How much VRAM does Qwen3-1.7B need?

Q4_K_M is exactly 1,107,409,472 bytes (1.03 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Qwen3-1.7B's KV cache?

3.50 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

Qwen3-1.7B

Qwen/Qwen3-1.7B

Qwen3-1.7B at Q4_K_M is exactly 1,107,409,472 bytes (1.03 GiB / 1.11 GB) — an effective 4.360 bits per weight, not the nominal 4. Its KV cache at 32K is 3.50 GiB.

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

Parameters

2.0B

Architecture

qwen3

28 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	0.50 GiB	537,829,952	2.118	—	unsloth
UD-IQ1_M	0.52 GiB	561,947,200	2.213	—	unsloth
UD-IQ2_XXS	0.56 GiB	605,790,784	2.385	—	unsloth
UD-IQ2_M	0.66 GiB	708,715,072	2.791	—	unsloth
UD-IQ3_XXS	0.71 GiB	765,174,336	3.013	—	unsloth
Q2_K	0.72 GiB	777,796,160	3.063	—	unsloth
Q2_K_L	0.72 GiB	777,796,160	3.063	—	unsloth
IQ2_M	0.77 GiB	828,885,408	3.264	—	bartowski
Q3_K_S	0.81 GiB	867,252,800	3.415	—	unsloth
Q2_K	0.82 GiB	879,896,992	3.465	—	bartowski
IQ3_XXS	0.83 GiB	888,064,416	3.497	—	bartowski
Q3_K_M	0.88 GiB	939,539,008	3.699	310	unsloth
IQ3_XS	0.90 GiB	967,926,176	3.811	—	bartowski
Q3_K_S	0.93 GiB	1,000,956,320	3.941	—	bartowski
IQ4_XS	0.94 GiB	1,010,383,424	3.978	310	unsloth
IQ3_M	0.96 GiB	1,029,366,176	4.053	—	bartowski
IQ4_NL	0.98 GiB	1,054,423,616	4.152	—	unsloth
Q4_0	0.98 GiB	1,056,782,912	4.161	310	unsloth
Q4_K_S	0.99 GiB	1,060,190,784	4.175	—	unsloth
Q3_K_M	1.00 GiB	1,073,242,528	4.226	—	bartowski
Q4_K_M	1.03 GiB	1,107,409,472	4.360	—	unsloth
Q3_K_L	1.06 GiB	1,137,205,408	4.478	—	lmstudio-community
Q3_K_L	1.06 GiB	1,137,205,664	4.478	—	bartowski
Q4_1	1.06 GiB	1,142,504,000	4.499	—	unsloth
IQ4_XS	1.09 GiB	1,175,689,632	4.629	311	bartowski
Q2_K_L	1.10 GiB	1,183,768,992	4.661	—	bartowski
IQ4_NL	1.15 GiB	1,229,453,728	4.841	—	bartowski
Q5_K_S	1.15 GiB	1,230,584,384	4.845	—	unsloth
Q4_0	1.15 GiB	1,231,813,024	4.850	—	bartowski
Q4_K_S	1.15 GiB	1,235,220,896	4.864	—	bartowski
Q5_K_M	1.17 GiB	1,257,880,128	4.953	310	unsloth
Q4_K_M	1.19 GiB	1,282,439,328	5.050	311	lmstudio-community
Q4_K_M	1.19 GiB	1,282,439,584	5.050	—	bartowski
Q4_1	1.25 GiB	1,336,981,920	5.264	—	bartowski
Q6_K	1.32 GiB	1,417,755,200	5.582	310	unsloth
Q5_K_S	1.35 GiB	1,444,510,112	5.688	—	bartowski
Q5_K_M	1.37 GiB	1,471,805,856	5.795	311	bartowski
Q4_K_L	1.41 GiB	1,513,382,304	5.959	—	bartowski
Q5_K_L	1.55 GiB	1,663,852,960	6.551	—	bartowski
Q6_K	1.56 GiB	1,673,007,264	6.588	311	lmstudio-community

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.44 GiB	0.44 GiB	—	28 / 0 / 0
8,192	0.88 GiB	0.88 GiB	—	28 / 0 / 0
16,384	1.75 GiB	1.75 GiB	—	28 / 0 / 0
32,768	3.50 GiB	3.50 GiB	—	28 / 0 / 0
65,536	7.00 GiB	7.00 GiB	—	28 / 0 / 0
131,072	14.00 GiB	14.00 GiB	—	28 / 0 / 0

Compare with

same modality, comparable size

vs Qwen3-4B vs HyperCLOVAX-SEED-Text-Instruct-1.5B vs Llama-3.2-1B-Instruct vs ced-base vs gemma-3-1b-it vs embeddinggemma-300m 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 1.06 GiB. The real file is 1.03 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

2048

Vocab

151,936

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

false

Questions people ask

How much VRAM does Qwen3-1.7B need?: Q4_K_M is exactly 1,107,409,472 bytes (1.03 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Qwen3-1.7B's KV cache?: 3.50 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 Qwen3-1.7B 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.