How much VRAM does grok-2 need?

Q4_K_M is exactly 164,055,568,672 bytes (152.79 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is grok-2'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.

Is grok-2 a mixture-of-experts model?

Yes — 8 experts with 2 routed per token. Every expert must be resident in memory, but only the routed ones are read per token, so its memory requirement and its speed behave very differently.

xai-org · text · mixture of experts

grok-2

xai-org/grok-2

grok-2 at Q4_K_M is exactly 164,055,568,672 bytes (152.79 GiB / 164.06 GB) — an effective 4.870 bits per weight, not the nominal 4. Its KV cache at 32K is 8.00 GiB.

From the file· summed from 4 file(s)From the file· KV per layer

Parameters

270B

total, not active

Architecture

grok

64 layers

Context

131,072

native (config.json)

License

other

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ1_M2 shards	57.16 GiB	61,379,006,624	1.822	—	bartowski
IQ2_XXS2 shards	63.81 GiB	68,517,712,064	2.034	—	bartowski
IQ2_XS2 shards	72.40 GiB	77,736,792,256	2.308	—	bartowski
IQ2_S2 shards	73.30 GiB	78,709,870,752	2.336	—	bartowski
UD-TQ1_0	76.17 GiB	81,784,426,368	2.428	—	oki692
UD-TQ1_0	76.17 GiB	81,784,426,368	2.428	—	unsloth
IQ2_M3 shards	82.15 GiB	88,205,775,136	2.618	—	bartowski
UD-IQ1_S2 shards	82.82 GiB	88,931,520,576	2.640	—	unsloth
UD-IQ1_S2 shards	82.82 GiB	88,931,520,576	2.640	—	oki692
UD-IQ1_M2 shards	88.02 GiB	94,513,483,840	2.805	—	unsloth
UD-IQ1_M2 shards	88.02 GiB	94,513,483,840	2.805	—	oki692
Q2_K3 shards	89.93 GiB	96,560,828,736	2.866	—	bartowski
Q2_K_L3 shards	90.91 GiB	97,609,404,736	2.897	—	bartowski
UD-IQ2_XXS2 shards	92.52 GiB	99,346,894,912	2.949	—	unsloth
UD-IQ2_XXS2 shards	92.52 GiB	99,346,894,912	2.949	—	oki692
Q2_K3 shards	93.18 GiB	100,050,489,536	2.970	—	oki692
Q2_K3 shards	93.18 GiB	100,050,489,536	2.970	—	unsloth
Q2_K_L3 shards	93.41 GiB	100,302,147,776	2.977	—	unsloth
Q2_K_L3 shards	93.41 GiB	100,302,147,776	2.977	—	oki692
UD-IQ2_M3 shards	96.02 GiB	103,104,991,424	3.061	—	oki692
UD-IQ2_M3 shards	96.02 GiB	103,104,991,424	3.061	—	unsloth
IQ3_XXS3 shards	99.62 GiB	106,962,702,656	3.175	—	bartowski
IQ3_XS3 shards	104.12 GiB	111,802,929,472	3.319	—	bartowski
UD-IQ3_XXS3 shards	109.27 GiB	117,330,497,728	3.483	—	unsloth
UD-IQ3_XXS3 shards	109.27 GiB	117,330,497,728	3.483	—	oki692
Q3_K_S3 shards	109.94 GiB	118,044,053,696	3.504	—	unsloth
Q3_K_S3 shards	109.94 GiB	118,044,053,696	3.504	—	oki692
Q3_K_S3 shards	109.94 GiB	118,044,053,824	3.504	—	bartowski
IQ3_M4 shards	115.25 GiB	123,748,307,392	3.673	—	bartowski
Q3_K_M4 shards	116.44 GiB	125,023,375,776	3.711	—	bartowski
Q3_K_M3 shards	121.19 GiB	130,123,649,216	3.862	—	unsloth
Q3_K_M3 shards	121.19 GiB	130,123,649,216	3.862	—	oki692
Q3_K_L4 shards	121.28 GiB	130,224,312,736	3.865	—	bartowski
IQ4_XS4 shards	134.82 GiB	144,761,770,432	4.297	—	bartowski
IQ4_XS3 shards	134.85 GiB	144,795,324,608	4.298	—	oki692
IQ4_XS3 shards	134.85 GiB	144,795,324,608	4.298	—	unsloth
IQ4_NL4 shards	142.48 GiB	152,982,606,144	4.541	—	unsloth
IQ4_NL4 shards	142.48 GiB	152,982,606,144	4.541	—	oki692
IQ4_NL4 shards	142.48 GiB	152,982,606,272	4.541	—	bartowski
Q4_04 shards	142.60 GiB	153,116,823,872	4.545	—	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 DeepSeek-V4-Flash vs Qwen3-30B-A3B-Thinking-2507 vs Laguna-XS-2.1 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 141.19 GiB. The real file is 152.79 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

8192

Vocab

131,072

Sliding window

none

SWA period

—

MLA

Experts

Experts per token

use_sliding_window

—

Questions people ask

How much VRAM does grok-2 need?: Q4_K_M is exactly 164,055,568,672 bytes (152.79 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is grok-2'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.
Is grok-2 a mixture-of-experts model?: Yes — 8 experts, 2 routed per token. Every expert must be resident, but only the routed ones are read per token, which is why its memory requirement and its speed behave very differently.
Which quantization of grok-2 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.