MiniMax-M2 VRAM & Memory Requirements

Specific checkpoints with sourced architecture and weight files. GGUF Q4_K_M (Unsloth): 130.90 GiB–133.99 GiB at 8,192 tokens, 1 sequence, FP16 KV, llama.cpp.

This result is an estimate, not a measured trace.Not verifiedFile-backed weights

Publisher and provenance

PublisherMiniMax
Hugging FaceMiniMaxAI/MiniMax-M2
Revision757303d492a50514c312788b5247a4f696a4c6a3
Total parameters228,689,764,864
Active parametersUnknown / not independently sourced
Layers62
Query heads48
KV heads8 (gqa)
Head dimension128
Recorded context capability196,608
Completenesscomplete · full

Active parameters describe tokens' routed compute, not the full resident model weight footprint. Full-resident weight estimates use total parameters.

Memory layers in this estimate

Artifact bytes, raw KV, engine extras, and total demand are separate. This is not a runtime trace.

Weight artifact: 128.84 GiB (138,342,385,120 bytes)
Raw KV cache: 1.94 GiB (2,080,374,784 bytes)
Engine extras / workspace: 0.13 GiB–3.21 GiB
Total estimated demand: 130.90 GiB–133.99 GiB

Published artifacts

Quantization	Selected files	Bytes	Engine	Evidence
FP8 safetensors fp8 Calculator: FP8 safetensors	125 safetensors shards Total selected bytes: 480,836,588,544 (447.81 GiB / 480.84 GB) First shard: model-00000-of-00130.safetensors Last shard: model-00124-of-00130.safetensors View all 125 shard files `model-00000-of-00130.safetensors` `model-00001-of-00130.safetensors` `model-00002-of-00130.safetensors` `model-00003-of-00130.safetensors` `model-00004-of-00130.safetensors` `model-00005-of-00130.safetensors` `model-00006-of-00130.safetensors` `model-00007-of-00130.safetensors` `model-00008-of-00130.safetensors` `model-00009-of-00130.safetensors` `model-00010-of-00130.safetensors` `model-00011-of-00130.safetensors` `model-00012-of-00130.safetensors` `model-00013-of-00130.safetensors` `model-00014-of-00130.safetensors` `model-00015-of-00130.safetensors` `model-00016-of-00130.safetensors` `model-00017-of-00130.safetensors` `model-00018-of-00130.safetensors` `model-00019-of-00130.safetensors` `model-00020-of-00130.safetensors` `model-00021-of-00130.safetensors` `model-00022-of-00130.safetensors` `model-00023-of-00130.safetensors` `model-00024-of-00130.safetensors` `model-00025-of-00130.safetensors` `model-00026-of-00130.safetensors` `model-00027-of-00130.safetensors` `model-00028-of-00130.safetensors` `model-00029-of-00130.safetensors` `model-00030-of-00130.safetensors` `model-00031-of-00130.safetensors` `model-00032-of-00130.safetensors` `model-00033-of-00130.safetensors` `model-00034-of-00130.safetensors` `model-00035-of-00130.safetensors` `model-00036-of-00130.safetensors` `model-00037-of-00130.safetensors` `model-00038-of-00130.safetensors` `model-00039-of-00130.safetensors` `model-00040-of-00130.safetensors` `model-00041-of-00130.safetensors` `model-00042-of-00130.safetensors` `model-00043-of-00130.safetensors` `model-00044-of-00130.safetensors` `model-00045-of-00130.safetensors` `model-00046-of-00130.safetensors` `model-00047-of-00130.safetensors` `model-00048-of-00130.safetensors` `model-00049-of-00130.safetensors` `model-00050-of-00130.safetensors` `model-00051-of-00130.safetensors` `model-00052-of-00130.safetensors` `model-00053-of-00130.safetensors` `model-00054-of-00130.safetensors` `model-00055-of-00130.safetensors` `model-00056-of-00130.safetensors` `model-00057-of-00130.safetensors` `model-00058-of-00130.safetensors` `model-00059-of-00130.safetensors` `model-00060-of-00130.safetensors` `model-00061-of-00130.safetensors` `model-00062-of-00130.safetensors` `model-00063-of-00130.safetensors` `model-00064-of-00130.safetensors` `model-00065-of-00130.safetensors` `model-00066-of-00130.safetensors` `model-00067-of-00130.safetensors` `model-00068-of-00130.safetensors` `model-00069-of-00130