Qwen3 32B hardware requirements

What is Qwen3 32B?

Qwen3 32B is the largest dense model in the Qwen3 family, released by Alibaba Cloud's Qwen team in April 2025. It switches between a default thinking mode for reasoning, math and coding and a non-thinking mode for general chat, with tool calling in both. Its native context is 32k tokens, and the model card explains how to extend it to 128k with YaRN.

Model card

About the model

Parameters: 32.8B
Attention: Full attention
Max context: 32,768 tokens
Released: 2025-04-29
Model card: Qwen3 32B on Hugging Face

Other sizes in this series

Graphics card memory each model needs at its baseline quant with 8k context, and its verdict on common graphics cards, smallest model first
Model	Parameters	Baseline quant	Memory needed (8k)	GeForce RTX 4060	GeForce RTX 5060 Ti 16GB	GeForce RTX 3090
Qwen3 8B	8.2B	Q4_K_M	6.5 GB	Runs great	Runs great	Runs great
Qwen3 14B	14.8B	Q4_K_M	10.5 GB	Runs slowly	Runs great	Runs great
Qwen3 30B-A3B (2507)	30.5B total · 3.3B active (MoE)	Q4_K_M	19.8 GB	Runs well	Runs well	Runs great
Qwen3 32Bthis model	32.8B	Q4_K_M	22.0 GB	Won't run	Runs slowly	Runs well

GGUF files

Tracked quantizations of Qwen3 32B, smallest first
Quant	File size	Quality	Published by
Q2_K	12.34 GB	Heavy loss	Community GGUF · bartowski
IQ4_XS	17.69 GB	Good balance	Community GGUF · bartowski
Q4_K_Mbaseline	19.76 GB	Good balance	Community GGUF · bartowski
Q8_0	34.82 GB	Near-lossless	Community GGUF · bartowski

Memory needed by quant and context

Graphics card memory in GB: weights loaded on the GPU + f16 KV cache + compute buffer. Add about 0.6 GB if the GPU also drives your display on Windows. Part of the file stays off the graphics card: llama.cpp keeps the input embedding table in system RAM. A Mac counts the whole file, so it needs a little more.
Quant	4k	8k	16k	32k
Q2_K	13.7	14.8	17.0	21.5
IQ4_XS	18.9	20.0	22.2	26.7
Q4_K_M	20.9	22.0	24.3	28.7
Q8_0	35.6	36.7	38.9	43.4

Which GPUs can run Qwen3 32B?

How to read the verdicts

Runs great: Fully on the GPU at 20 tok/s or more
Runs well: Fully on the GPU at 8–20 tok/s; MoE experts in system RAM at 20 tok/s or more; or at least 90% on the GPU at 8 tok/s or more
Runs slowly: 2–8 tok/s; CPU-only; less than 90% on the GPU; or MoE experts in system RAM below 20 tok/s
Won't run: Does not fit, or under 2 tok/s

Dense models split between GPU and CPU slow down sharply — the CPU side sets the pace. MoE models that keep only their experts in system RAM degrade far more gently.

Sort

Filter this list by budget, type and context in the calculator

Scroll sideways to see every column.

