Qwen3 4B · 39 cards measured first-party · Updated July 2026

How Fast Does Qwen3 4B Run on Each GPU?

Qwen3 4B is the entry point: ~5GB, runs on essentially anything with a modern GPU in it, and fast enough that the bottleneck stops being the card and starts being how quickly you can read. It's the model to reach for on a laptop or an old 8GB card.

Benchmarked weights: Qwen/Qwen3-4B-GGUF

Fastest we measured

NVIDIA GeForce RTX 5090

375.6 tok/s on Qwen3 4B. Measured on our bench. 32GB of VRAM, 575W board rating. AI Score 22.3/100 across our full 12-workload suite.

Pros

• 375.6 tok/s on Qwen3 4B
• 32GB, clears the Qwen3 4B floor
• Rentable by the hour rather than bought

Cons

• 575W board rating
• Datacenter or workstation hardware, not a retail purchase

Best for: Qwen3 4B work where you want the ceiling gone rather than the cheapest entry.

Runner-up

NVIDIA H100 NVL

364.7 tok/s on Qwen3 4B. Anchored estimate. 94GB of VRAM, 400W board rating. AI Score 67.0/100 across our full 12-workload suite.

Pros

• 364.7 tok/s on Qwen3 4B
• 94GB, clears the Qwen3 4B floor
• Rentable by the hour rather than bought

Cons

• 400W board rating
• Datacenter or workstation hardware, not a retail purchase

Best for: Qwen3 4B work where you want the ceiling gone rather than the cheapest entry.

Third

NVIDIA RTX PRO 6000 Blackwell Workstation Edition

362.81 tok/s on Qwen3 4B. Measured on our bench. 96GB of VRAM, 600W board rating. AI Score 53.1/100 across our full 12-workload suite.

Pros

• 362.81 tok/s on Qwen3 4B
• 96GB, clears the Qwen3 4B floor
• Rentable by the hour rather than bought

Cons

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

Best for: Qwen3 4B work where you want the ceiling gone rather than the cheapest entry.

Bandwidth-bound, but with a twist worth knowing: at 4B the model is so small that the fastest cards stop being fully occupied. On our measured B300 this workload sat at just 12-30% utilisation, the chip spends its time waiting rather than computing. Past a certain point, buying more GPU stops buying more tokens.

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 GeForce RTX 5090 tops our Qwen3 4B leaderboard at 375.6 tok/s (measured), 530% of the way clear of the slowest card that still fits. This is a bandwidth workload: buy memory speed, not tensor cores.

FAQ

What is the fastest GPU for Qwen3 4B?

NVIDIA GeForce RTX 5090, at 375.6 tok/s on our bench, a first-party measurement. It carries 32GB of VRAM. Of the 61 cards we have Qwen3 4B data for, 61 can run it at all.

How much VRAM do I need for Qwen3 4B?

~5GB at Q4_K_M. If a card can't run this, it can't run anything in our LLM ladder.

Why does the Qwen3 4B 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 4B numbers measured or estimated?

Both, and every row says which. 39 of the 61 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 4B 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 4B?

Nearly every card in our fleet clears this workload's VRAM floor, so there are no gates on this page. That's unusual, most of our suite excludes a meaningful chunk of the market.

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'.