.safetensors` `model-00070-of-00130.safetensors` `model-00071-of-00130.safetensors` `model-00072-of-00130.safetensors` `model-00073-of-00130.safetensors` `model-00074-of-00130.safetensors` `model-00075-of-00130.safetensors` `model-00076-of-00130.safetensors` `model-00077-of-00130.safetensors` `model-00078-of-00130.safetensors` `model-00079-of-00130.safetensors` `model-00080-of-00130.safetensors` `model-00081-of-00130.safetensors` `model-00082-of-00130.safetensors` `model-00083-of-00130.safetensors` `model-00084-of-00130.safetensors` `model-00085-of-00130.safetensors` `model-00086-of-00130.safetensors` `model-00087-of-00130.safetensors` `model-00088-of-00130.safetensors` `model-00089-of-00130.safetensors` `model-00090-of-00130.safetensors` `model-00091-of-00130.safetensors` `model-00092-of-00130.safetensors` `model-00093-of-00130.safetensors` `model-00094-of-00130.safetensors` `model-00095-of-00130.safetensors` `model-00096-of-00130.safetensors` `model-00097-of-00130.safetensors` `model-00098-of-00130.safetensors` `model-00099-of-00130.safetensors` `model-00100-of-00130.safetensors` `model-00101-of-00130.safetensors` `model-00102-of-00130.safetensors` `model-00103-of-00130.safetensors` `model-00104-of-00130.safetensors` `model-00105-of-00130.safetensors` `model-00106-of-00130.safetensors` `model-00107-of-00130.safetensors` `model-00108-of-00130.safetensors` `model-00109-of-00130.safetensors` `model-00110-of-00130.safetensors` `model-00111-of-00130.safetensors` `model-00112-of-00130.safetensors` `model-00113-of-00130.safetensors` `model-00114-of-00130.safetensors` `model-00115-of-00130.safetensors` `model-00116-of-00130.safetensors` `model-00117-of-00130.safetensors` `model-00118-of-00130.safetensors` `model-00119-of-00130.safetensors` `model-00120-of-00130.safetensors` `model-00121-of-00130.safetensors` `model-00122-of-00130.safetensors` `model-00123-of-00130.safetensors` `model-00124-of-00130.safetensors`	480,836,588,544 (447.81 GiB / 480.84 GB)	`vllm, sglang, unspecified`	`file-metadata`
GGUF Q4_K_M (Unsloth) gguf-q4_k_m Calculator: GGUF Q4_K_M (Unsloth)	Q4_K_M/MiniMax-M2-Q4_K_M-00001-of-00003.gguf, Q4_K_M/MiniMax-M2-Q4_K_M-00002-of-00003.gguf, Q4_K_M/MiniMax-M2-Q4_K_M-00003-of-00003.gguf	138,342,385,120 (128.84 GiB / 138.34 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q5_K_M (Unsloth) gguf-q5_k_m Calculator: GGUF Q5_K_M (Unsloth)	Q5_K_M/MiniMax-M2-Q5_K_M-00001-of-00004.gguf, Q5_K_M/MiniMax-M2-Q5_K_M-00002-of-00004.gguf, Q5_K_M/MiniMax-M2-Q5_K_M-00003-of-00004.gguf, Q5_K_M/MiniMax-M2-Q5_K_M-00004-of-00004.gguf	162,303,936,128 (151.16 GiB / 162.30 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q6_K (Unsloth) gguf-q6_k Calculator: GGUF Q6_K (Unsloth)	Q6_K/MiniMax-M2-Q6_K-00001-of-00004.gguf, Q6_K/MiniMax-M2-Q6_K-00002-of-00004.gguf, Q6_K/MiniMax-M2-Q6_K-00003-of-00004.gguf, Q6_K/MiniMax-M2-Q6_K-00004-of-00004.gguf	187,763,083,904 (174.87 GiB / 187.76 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q8_0 (Unsloth) gguf-q8_0 Calculator: GGUF Q8_0 (Unsloth)	5 safetensors shards Total selected bytes: 243,136,873,184 (226.44 GiB / 243.14 GB) First shard: Q8_0/MiniMax-M2-Q8_0-00001-of-00005.gguf Last shard: Q8_0/MiniMax-M2-Q8_0-00005-of-00005.gguf View all 5 shard files `Q8_0/MiniMax-M2-Q8_0-00001-of-00005.gguf` `Q8_0/MiniMax-M2-Q8_0-00002-of-00005.gguf` `Q8_0/MiniMax-M2-Q8_0-00003-of-00005.gguf` `Q8_0/MiniMax-M2-Q8_0-00004-of-00005.gguf` `Q8_0/MiniMax-M2-Q8_0-00005-of-00005.gguf`	243,136,873,184 (226.44 GiB / 243.14 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
INT4 theoretical	No published file	114,344,882,432 (theoretical)	—	theoretical

Memory by quantization and context

Concurrent sequences = 1. KV defaults to FP16. The range includes documented runtime extras for the listed engine. These figures use the same core as the interactive calculator.