58 GPUs, Macs and CPU setups at Q4_K_M with 8k context, sorted by verdict and speed
Hardware	Verdict	Speed	Quant	Memory	Runs as	Context	Notes	Price
NVIDIA RTX PRO 6000 Blackwell	Runs great	est. 57.3 tok/s50.4–64.1calibrated estimate ±12%	Q4_K_M	22.6 / 96.0 GB	Full GPU	up to 32k	—	—
NVIDIA RTX PRO 6000 Blackwell Max-Q	Runs great	est. 57.3 tok/s50.4–64.1calibrated estimate ±12%	Q4_K_M	22.6 / 96.0 GB	Full GPU	up to 32k	—	—
GeForce RTX 5090	Runs greatQwen3 32B verdict on the GeForce RTX 5090	53.3 tok/s8k estimate 57.3 tok/s (50.4–64.1, calibrated estimate ±12%)measured (1 run, 512 context)	Q4_K_M	22.6 / 32.0 GB	Full GPU	up to 32k	—	$4,200 street (as of 2026-08-09)
GeForce RTX 4090	Runs greatQwen3 32B verdict on the GeForce RTX 4090	42.5 tok/s8k estimate 32.2 tok/s (28.3–36.1, calibrated estimate ±12%)measured (1 run, 512 context)	Q4_K_M	22.6 / 24.0 GB	Full GPU	up to 8k	—	$2,755 street (as of 2026-08-09)
Radeon RX 7900 XTX	Runs great	est. 33.3 tok/s29.3–37.3calibrated estimate ±12%	Q4_K_M	22.6 / 24.0 GB	Full GPU	up to 8k	—	$999 launch MSRP
GeForce RTX 3090 Ti	Runs great	est. 32.2 tok/s28.3–36.1calibrated estimate ±12%	Q4_K_M	22.6 / 24.0 GB	Full GPU	up to 8k	—	$1,999 launch MSRP
Apple M5 Ultra · 96 GB	Runs great	est. 32.0 tok/s25.6–38.3calibrated estimate ±20%	Q4_K_M	22.5 / 67.2 GB	Unified memory	up to 8k	—	—
GeForce RTX 3090	Runs wellQwen3 32B verdict on the GeForce RTX 3090	est. 29.9 tok/s26.3–33.5calibrated estimate ±12%	Q4_K_M	22.6 / 24.0 GB	Full GPU	up to 8k	Try IQ4_XS: Runs great	$1,050 street (as of 2026-08-09)
GeForce RTX 5090 Laptop	Runs well	est. 28.6 tok/s25.2–32.1calibrated estimate ±12%	Q4_K_M	22.6 / 24.0 GB	Full GPU	up to 8k	Try IQ4_XS: Runs great	—
Apple M3 Ultra · 96 GB	Runs well	26.7 tok/s8k estimate 21.3 tok/s (17.0–25.6, calibrated estimate ±20%)measured (1 run, 512 context)	Q4_K_M	22.5 / 67.2 GB	Unified memory	up to 32k	Q2_K (heavy quality loss): Runs great	—
Apple M5 Max (40-core GPU) · 48 GB	Runs well	est. 21.3 tok/s17.0–25.6calibrated estimate ±20%	Q4_K_M	22.5 / 33.6 GB	Unified memory	up to 32k	Q2_K (heavy quality loss): Runs great	—
Apple M2 Ultra · 64 GB	Runs well	est. 20.8 tok/s16.7–25.0calibrated estimate ±20%	Q4_K_M	22.5 / 44.8 GB	Unified memory	up to 32k	Q2_K (heavy quality loss): Runs great	—
Apple M4 Max (40-core GPU) · 48 GB	Runs well	est. 18.9 tok/s15.2–22.7calibrated estimate ±20%	Q4_K_M	22.5 / 33.6 GB	Unified memory	up to 32k	—	—
Apple M5 Max (32-core GPU) · 36 GB	Runs well	est. 16.0 tok/s12.8–19.1calibrated estimate ±20%	Q4_K_M	22.5 / 25.2 GB	Unified memory	up to 16k	—	—
Apple M4 Max (32-core GPU) · 36 GB	Runs well	est. 14.2 tok/s11.4–17.1calibrated estimate ±20%	Q4_K_M	22.5 / 25.2 GB	Unified memory	up to 16k	—	—
