How much VRAM does Qwen3-32B need?

Q4_K_M is exactly 19,762,149,024 bytes (18.40 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Qwen3-32B's KV cache?

8.00 GiB at 32K context with an f16 cache, computed per layer from the model's architecture. Quantizing the cache to q8_0 roughly halves it.

Qwen · text

Qwen3-32B

Qwen/Qwen3-32B

Qwen3-32B at Q4_K_M is exactly 19,762,149,024 bytes (18.40 GiB / 19.76 GB) — an effective 4.826 bits per weight, not the nominal 4. Its KV cache at 32K is 8.00 GiB.

From the file· summed from 1 file(s)From the file· KV per layer

Parameters

32.8B

Architecture

qwen3

64 layers

Context

40,960

native (config.json)

License

—

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
UD-IQ1_S	7.21 GiB	7,738,884,768	1.890	—	unsloth
UD-IQ1_M	7.73 GiB	8,302,084,768	2.027	—	unsloth
UD-IQ2_XXS	8.64 GiB	9,277,096,608	2.265	—	unsloth
UD-IQ2_M	10.76 GiB	11,553,366,688	2.821	—	unsloth
Q2_K	11.50 GiB	12,344,652,448	3.014	—	unsloth
Q2_K_L	11.67 GiB	12,526,975,648	3.059	—	unsloth
UD-IQ3_XXS	12.08 GiB	12,967,551,648	3.167	—	unsloth
Q3_K_S	13.40 GiB	14,389,739,168	3.514	—	unsloth
Q3_K_M	14.87 GiB	15,971,778,208	3.900	707	unsloth
Q3_K_L	16.14 GiB	17,330,994,240	4.232	—	lmstudio-community
IQ4_XS	16.50 GiB	17,714,805,408	4.326	707	unsloth
IQ4_NL	17.40 GiB	18,679,495,328	4.561	—	unsloth
Q4_0	17.42 GiB	18,703,088,288	4.567	707	unsloth
Q4_K_S	17.48 GiB	18,771,245,728	4.584	—	unsloth
Q4_K_M	18.40 GiB	19,762,149,024	4.826	707	Qwen
Q4_K_M	18.40 GiB	19,762,149,440	4.826	—	lmstudio-community
Q4_K_M	18.40 GiB	19,762,150,048	4.826	707	unsloth
Q4_1	19.22 GiB	20,636,523,168	5.039	—	unsloth
Q5_0	21.08 GiB	22,635,493,024	5.527	—	Qwen
Q5_K_S	21.08 GiB	22,635,494,048	5.527	—	unsloth
Q5_K_M	21.62 GiB	23,214,831,232	5.669	707	Qwen
Q5_K_M	21.62 GiB	23,214,832,288	5.669	707	unsloth
Q6_K	25.04 GiB	26,883,306,112	6.564	707	Qwen
Q6_K	25.04 GiB	26,883,306,560	6.564	—	lmstudio-community
Q6_K	25.04 GiB	26,883,307,168	6.564	707	unsloth
Q8_0	32.43 GiB	34,817,718,912	8.502	707	Qwen
Q8_0	32.43 GiB	34,817,719,360	8.502	—	lmstudio-community
Q8_0	32.43 GiB	34,817,719,968	8.502	707	unsloth
BF162 shards	61.03 GiB	65,531,575,936	16.002	—	unsloth

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	1.00 GiB	1.00 GiB	—	64 / 0 / 0
8,192	2.00 GiB	2.00 GiB	—	64 / 0 / 0
16,384	4.00 GiB	4.00 GiB	—	64 / 0 / 0
32,768	8.00 GiB	8.00 GiB	—	64 / 0 / 0
65,536	16.00 GiB	16.00 GiB	—	64 / 0 / 0
131,072	32.00 GiB	32.00 GiB	—	64 / 0 / 0

Compare with

same modality, comparable size

vs Qwen3-Coder-30B-A3B-Instruct vs Qwen3.6-27B vs Qwen3.8-27B vs gemma-4-12B-it-qat-q4_0-unquantized vs gemma-4-E4B-it vs Qwen3-30B-A3B-Thinking-2507 All alternatives →Will it fit on my card? →

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 17.16 GiB. The real file is 18.40 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json

Layers

Attention heads

KV heads

Head dim

128

Hidden size

5120

Vocab

151,936

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

false

Questions people ask

How much VRAM does Qwen3-32B need?: Q4_K_M is exactly 19,762,149,024 bytes (18.40 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Qwen3-32B's KV cache?: 8.00 GiB at 32K context with an f16 cache, computed per layer. Quantizing the cache to q8_0 roughly halves it, which is often the difference between a context length fitting and not.
Which quantization of Qwen3-32B should I use?: Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.