Quantization	2,048 tokens	8,192 tokens	32,768 tokens
FP8 safetensors	448.92 GiB–450.81 GiB	450.38 GiB–452.31 GiB	456.19 GiB–458.30 GiB
GGUF Q4_K_M (Unsloth)	129.45 GiB–132.41 GiB	130.90 GiB–133.99 GiB	136.72 GiB–140.21 GiB
GGUF Q5_K_M (Unsloth)	151.77 GiB–155.17 GiB	153.22 GiB–156.75 GiB	159.03 GiB–162.97 GiB
GGUF Q6_K (Unsloth)	175.48 GiB–179.36 GiB	176.93 GiB–180.94 GiB	182.74 GiB–187.16 GiB
GGUF Q8_0 (Unsloth)	227.05 GiB–231.96 GiB	228.50 GiB–233.54 GiB	234.31 GiB–239.76 GiB
INT4 theoretical Theoretical rows are not downloadable artifacts.	107.10 GiB–138.30 GiB	108.55 GiB–140.12 GiB	114.37 GiB–147.27 GiB

Raw KV contribution

Standard GQA/MHA raw KV is 2 × layers × kvHeads × headDim × tokens × sequences × bytes per element. This page uses KV heads (8), not query heads (48). Using query heads would overstate this cache by 6×.

8,192 tokens, 1 sequence, FP16: 2,080,374,784 bytes (1.9375 GiB)
32,768 tokens: 8,321,499,136 bytes (7.7500 GiB)

Open the calculator with this model

Sources

MiniMax-M2 model card
config.json — revision 757303d492a50514c312788b5247a4f696a4c6a3
safetensors index — metadata.total_size=480836588544
Artifact repo unsloth/MiniMax-M2-GGUF

Assumptions that affect this result

Requested context is within the recorded native window of 196608 tokens.
Full-resident weight estimates use total parameters, including inactive MoE experts.
Activated parameter count is unknown / not independently sourced. Active parameters describe routed compute, not the full resident model weight footprint.
Weight bytes come from the selected weight files only, not from every file in the repository.
Download file size is not peak GPU memory. Runtime layout, allocator padding, and KV are extra.
Counted files: Q4_K_M/MiniMax-M2-Q4_K_M-00001-of-00003.gguf, Q4_K_M/MiniMax-M2-Q4_K_M-00002-of-00003.gguf, Q4_K_M/MiniMax-M2-Q4_K_M-00003-of-00003.gguf.
Multi-token prediction / draft weights exist and are not included unless enabled.
Context budget is the number of cached tokens per sequence (prompt plus reserved generation).

Calculator

Loading catalog snapshot…

Choose a model first, then check a device.

Model

Show historical models

Quantization

Device type

Hardware

Workload

Context length primarily affects KV cache memory.

Advanced settings

Runtime

Engine and KV cache type for this estimate.

EngineKV cache typeRecurrent / conv state dtype Include theoretical estimates (expert) Allow a substitute if the filter has no match

Workload

Overhead scenario and the prefill workspace.

Overhead scenario Include heavier prefill workspace

Topology

How the model is split across devices. Memory is not added together.

DeploymentIntra-replica splitDevice count

Replica count is an advanced deployment option. Quantity is nodes or GPUs, not replicas.

Replica groupsDevices per replica

Budget

System reserve and safety margin. An empty number keeps the previous value.

System / display reserve

GiB (2^30 bytes)

Budget already excludes system reserveExtra safety margin

Custom bits per weightDisplay unit

Custom model

Import a config or type the architecture. Nothing is scanned until you import.

Import local config.json

Parsed in the browser. The file is not uploaded.

Public Hugging Face repoRevision

Only huggingface.co, config.json, no token. Private repos are rejected.

Total parametersActive parametersLayersQuery headsKV headsHead dimension

Search or pick a catalog checkpoint to see memory demand.

Search or pick a catalog checkpoint to see memory demand.