Apple M5 Pro · 48 GB	Runs slowly	est. 10.7 tok/s8.5–12.8calibrated estimate ±20%	Q4_K_M	22.5 / 33.6 GB	Unified memory	up to 32k	Q2_K (heavy quality loss): Runs well	—
NVIDIA DGX Spark · 128 GB	Runs slowly	10.7 tok/s8k estimate 9.7 tok/s (7.8–11.7, calibrated estimate ±20%)measured (2 runs, 512 context)	Q4_K_M	22.5 / 116.9 GB	Unified memory	up to 32k	Q2_K (heavy quality loss): Runs well	$3,999 launch MSRP (128 GB)
Ryzen AI Max+ 395 (Strix Halo) · 32 GB	Runs slowly	10.2 tok/s8k estimate 5.8 tok/s (4.1–7.6, theoretical estimate ±30%)measured (2 runs, 512 context)	Q4_K_M	21.9 GB RAM	CPU only	up to 16k	Q2_K (heavy quality loss): Runs well	$1,999 launch MSRP (128 GB)
Apple M4 Pro · 48 GB	Runs slowlyQwen3 32B verdict on the Apple M4 Pro	est. 9.5 tok/s7.6–11.4calibrated estimate ±20%	Q4_K_M	22.5 / 33.6 GB	Unified memory	up to 32k	Q2_K (heavy quality loss): Runs well	—
Radeon RX 7900 XT	Runs slowly	est. 9.4 tok/s7.5–11.2calibrated estimate ±20%	Q4_K_M	20.0 / 20.0 GB + 3.1 GB RAM	Partial offload	up to 16k	Try IQ4_XS: Runs well Q2_K (heavy quality loss): Runs great	$899 launch MSRP
GeForce RTX 5080	Runs slowly	est. 5.1 tok/s4.1–6.1calibrated estimate ±20%	Q4_K_M	16.0 / 16.0 GB + 7.1 GB RAM	Partial offload	up to 16k	Q2_K (heavy quality loss): Runs great	$1,256 street (as of 2026-08-09)
GeForce RTX 5070 Ti	Runs slowly	est. 5.1 tok/s4.0–6.1calibrated estimate ±20%	Q4_K_M	16.0 / 16.0 GB + 7.1 GB RAM	Partial offload	up to 16k	Q2_K (heavy quality loss): Runs great	$949 street (as of 2026-08-09)
GeForce RTX 5080 Laptop	Runs slowly	est. 5.1 tok/s4.0–6.1calibrated estimate ±20%	Q4_K_M	16.0 / 16.0 GB + 7.1 GB RAM	Partial offload	up to 16k	Q2_K (heavy quality loss): Runs great	—
GeForce RTX 4080 Super	Runs slowly	est. 4.9 tok/s3.9–5.9calibrated estimate ±20%	Q4_K_M	16.0 / 16.0 GB + 7.1 GB RAM	Partial offload	up to 16k	Q2_K (heavy quality loss): Runs great	$999 launch MSRP
GeForce RTX 4080	Runs slowly	est. 4.9 tok/s3.9–5.9calibrated estimate ±20%	Q4_K_M	16.0 / 16.0 GB + 7.1 GB RAM	Partial offload	up to 16k	Q2_K (heavy quality loss): Runs great	$1,199 launch MSRP
Radeon RX 9070	Runs slowly	est. 4.9 tok/s3.9–5.9calibrated estimate ±20%	Q4_K_M	16.0 / 16.0 GB + 7.1 GB RAM	Partial offload	up to 16k	Q2_K (heavy quality loss): Runs great	$639 street (as of 2026-08-09)
Radeon RX 9070 XT	Runs slowly	est. 4.9 tok/s3.9–5.9calibrated estimate ±20%	Q4_K_M	16.0 / 16.0 GB + 7.1 GB RAM	Partial offload	up to 16k	Q2_K (heavy quality loss): Runs great	$689 street (as of 2026-08-09)
Radeon RX 7800 XT	Runs slowly	est. 4.9 tok/s3.9–5.9calibrated estimate ±20%	Q4_K_M	16.0 / 16.0 GB + 7.1 GB RAM	Partial offload	up to 16k	Q2_K (heavy quality loss): Runs great	$499 launch MSRP
