How much VRAM does DeepSeek-R1-Distill-Qwen-14B need?

Q4_K_M is exactly 8,988,109,952 bytes (8.37 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is DeepSeek-R1-Distill-Qwen-14B's KV cache?

6.00 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.

deepseek-ai · text

DeepSeek-R1-Distill-Qwen-14B

deepseek-ai/DeepSeek-R1-Distill-Qwen-14B

DeepSeek-R1-Distill-Qwen-14B at Q4_K_M is exactly 8,988,109,952 bytes (8.37 GiB / 8.99 GB) — an effective 4.868 bits per weight, not the nominal 4. Its KV cache at 32K is 6.00 GiB.

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

Parameters

14.8B

Architecture

qwen2

48 layers

Context

131,072

native (config.json)

License

—

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ2_XS	4.38 GiB	4,704,574,880	2.548	—	bartowski
IQ2_S	4.66 GiB	5,003,726,240	2.710	—	bartowski
IQ2_M	4.99 GiB	5,356,146,080	2.901	—	bartowski
Q2_K	5.37 GiB	5,770,497,184	3.126	—	unsloth
Q2_K	5.37 GiB	5,770,497,440	3.126	—	bartowski
Q2_K_L	5.54 GiB	5,952,973,984	3.224	—	unsloth
IQ3_XS	5.94 GiB	6,383,361,440	3.458	—	bartowski
Q2_K_L	6.08 GiB	6,530,817,440	3.537	—	bartowski
Q3_K_S	6.20 GiB	6,659,595,680	3.607	—	bartowski
IQ3_M	6.44 GiB	6,916,537,760	3.746	—	bartowski
Q3_K_M	6.84 GiB	7,339,203,744	3.975	579	unsloth
Q3_K_M	6.84 GiB	7,339,204,000	3.975	—	bartowski
Q3_K_L	7.38 GiB	7,924,767,872	4.292	—	lmstudio-community
Q3_K_L	7.38 GiB	7,924,768,160	4.292	—	bartowski
IQ4_XS	7.56 GiB	8,119,840,160	4.398	579	bartowski
Q4_0	7.96 GiB	8,544,267,680	4.628	579	bartowski
IQ4_NL	7.96 GiB	8,549,182,880	4.631	—	bartowski
Q4_K_S	7.98 GiB	8,573,431,200	4.644	—	bartowski
Q4_K_M	8.37 GiB	8,988,109,952	4.868	579	lmstudio-community
Q4_K_M	8.37 GiB	8,988,109,984	4.868	—	unsloth
Q4_K_M	8.37 GiB	8,988,110,240	4.868	579	bartowski
Q4_1	8.75 GiB	9,392,139,680	5.087	—	bartowski
Q4_K_L	8.91 GiB	9,565,953,440	5.181	—	bartowski
Q5_K_S	9.56 GiB	10,266,553,760	5.561	—	bartowski
Q5_K_M	9.79 GiB	10,508,872,864	5.692	579	unsloth
Q5_K_M	9.79 GiB	10,508,873,120	5.692	—	bartowski
Q5_K_L	10.23 GiB	10,989,395,360	5.952	—	bartowski
Q6_K	11.29 GiB	12,124,683,392	6.567	579	lmstudio-community
Q6_K	11.29 GiB	12,124,683,424	6.567	579	unsloth
Q6_K	11.29 GiB	12,124,683,680	6.567	—	bartowski
Q6_K_L	11.64 GiB	12,501,802,400	6.771	—	bartowski
Q8_0	14.62 GiB	15,701,597,312	8.505	579	lmstudio-community
Q8_0	14.62 GiB	15,701,597,344	8.505	579	unsloth
Q8_0	14.62 GiB	15,701,597,600	8.505	—	bartowski
F16	27.52 GiB	29,547,715,744	16.004	579	unsloth
F16	27.52 GiB	29,547,716,000	16.004	579	bartowski
F322 shards	55.03 GiB	59,086,101,824	32.003	—	bartowski

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.75 GiB	0.75 GiB	—	48 / 0 / 0
8,192	1.50 GiB	1.50 GiB	—	48 / 0 / 0
16,384	3.00 GiB	3.00 GiB	—	48 / 0 / 0
32,768	6.00 GiB	6.00 GiB	—	48 / 0 / 0
65,536	12.00 GiB	12.00 GiB	—	48 / 0 / 0
131,072	24.00 GiB	24.00 GiB	—	48 / 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 7.74 GiB. The real file is 8.37 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 DeepSeek-R1-Distill-Qwen-14B need?: Q4_K_M is exactly 8,988,109,952 bytes (8.37 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is DeepSeek-R1-Distill-Qwen-14B's KV cache?: 6.00 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 DeepSeek-R1-Distill-Qwen-14B 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.