across 3 GPUs measured on our own bench, H100 80GB HBM3 is fastest at 4.59 images/min. The slowest, L40S, manages 1.78, so the spread is 2.6x from top to bottom. Per dollar of launch price, RTX PRO 6000 Blackwell Workstation Edition gives the most (0.4 images/min per $1,000).
About Qwen-Image 2.1. Qwen-Image 2.1: from Qwen, 7.1B parameters, on Hugging Face since September 2026. 8,588,141 downloads in the last 30 days and 18 community quantizations.
How it compares. L40S: Qwen-Image 2.1 1.78 images/min, ERNIE-Image Turbo 4.41 (8B), Z-Image 1.31 (6B), Z-Image Turbo (1024px) 18.2 (6B), Stable Diffusion 3.5 Large 4.54 (8B). 3 of 4 beat Qwen-Image 2.1 here.
Cost on a rented GPU. 1,000 images of Qwen-Image 2.1: $5.59 on a RTX PRO 6000 Blackwell Workstation Edition ($1.08/hr, 5.2 hours), $7.76 on a H100 80GB HBM3 ($2.14/hr, 3.6 hours, 1.4x the cost).
Qwen-Image 2.1: cost per 1,000 images on rented GPUs
| GPU | Cheapest rate | Speed (images/min) | Cost per 1,000 images |
|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | $1.08/hr | 3.21 | $5.59 |
| NVIDIA L40S | $0.79/hr | 1.78 | $7.40 |
| NVIDIA A100 80GB SXM4 | $0.95/hr | 2.06 | $7.66 |
| NVIDIA H100 80GB HBM3 | $2.14/hr | 4.59 | $7.76 |
Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.
Speed tiers for Qwen-Image 2.1. under 6 images/min: 4 (H100 80GB HBM3, RTX PRO 6000 Blackwell Workstation Edition, A100 80GB SXM4). 30 images/min means two seconds or less per picture.
Time per image. Qwen-Image 2.1 at 1024px and 40 steps: 13.1s per image on the H100 80GB HBM3, 33.7s on the L40S. A batch of 100 takes 22 min on the fastest card and 56 min on the slowest.
VRAM for Qwen-Image 2.1. Measured peak 38.8GB, so 48GB is the smallest common card size; smallest card it ran on: L40S (48GB).
Power on Qwen-Image 2.1. Most efficient: H100 80GB HBM3, 690W, 2.50 kWh per 1,000 images. At $0.15/kWh: $0.38 per 1,000 images.
Our verdict
Fastest on Qwen-Image 2.1: NVIDIA H100 80GB HBM3, 4.59 images/min. Cheapest to rent per job: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, $5.59 per 1,000 images.
FAQ
- What GPU do I need to run Qwen-Image 2.1?
- About 41GB. Smallest card that ran it: NVIDIA L40S (48GB).
- How much does it cost to run Qwen-Image 2.1 in the cloud?
- $5.59 per 1,000 images on a NVIDIA RTX PRO 6000 Blackwell Workstation Edition at $1.08/hr, cheapest of 4 rentable cards we measured.
- Can I run Qwen-Image 2.1 on a 12GB, 16GB or 24GB card?
- It used 38.8GB at the precision we tested. 12GB: no; 16GB: no; 24GB: no.
- Is the H100 80GB HBM3 or the A100 80GB SXM4 faster for Qwen-Image 2.1?
- The H100 80GB HBM3: 4.59 vs 2.06 images/min, 123% faster on our bench.
How we test
Qwen-Image 2.1 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.