GeForce RTX 4070 Ti Super	Runs slowly	est. 4.9 tok/s3.9–5.8calibrated estimate ±20%	Q4_K_M	16.0 / 16.0 GB + 7.1 GB RAM	Partial offload	up to 16k	Q2_K (heavy quality loss): Runs great	$799 launch MSRP
GeForce RTX 4090 Laptop	Runs slowly	est. 4.8 tok/s3.8–5.7calibrated estimate ±20%	Q4_K_M	16.0 / 16.0 GB + 7.1 GB RAM	Partial offload	up to 16k	Q2_K (heavy quality loss): Runs well	—
GeForce RTX 5060 Ti 16GB	Runs slowlyQwen3 32B verdict on the GeForce RTX 5060 Ti 16GB	est. 4.5 tok/s3.6–5.5calibrated estimate ±20%	Q4_K_M	16.0 / 16.0 GB + 7.1 GB RAM	Partial offload	up to 16k	Q2_K (heavy quality loss): Runs well	$569 street (as of 2026-08-09)
Radeon RX 9060 XT 16GB	Runs slowly	est. 4.3 tok/s3.4–5.2calibrated estimate ±20%	Q4_K_M	16.0 / 16.0 GB + 7.1 GB RAM	Partial offload	up to 16k	Q2_K (heavy quality loss): Runs well	$449 street (as of 2026-08-09)
GeForce RTX 4060 Ti 16GB	Runs slowly	est. 4.1 tok/s3.3–4.9calibrated estimate ±20%	Q4_K_M	16.0 / 16.0 GB + 7.1 GB RAM	Partial offload	up to 16k	Q2_K (heavy quality loss): Runs well	$499 launch MSRP
Apple M6 · 32 GB	Runs slowly	est. 3.9 tok/s2.7–5.1theoretical estimate ±30%	Q4_K_M	21.9 GB RAM	CPU only	up to 16k	—	—
Apple M5 · 32 GB	Runs slowly	est. 3.5 tok/s2.5–4.6theoretical estimate ±30%	Q4_K_M	21.9 GB RAM	CPU only	up to 16k	—	—
GeForce RTX 3080 12GB	Runs slowly	est. 3.5 tok/s2.8–4.1calibrated estimate ±20%	Q4_K_M	12.0 / 12.0 GB + 11.1 GB RAM	Partial offload	up to 8k	—	$799 launch MSRP
GeForce RTX 5070	Runs slowlyQwen3 32B verdict on the GeForce RTX 5070	est. 3.4 tok/s2.7–4.1calibrated estimate ±20%	Q4_K_M	12.0 / 12.0 GB + 11.1 GB RAM	Partial offload	up to 8k	—	$629 street (as of 2026-08-09)
GeForce RTX 5070 Ti Laptop	Runs slowly	est. 3.4 tok/s2.7–4.1calibrated estimate ±20%	Q4_K_M	12.0 / 12.0 GB + 11.1 GB RAM	Partial offload	up to 8k	—	—
GeForce RTX 4070	Runs slowlyQwen3 32B verdict on the GeForce RTX 4070	est. 3.3 tok/s2.7–4.0calibrated estimate ±20%	Q4_K_M	12.0 / 12.0 GB + 11.1 GB RAM	Partial offload	up to 8k	—	$599 launch MSRP
GeForce RTX 4070 Super	Runs slowly	est. 3.3 tok/s2.7–4.0calibrated estimate ±20%	Q4_K_M	12.0 / 12.0 GB + 11.1 GB RAM	Partial offload	up to 8k	—	$599 launch MSRP
GeForce RTX 4080 Laptop	Runs slowly	est. 3.3 tok/s2.6–3.9calibrated estimate ±20%	Q4_K_M	12.0 / 12.0 GB + 11.1 GB RAM	Partial offload	up to 8k	—	—
Intel Arc B580	Runs slowly	est. 3.2 tok/s2.6–3.9calibrated estimate ±20%	Q4_K_M	12.0 / 12.0 GB + 11.1 GB RAM	Partial offload	up to 8k	—	$290 street (as of 2026-08-09)
GeForce RTX 3060 12GB	Runs slowlyQwen3 32B verdict on the GeForce RTX 3060 12GB	est. 3.2 tok/s2.6–3.8calibrated estimate ±20%	Q4_K_M	12.0 / 12.0 GB + 11.1 GB RAM	Partial offload	up to 8k	—	$250 street (as of 2026-08-09)
