Qwen-Image-Edit · 47 cards measured first-party · Updated October 2026

How Fast Is Qwen-Image-Edit on Each GPU?

Qwen-Image-Edit is the heaviest image workload in our suite: ~42GB of VRAM minimum, ~58GB to run clean, and it wants 58GB+ of system RAM on top. It excludes more hardware than anything except Llama 3.3 70B, and for the same reason. You can't optimise your way past a memory ceiling.

Benchmarked weights: Qwen/Qwen-Image-Edit

Fastest we measured

NVIDIA B300

8.14 images/min on Qwen-Image-Edit, the ceiling. Measured on our bench. 288GB of VRAM, $40,000 at launch.

Pros

• 8.14 images/min on Qwen-Image-Edit
• 288GB, clears the Qwen-Image-Edit floor
• Rentable by the hour rather than bought

Cons

• 1400W board rating
• Datacenter or workstation hardware, not a retail purchase

Best for: Qwen-Image-Edit work where you want the ceiling gone rather than the cheapest entry.

Cheapest card that runs it

NVIDIA RTX PRO 5000 Blackwell

0.8 images/min on Qwen-Image-Edit, lowest launch price that still fits. Measured on our bench. 48GB of VRAM, $4,500 at launch.

Pros

• 0.8 images/min on Qwen-Image-Edit
• 48GB, clears the Qwen-Image-Edit floor
• Rentable by the hour rather than bought

Cons

• 300W board rating
• Datacenter or workstation hardware, not a retail purchase

Best value

NVIDIA RTX PRO 6000 Blackwell Workstation Edition

2.64 images/min on Qwen-Image-Edit, most speed per dollar. Measured on our bench. 96GB of VRAM, $8,565 at launch. That is 0.31 images/min per $1,000 of launch price.

Pros

• 2.64 images/min on Qwen-Image-Edit
• 96GB, clears the Qwen-Image-Edit floor
• Rentable by the hour rather than bought

Cons

• 600W board rating
• Datacenter or workstation hardware, not a retail purchase

Compute-bound, but capacity is what decides whether you're in the conversation at all. The ~42GB floor puts this workload out of reach of every consumer card and most workstation cards, which is why the leaderboard below is almost entirely datacenter silicon.

Because this workload is tensor-compute bound, the ranking tracks architecture generation and tensor throughput rather than memory bandwidth, the reverse of our LLM charts. The same two cards can swap places entirely depending on which of these pages you're reading. That's the reason we run twelve workloads instead of publishing one score. A GPU isn't fast or slow. It's fast at some things and gated out of others, and which of those matters depends entirely on what you're actually going to run.

Our verdict

NVIDIA B300 tops our Qwen-Image-Edit leaderboard at 8.14 images/min (measured), 4688% of the way clear of the slowest card that still fits. But the number that decides most purchases isn't on the chart. It's the 60 cards that can't run Qwen-Image-Edit at all. This is a compute workload: buy architecture generation, not raw VRAM, as long as you clear the floor first.

FAQ

What is the fastest GPU for Qwen-Image-Edit?

NVIDIA B300, at 8.14 images/min on our bench, a first-party measurement. It carries 288GB of VRAM. Of the 79 cards we have Qwen-Image-Edit data for, 19 can run it at all.

How much VRAM do I need for Qwen-Image-Edit?

~42GB VRAM minimum, ~58GB for the full path, plus substantial system RAM for weight staging. This and Llama 3.3 70B are the two workloads that define the top of our board.

Why does the Qwen-Image-Edit ranking look different from your other benchmarks?

Because this workload is tensor-compute bound, the ranking tracks architecture generation and tensor throughput rather than memory bandwidth, the reverse of our LLM charts. The same two cards can swap places entirely depending on which of these pages you're reading. That's why we publish twelve separate workloads rather than one blended score, the ordering genuinely changes depending on the job.

Are these Qwen-Image-Edit numbers measured or estimated?

Both, and every row says which. 47 of the 79 cards here were rented and run by us on the same harness. The remainder are anchored estimates interpolated per workload against those measurements. We never blend the two silently, if a row says Estimated, we have not run that card.

Can I rent a GPU to run Qwen-Image-Edit instead of buying one?

Yes, and for the cards at the top of this leaderboard it's the only realistic option, most of them have no retail channel at all. It's also how we got these numbers: we rented the hardware by the hour rather than buying it. That's worth considering before you spend on a card to find out whether it's fast enough.

Why publish cards that can't run Qwen-Image-Edit?

Because it's the most useful thing we know. A card that can't load a model doesn't run it slowly, it doesn't run it. Most benchmark sites leave that as a blank cell or quietly drop to a smaller quantisation to produce a number. We publish it as a hard gate and score it zero, because 'this card cannot do the thing you want' is the answer to the question you were actually asking.

How we test

Every ranking on this page comes from our own benchmark runs, not vendor claims. Cards marked Measured were rented and run by us; cards marked Estimated are interpolated per workload against those measured anchors and are labelled on every row, we never blend the two silently. LLMs run on llama.cpp (llama-bench) at Q4_K_M with -p 512 -n 128. Diffusion and video run on diffusers/ComfyUI at BF16, with SDXL at FP16. Each workload gets a warmup pass plus multiple timed runs (5 for small LLMs, 3 for large models and images, 2 for video); we publish the mean as the result and the minimum as the 1% low. Run-to-run variance is under 0.5%. Telemetry, power, temperature, utilisation, clocks, peak VRAM, is sampled at 1 Hz for the duration of every run. Where a model exceeds a card's VRAM we publish a hard won't-fit result rather than quietly dropping to a smaller quantisation. A card that can't run a model scores zero on it. Silently swapping precision to make a number appear would make every number on this site meaningless. All figures are single-GPU, single-stream, batch-size-1. That is the honest way to measure what one card does for one user, and it is deliberately not how a datacenter serves a model. Vendor and MLPerf figures use large batches across many GPUs and will be far higher. Neither is wrong, they answer different questions. Ours answers 'what will this card do for me'.