How much VRAM does Dolphin3.0-R1-Mistral-24B need?

Q4_K_M is exactly 13,070,715,520 bytes (12.17 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Dolphin3.0-R1-Mistral-24B's KV cache?

5.00 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.

dphn · text

Dolphin3.0-R1-Mistral-24B

dphn/Dolphin3.0-R1-Mistral-24B

Dolphin3.0-R1-Mistral-24B at Q4_K_M is exactly 13,070,715,520 bytes (12.17 GiB / 13.07 GB) — an effective 4.436 bits per weight, not the nominal 4. Its KV cache at 32K is 5.00 GiB.

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

Parameters

23.6B

Architecture

llama

40 layers

Context

32,768

native (config.json)

License

—

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ2_XXS	6.10 GiB	6,545,132,352	2.221	—	bartowski
IQ2_XS	6.71 GiB	7,207,045,952	2.446	—	bartowski
IQ2_S	6.96 GiB	7,478,366,048	2.538	—	bartowski
IQ2_M	7.56 GiB	8,114,065,248	2.754	—	bartowski
Q2_K	8.28 GiB	8,890,339,488	3.017	—	bartowski
Q2_K_L	8.89 GiB	9,545,709,472	3.240	—	bartowski
Q3_K_S	9.11 GiB	9,780,805,920	3.319	—	eaddario
IQ3_XS	9.23 GiB	9,907,131,584	3.362	—	bartowski
IQ3_S	9.48 GiB	10,181,722,496	3.455	—	eaddario
Q3_K_S	9.69 GiB	10,400,289,984	3.530	—	bartowski
Q3_K_M	9.70 GiB	10,419,946,816	3.536	—	eaddario
IQ3_M	9.92 GiB	10,650,965,184	3.615	—	bartowski
IQ3_M	10.07 GiB	10,815,620,416	3.671	—	eaddario
Q3_K_L	10.21 GiB	10,958,161,216	3.719	—	eaddario
Q3_K_M	10.69 GiB	11,474,097,344	3.894	—	bartowski
Q3_K_L	11.55 GiB	12,400,776,384	4.209	—	bartowski
Q4_K_S	11.77 GiB	12,638,177,920	4.289	—	eaddario
IQ4_XS	11.88 GiB	12,758,931,648	4.330	—	bartowski
IQ4_NL	12.16 GiB	13,054,331,520	4.430	—	eaddario
Q4_K_M	12.17 GiB	13,070,715,520	4.436	—	eaddario
IQ4_NL	12.54 GiB	13,468,031,488	4.571	—	bartowski
Q4_0	12.57 GiB	13,494,245,888	4.580	—	bartowski
Q4_K_S	12.62 GiB	13,549,296,128	4.598	—	bartowski
Q4_K_M	13.35 GiB	14,333,925,888	4.865	—	bartowski
Q4_K_L	13.81 GiB	14,832,007,072	5.034	—	bartowski
Q4_1	13.85 GiB	14,873,123,968	5.048	—	bartowski
Q5_K_S	13.88 GiB	14,908,510,080	5.060	—	eaddario
Q5_K_M	14.51 GiB	15,583,940,480	5.289	—	eaddario
Q5_K_S	15.18 GiB	16,304,430,848	5.533	—	bartowski
Q5_K_M	15.61 GiB	16,764,002,048	5.689	—	bartowski
Q5_K_L	16.00 GiB	17,178,195,872	5.830	—	bartowski
Q6_K	17.96 GiB	19,280,418,016	6.543	—	eaddario
Q6_K	18.02 GiB	19,345,957,984	6.566	—	bartowski
Q6_K_L	18.32 GiB	19,671,021,472	6.676	—	bartowski
Q8_0	21.40 GiB	22,976,159,360	7.798	—	eaddario
Q8_0	23.33 GiB	25,054,803,872	8.503	—	bartowski
F16	43.92 GiB	47,153,562,016	16.003	—	eaddario
F323 shards	87.82 GiB	94,297,580,224	32.003	—	bartowski

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.63 GiB	0.63 GiB	—	40 / 0 / 0
8,192	1.25 GiB	1.25 GiB	—	40 / 0 / 0
16,384	2.50 GiB	2.50 GiB	—	40 / 0 / 0
32,768	5.00 GiB	5.00 GiB	—	40 / 0 / 0
65,536	10.00 GiB	10.00 GiB	—	40 / 0 / 0
131,072	20.00 GiB	20.00 GiB	—	40 / 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 12.35 GiB. The real file is 12.17 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

131,074

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does Dolphin3.0-R1-Mistral-24B need?: Q4_K_M is exactly 13,070,715,520 bytes (12.17 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Dolphin3.0-R1-Mistral-24B's KV cache?: 5.00 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 Dolphin3.0-R1-Mistral-24B 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.