How much VRAM does Qwen2.5-32B-Instruct need?

Q4_K_M is exactly 19,851,336,288 bytes (18.49 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Qwen2.5-32B-Instruct'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

Qwen2.5-32B-Instruct

Qwen/Qwen2.5-32B-Instruct

Qwen2.5-32B-Instruct at Q4_K_M is exactly 19,851,336,288 bytes (18.49 GiB / 19.85 GB) — an effective 4.847 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

qwen2

64 layers

Context

32,768

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ2_XXS	8.41 GiB	9,028,250,496	2.204	—	bartowski
IQ2_XS	9.27 GiB	9,957,550,976	2.431	—	bartowski
IQ2_S	9.67 GiB	10,387,569,536	2.536	—	bartowski
IQ2_M	10.49 GiB	11,264,441,216	2.751	—	bartowski
Q2_K4 shards	11.47 GiB	12,313,098,784	3.006	—	Qwen
Q2_K	11.47 GiB	12,313,099,136	3.006	—	bartowski
Q2_K_L	12.18 GiB	13,073,419,136	3.192	—	bartowski
IQ3_XS	12.76 GiB	13,705,513,856	3.346	—	bartowski
Q3_K_S	13.40 GiB	14,392,331,136	3.514	—	bartowski
IQ3_M	13.79 GiB	14,810,123,136	3.616	—	bartowski
Q3_K_M5 shards	14.84 GiB	15,935,048,320	3.891	—	Qwen
Q3_K_M	14.84 GiB	15,935,048,576	3.891	771	bartowski
Q3_K_L	16.06 GiB	17,247,079,008	4.211	—	lmstudio-community
Q3_K_L	16.06 GiB	17,247,079,296	4.211	—	bartowski
IQ4_XS	16.48 GiB	17,693,154,176	4.320	771	bartowski
Q4_05 shards	17.36 GiB	18,640,231,072	4.551	—	Qwen
Q4_0	17.43 GiB	18,711,010,176	4.569	771	bartowski
Q4_K_S	17.49 GiB	18,784,410,496	4.587	—	bartowski
Q4_K_M	18.49 GiB	19,851,336,288	4.847	771	lmstudio-community
Q4_K_M5 shards	18.49 GiB	19,851,336,384	4.847	—	Qwen
Q4_K_M	18.49 GiB	19,851,336,576	4.847	771	bartowski
Q4_K_L	19.03 GiB	20,429,179,776	4.988	—	bartowski
Q5_06 shards	21.08 GiB	22,638,254,880	5.528	—	Qwen
Q5_K_S	21.08 GiB	22,638,254,976	5.528	—	bartowski
Q5_K_M6 shards	21.66 GiB	23,262,157,568	5.680	—	Qwen
Q5_K_M	21.66 GiB	23,262,157,696	5.680	771	bartowski
Q5_K_L	22.11 GiB	23,742,679,936	5.797	—	bartowski
Q6_K	25.04 GiB	26,886,154,848	6.565	771	lmstudio-community
Q6_K7 shards	25.04 GiB	26,886,155,104	6.565	—	Qwen
Q6_K	25.04 GiB	26,886,155,136	6.565	771	bartowski
Q6_K_L	25.39 GiB	27,263,273,856	6.657	—	bartowski
Q8_0	32.43 GiB	34,820,885,088	8.502	771	lmstudio-community
Q8_0	32.43 GiB	34,820,885,376	8.502	771	bartowski
Q8_09 shards	32.43 GiB	34,820,885,632	8.502	—	Qwen
F162 shards	61.04 GiB	65,535,970,112	16.002	—	bartowski

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 18.49 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

152,064

Sliding window

131072

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

false

This model declares a sliding window but sets use_sliding_window: false, so the window is not applied. Honouring the field without the flag understates KV for the whole family.

Questions people ask

How much VRAM does Qwen2.5-32B-Instruct need?: Q4_K_M is exactly 19,851,336,288 bytes (18.49 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Qwen2.5-32B-Instruct'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 Qwen2.5-32B-Instruct 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.