GeForce RTX 2080 Ti	Runs slowly	est. 3.1 tok/s2.5–3.8calibrated estimate ±20%	Q4_K_M	11.0 / 11.0 GB + 12.1 GB RAM	Partial offload	up to 8k	—	$999 launch MSRP
GeForce GTX 1080 Ti	Runs slowly	est. 3.1 tok/s2.2–4.0theoretical estimate ±30%	Q4_K_M	11.0 / 11.0 GB + 12.1 GB RAM	Partial offload	up to 8k	—	$699 launch MSRP
Apple M4 · 32 GB	Won't runTry IQ4_XS: Runs slowlyQwen3 32B verdict on the Apple M4	est. 4.6 tok/s3.7–5.5calibrated estimate ±20%	IQ4_XS	20.4 / 22.4 GB	Unified memory	up to 8k	—	—
GeForce RTX 3070	Won't runTry Q2_K (heavy quality loss): Runs slowly	est. 4.6 tok/s3.7–5.5calibrated estimate ±20%	Q2_K	8.0 / 8.0 GB + 7.7 GB RAM	Partial offload	up to 16k	—	$499 launch MSRP
GeForce RTX 5060	Won't runTry Q2_K (heavy quality loss): Runs slowly	est. 4.6 tok/s3.7–5.5calibrated estimate ±20%	Q2_K	8.0 / 8.0 GB + 7.7 GB RAM	Partial offload	up to 16k	—	$339 street (as of 2026-08-09)
GeForce RTX 5060 Ti 8GB	Won't runTry Q2_K (heavy quality loss): Runs slowly	est. 4.6 tok/s3.7–5.5calibrated estimate ±20%	Q2_K	8.0 / 8.0 GB + 7.7 GB RAM	Partial offload	up to 16k	—	$429 street (as of 2026-08-09)
GeForce RTX 4060 Ti 8GB	Won't runTry Q2_K (heavy quality loss): Runs slowly	est. 4.4 tok/s3.5–5.3calibrated estimate ±20%	Q2_K	8.0 / 8.0 GB + 7.7 GB RAM	Partial offload	up to 16k	—	$399 launch MSRP
GeForce RTX 4060	Won't runTry Q2_K (heavy quality loss): Runs slowlyQwen3 32B verdict on the GeForce RTX 4060	est. 4.4 tok/s3.5–5.2calibrated estimate ±20%	Q2_K	8.0 / 8.0 GB + 7.7 GB RAM	Partial offload	up to 16k	—	$299 launch MSRP
GeForce RTX 4060 Laptop	Won't runTry Q2_K (heavy quality loss): Runs slowly	est. 4.3 tok/s3.5–5.2calibrated estimate ±20%	Q2_K	8.0 / 8.0 GB + 7.7 GB RAM	Partial offload	up to 16k	—	—
GeForce RTX 4070 Laptop	Won't runTry Q2_K (heavy quality loss): Runs slowly	est. 4.3 tok/s3.5–5.2calibrated estimate ±20%	Q2_K	8.0 / 8.0 GB + 7.7 GB RAM	Partial offload	up to 16k	—	—
GeForce RTX 3050 8GB	Won't runTry Q2_K (heavy quality loss): Runs slowly	est. 4.2 tok/s3.4–5.1calibrated estimate ±20%	Q2_K	8.0 / 8.0 GB + 7.7 GB RAM	Partial offload	up to 16k	—	$249 launch MSRP
GeForce RTX 3080 10GB	Won't runTry IQ4_XS: Runs slowly	est. 3.5 tok/s2.8–4.1calibrated estimate ±20%	IQ4_XS	10.0 / 10.0 GB + 11.0 GB RAM	Partial offload	up to 8k	—	$699 launch MSRP
DDR5-6000 dual-channel CPU · 96 GB	Won't runTry Q2_K (heavy quality loss): Runs slowly	est. 3.3 tok/s2.7–4.0calibrated estimate ±20%	Q2_K	14.5 GB RAM	CPU only	up to 8k	—	—
DDR5-5600 dual-channel CPU · 96 GB	Won't runTry Q2_K (heavy quality loss): Runs slowly	est. 3.1 tok/s2.5–3.7calibrated estimate ±20%	Q2_K	14.5 GB RAM	CPU only	up to 8k	—	—
DDR4-3200 dual-channel CPU · 64 GB	Won't run	est. 1.2 tok/s0.9–1.4calibrated estimate ±20%	Q4_K_M	21.9 GB RAM	CPU only	—	Fits, but too slow to use	—

