Qwen3 32B · 56 cards measured first-party · Updated October 2026

How Fast Does Qwen3 32B Run on Each GPU?

Qwen3 32B is the largest model most people can realistically run at home. It needs roughly 20GB at Q4_K_M, which clears a 24GB consumer card with room to spare and stops a 16GB card dead. That makes it the most useful single benchmark on this site for anyone choosing between the two.

Benchmarked weights: unsloth/Qwen3-32B-GGUF

Fastest we measured

NVIDIA B300

83.68 tok/s on Qwen3 32B, the ceiling. Measured on our bench. 288GB of VRAM, $40,000 at launch.

Pros

• 83.68 tok/s on Qwen3 32B
• 288GB, clears the Qwen3 32B floor
• Rentable by the hour rather than bought

Cons

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

Best consumer card

NVIDIA GeForce RTX 5090

71.15 tok/s on Qwen3 32B, fastest card you can buy at retail. Measured on our bench. 32GB of VRAM, $1,999 at launch.

Pros

• 71.15 tok/s on Qwen3 32B
• 32GB, clears the Qwen3 32B floor

Cons

• 575W board rating

Cheapest card that runs it

AMD Radeon RX 7900 XTX

41.4 tok/s on Qwen3 32B, lowest launch price that still fits. Anchored estimate. 24GB of VRAM, $999 at launch.

Pros

• 41.4 tok/s on Qwen3 32B
• 24GB, clears the Qwen3 32B floor

Cons

• 355W board rating

Best value

NVIDIA GeForce RTX 4090

44.28 tok/s on Qwen3 32B, most speed per dollar. Measured on our bench. 24GB of VRAM, $1,599 at launch. That is 27.69 tok/s per $1,000 of launch price.

Pros

• 44.28 tok/s on Qwen3 32B
• 24GB, clears the Qwen3 32B floor

Cons

• 450W board rating

Like every LLM in our suite, 32B token generation is bandwidth-bound. The card reads ~20GB of weights per token, so the ranking follows memory bandwidth and largely ignores compute. A card with half the tensor cores and the same bandwidth will land in roughly the same place.

Because this workload is bandwidth-bound, the ranking above tracks memory bandwidth far more closely than core counts or price. A card with fewer tensor cores and faster memory will beat a card with the opposite.

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 Qwen3 32B leaderboard at 83.68 tok/s (measured), 574% 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 59 cards that can't run Qwen3 32B at all. This is a bandwidth workload: buy memory speed, not tensor cores.

FAQ

What is the fastest GPU for Qwen3 32B?

NVIDIA B300, at 83.68 tok/s on our bench, a first-party measurement. It carries 288GB of VRAM. Of the 101 cards we have Qwen3 32B data for, 42 can run it at all.

How much VRAM do I need for Qwen3 32B?

~20GB at Q4_K_M. This is the cliff that decides the 16GB-versus-24GB question, and it's why we treat VRAM as the first spec and speed as the second.

Why does the Qwen3 32B ranking look different from your other benchmarks?

Because this workload is bandwidth-bound, the ranking above tracks memory bandwidth far more closely than core counts or price. A card with fewer tensor cores and faster memory will beat a card with the opposite. That's why we publish twelve separate workloads rather than one blended score, the ordering genuinely changes depending on the job.

Are these Qwen3 32B numbers measured or estimated?

Both, and every row says which. 56 of the 101 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 Qwen3 32B 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 Qwen3 32B?

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'.