How much VRAM does Llama-3.1-8B need?

Q4_K_M is exactly 4,920,733,824 bytes (4.58 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Llama-3.1-8B's KV cache?

4.00 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.

meta-llama · text

Llama-3.1-8B

meta-llama/Llama-3.1-8B

Llama-3.1-8B at Q4_K_M is exactly 4,920,733,824 bytes (4.58 GiB / 4.92 GB) — an effective 4.902 bits per weight, not the nominal 4. Its KV cache at 32K is 4.00 GiB.

From the file· summed from 1 file(s)From the file· KV from mirror (mirror:unsloth/Llama-3.1-8B)

Parameters

8.0B

Architecture

llama

32 layers

Context

131,072

native (config.json)

License

llama3.1

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
Q2_K	2.96 GiB	3,179,141,120	3.167	—	dphn
Q3_K_S	3.41 GiB	3,664,509,760	3.651	—	dphn
Q3_K_M	3.74 GiB	4,018,928,448	4.004	292	dphn
Q3_K	3.74 GiB	4,018,928,448	4.004	—	dphn
Q3_K_L	4.03 GiB	4,321,966,912	4.306	—	dphn
Q4_0	4.34 GiB	4,661,223,296	4.644	292	dphn
Q4_K_S	4.37 GiB	4,692,680,576	4.675	—	dphn
Q4_K_M	4.58 GiB	4,920,733,824	4.902	292	NousResearch
Q4_K_M	4.58 GiB	4,920,734,336	4.902	292	Delentia
Q4_K_M	4.58 GiB	4,920,745,856	4.902	292	dphn
Q4_K	4.58 GiB	4,920,745,856	4.902	—	dphn
Q4_1	4.78 GiB	5,130,264,960	5.111	—	dphn
Q5_0	5.21 GiB	5,599,306,624	5.578	—	dphn
Q5_K_S	5.21 GiB	5,599,306,624	5.578	—	dphn
Q5_K_M	5.34 GiB	5,732,987,008	5.711	292	NousResearch
Q5_K	5.34 GiB	5,733,000,064	5.711	—	dphn
Q5_K_M	5.34 GiB	5,733,000,064	5.711	292	dphn
Q5_1	5.65 GiB	6,068,348,288	6.045	—	dphn
Q6_K	6.14 GiB	6,596,006,016	6.571	292	NousResearch
Q6_K	6.14 GiB	6,596,020,160	6.571	292	dphn
Q8_0	7.95 GiB	8,540,770,432	8.509	292	NousResearch
Q8_0	7.95 GiB	8,540,770,944	8.509	—	Delentia
Q8_0	7.95 GiB	8,540,788,544	8.509	292	dphn
Q4_K_M3 shards	13.75 GiB	14,762,205,536	14.707	—	Delentia
F16	14.97 GiB	16,068,924,224	16.008	292	dphn

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.50 GiB	0.50 GiB	—	32 / 0 / 0
8,192	1.00 GiB	1.00 GiB	—	32 / 0 / 0
16,384	2.00 GiB	2.00 GiB	—	32 / 0 / 0
32,768	4.00 GiB	4.00 GiB	—	32 / 0 / 0
65,536	8.00 GiB	8.00 GiB	—	32 / 0 / 0
131,072	16.00 GiB	16.00 GiB	—	32 / 0 / 0

Compare with

same modality, comparable size

vs gemma-4-12B-it-qat-q4_0-unquantized vs gemma-4-E4B-it vs Qwen3-30B-A3B-Thinking-2507 vs Qwen3-4B vs Qwen3-8B vs Llama-3.2-1B-Instruct 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 4.21 GiB. The real file is 4.58 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from mirror:unsloth/Llama-3.1-8B

Layers

Attention heads

KV heads

Head dim

128

Hidden size

4096

Vocab

128,256

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does Llama-3.1-8B need?: Q4_K_M is exactly 4,920,733,824 bytes (4.58 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Llama-3.1-8B's KV cache?: 4.00 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 Llama-3.1-8B 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.