Models with a thinking mode spend extra tokens thinking, so their speed thresholds are 1.5× stricter (30 / 12 / 3 tok/s). With thinking turned off, a model may do one grade better.

Cheapest GPUs that run it

Runs greatGeForce RTX 4090$2,755 street (as of 2026-08-09), 42.5 tok/s measured at 512 context (1 run; 8k estimate 32.2 tok/s, 28.3–36.1, calibrated estimate ±12%).
Runs wellGeForce RTX 3090$1,050 street (as of 2026-08-09), est. 29.9 tok/s (26.3–33.5, calibrated estimate ±12%).

Measured results for Qwen3 32B

Public benchmarks we calibrate against. Their conditions (context, backend, flags) can differ from the estimates above.

Measured results for Qwen3 32B
Hardware	Quant	Backend	Context	Prompt (tok/s)	Generation (tok/s)	Flags	Source	Measured
GeForce RTX 5090	Q4_K_XL	llama.cpp	4k	2,931.0	61.4	CUDA	hardware-corner.net	2025-12-09
GeForce RTX 4090	Q4_K_XL	llama.cpp	4k	2,448.0	39.6	CUDA	hardware-corner.net	2026-03-01
GeForce RTX 3090	Q4_K_XL	llama.cpp	4k	1,088.0	35.1	CUDA	hardware-corner.net	2025-12-09
GeForce RTX 3090 Ti	Q4_K_XL	llama.cpp	4k	1,239.0	38.0	CUDA	hardware-corner.net	2025-12-09
GeForce RTX 5090	Q4_K_M	llama.cpp	512	3,446.0	53.3	CUDA	github.com	2026-09-15
GeForce RTX 4090	Q4_K_M	llama.cpp	512	2,302.0	42.5	Vulkan	github.com	2025-12-09
Ryzen AI Max+ 395 (Strix Halo)	Q4_K_M	llama.cpp	512	256.0	9.9	ROCm	blog.tosukui.xyz	2025-07-13
Ryzen AI Max+ 395 (Strix Halo)	Q4_K_M	llama.cpp	512	140.0	10.4	Vulkan	blog.tosukui.xyz	2025-05-16
Ryzen AI Max+ 395 (Strix Halo)	Q8_0	llama.cpp	512	77.0	6.4	ROCm	llm-tracker.info	2025-05-11
NVIDIA DGX Spark	Q4_K_M	llama.cpp	512	630.0	10.5	CUDA, Vulkan	github.com	2025-11-22
NVIDIA DGX Spark	Q4_K_M	llama.cpp	512	762.0	10.8	CUDA	github.com	2025-12-05
Apple M3 Ultra	Q4_K_M	llama.cpp	512	335.0	26.7	Metal	github.com	2025-12-08

Frequently asked questions

How much VRAM does Qwen3 32B need?

At Q4_K_M the file is 19.76 GB; with 8k context a graphics card needs about 22.0 GB (weights 19.32 GB, KV cache 2.1 GB, compute buffer 0.6 GB). The smallest tracked file (Q2_K (heavy quality loss)) needs about 14.8 GB. Part of the file stays off the graphics card: llama.cpp keeps the input embedding table in system RAM. A Mac counts the whole file, so it needs a little more.

What is the cheapest GPU that runs Qwen3 32B well?

Cheapest that runs it great: GeForce RTX 4090 at $2,755 (street price as of 2026-08-09), 42.5 tok/s measured at 512 context (1 run; 8k estimate 32.2 tok/s, 28.3–36.1, calibrated estimate ±12%). Cheaper and still Runs well: GeForce RTX 3090 at $1,050 (street price as of 2026-08-09), est. 29.9 tok/s (26.3–33.5, calibrated estimate ±12%).

Can Qwen3 32B run on an 8 GB, 16 GB or 24 GB GPU?

At Q4_K_M with 8k context: GeForce RTX 4060 — Won't run, est. 2.5 tok/s (2.0–3.0, calibrated estimate ±20%); GeForce RTX 5060 Ti 16GB — Runs slowly, est. 4.5 tok/s (3.6–5.5, calibrated estimate ±20%); GeForce RTX 3090 — Runs well, est. 29.9 tok/s (26.3–33.5, calibrated estimate ±12%).

Speeds are estimates from memory bandwidth, calibrated against public benchmarks, and each one comes with an error band and a confidence label. Real results vary with drivers, backend, context length and thermals.