Qwen3-235B-A22B-Instruct-2507 VRAM & Memory Requirements

Specific checkpoints with sourced architecture and weight files. GGUF Q4_K_M (Unsloth): 133.99 GiB–137.11 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

PublisherQwen
Hugging FaceQwen/Qwen3-235B-A22B-Instruct-2507
Revisionac9c66cc9b46af7306746a9250f23d47083d689e
Total parameters235,093,634,560
Active parameters22,000,000,000
Layers94
Query heads64
KV heads4 (gqa)
Head dimension128
Recorded context capability262,144
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: 132.39 GiB (142,154,074,880 bytes)
Raw KV cache: 1.47 GiB (1,577,058,304 bytes)
Engine extras / workspace: 0.13 GiB–3.25 GiB
Total estimated demand: 133.99 GiB–137.11 GiB

Published artifacts

Quantization	Selected files	Bytes	Engine	Evidence
BF16 safetensors bf16 Calculator: BF16 safetensors	118 safetensors shards Total selected bytes: 470,187,269,120 (437.90 GiB / 470.19 GB) First shard: model-00001-of-00118.safetensors Last shard: model-00118-of-00118.safetensors View all 118 shard files `model-00001-of-00118.safetensors` `model-00002-of-00118.safetensors` `model-00003-of-00118.safetensors` `model-00004-of-00118.safetensors` `model-00005-of-00118.safetensors` `model-00006-of-00118.safetensors` `model-00007-of-00118.safetensors` `model-00008-of-00118.safetensors` `model-00009-of-00118.safetensors` `model-00010-of-00118.safetensors` `model-00011-of-00118.safetensors` `model-00012-of-00118.safetensors` `model-00013-of-00118.safetensors` `model-00014-of-00118.safetensors` `model-00015-of-00118.safetensors` `model-00016-of-00118.safetensors` `model-00017-of-00118.safetensors` `model-00018-of-00118.safetensors` `model-00019-of-00118.safetensors` `model-00020-of-00118.safetensors` `model-00021-of-00118.safetensors` `model-00022-of-00118.safetensors` `model-00023-of-00118.safetensors` `model-00024-of-00118.safetensors` `model-00025-of-00118.safetensors` `model-00026-of-00118.safetensors` `model-00027-of-00118.safetensors` `model-00028-of-00118.safetensors` `model-00029-of-00118.safetensors` `model-00030-of-00118.safetensors` `model-00031-of-00118.safetensors` `model-00032-of-00118.safetensors` `model-00033-of-00118.safetensors` `model-00034-of-00118.safetensors` `model-00035-of-00118.safetensors` `model-00036-of-00118.safetensors` `model-00037-of-00118.safetensors` `model-00038-of-00118.safetensors` `model-00039-of-00118.safetensors` `model-00040-of-00118.safetensors` `model-00041-of-00118.safetensors` `model-00042-of-00118.safetensors` `model-00043-of-00118.safetensors` `model-00044-of-00118.safetensors` `model-00045-of-00118.safetensors` `model-00046-of-00118.safetensors` `model-00047-of-00118.safetensors` `model-00048-of-00118.safetensors` `model-00049-of-00118.safetensors` `model-00050-of-00118.safetensors` `model-00051-of-00118.safetensors` `model-00052-of-00118.safetensors` `model-00053-of-00118.safetensors` `model-00054-of-00118.safetensors` `model-00055-of-00118.safetensors` `model-00056-of-00118.safetensors` `model-00057-of-00118.safetensors` `model-00058-of-00118.safetensors` `model-00059-of-00118.safetensors` `model-00060-of-00118.safetensors` `model-00061-of-00118.safetensors` `model-00062-of-00118.safetensors` `model-00063-of-00118.safetensors` `model-00064-of-00118.safetensors` `model-00065-of-00118.safetensors` `model-00066-of-00118.safetensors` `model-00067-of-00118.safetensors` `model-00068-of-00118.safetensors` `model-00069-of-00118.safetensors` `model-00070-of-00118.safetensors` `model-00071-of-00118.safetensors` `model-00072-of-00118.safetensors` `model-00073-of-00118.safetensors` `model-00074-of-00118.safetensors` `model-00075-of-00118.safetensors` `model-00076-of-00118.safetensors` `model-00077-of-00118.safetensors` `model-00078-of-00118.safetensors` `model-00079-of-00118.safetensors` `model-00080-of-00118.safetensors` `model-00081-of-00118.safetensors` `model-00082-of-00118.safetensors` `model-00083-of-00118.safetensors` `model-00084-of-00118.safetensors` `model-00085-of-00118.safetensors` `model-00086-of-00118.safetensors` `model-00087-of-00118.safetensors` `model-00088-of-00118.safetensors` `model-00089-of-00118.safetensors` `model-00090-of-00118.safetensors` `model-00091-of-00118.safetensors` `model-00092-of-00118.safetensors` `model-00093-of-00118.safetensors` `model-00094-of-00118.safetensors` `model-00095-of-00118.safetensors` `model-00096-of-00118.safetensors` `model-00097-of-00118.safetensors` `model-00098-of-00118.safetensors` `model-00099-of-00118.safetensors` `model-00100-of-00118.safetensors` `model-00101-of-00118.safetensors` `model-00102-of-00118.safetensors` `model-00103-of-00118.safetensors` `model-00104-of-00118.safetensors` `model-00105-of-00118.safetensors` `model-00106-of-00118.safetensors` `model-00107-of-00118.safetensors` `model-00108-of-00118.safetensors` `model-00109-of-00118.safetensors` `model-00110-of-00118.safetensors` `model-00111-of-00118.safetensors` `model-00112-of-00118.safetensors` `model-00113-of-00118.safetensors` `model-00114-of-00118.safetensors` `model-00115-of-00118.safetensors` `model-00116-of-00118.safetensors` `model-00117-of-00118.safetensors` `model-00118-of-00118.safetensors`	470,187,269,120 (437.90 GiB / 470.19 GB)	`vllm, sglang, unspecified`	`file-metadata`
GGUF Q4_K_M (Unsloth) gguf-q4_k_m Calculator: GGUF Q4_K_M (Unsloth)	Q4_K_M/Qwen3-235B-A22B-Instruct-2507-Q4_K_M-00001-of-00003.gguf, Q4_K_M/Qwen3-235B-A22B-Instruct-2507-Q4_K_M-00002-of-00003.gguf, Q4_K_M/Qwen3-235B-A22B-Instruct-2507-Q4_K_M-00003-of-00003.gguf	142,154,074,880 (132.39 GiB / 142.15 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/Qwen3-235B-A22B-Instruct-2507-Q5_K_M-00001-of-00004.gguf, Q5_K_M/Qwen3-235B-A22B-Instruct-2507-Q5_K_M-00002-of-00004.gguf, Q5_K_M/Qwen3-235B-A22B-Instruct-2507-Q5_K_M-00003-of-00004.gguf, Q5_K_M/Qwen3-235B-A22B-Instruct-2507-Q5_K_M-00004-of-00004.gguf	166,814,157,632 (155.36 GiB / 166.81 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q6_K (Unsloth) gguf-q6_k Calculator: GGUF Q6_K (Unsloth)	Q6_K/Qwen3-235B-A22B-Instruct-2507-Q6_K-00001-of-00004.gguf, Q6_K/Qwen3-235B-A22B-Instruct-2507-Q6_K-00002-of-00004.gguf, Q6_K/Qwen3-235B-A22B-Instruct-2507-Q6_K-00003-of-00004.gguf, Q6_K/Qwen3-235B-A22B-Instruct-2507-Q6_K-00004-of-00004.gguf	193,015,495,552 (179.76 GiB / 193.02 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
GGUF Q8_0 (Unsloth) gguf-q8_0 Calculator: GGUF Q8_0 (Unsloth)	6 safetensors shards Total selected bytes: 249,940,106,336 (232.77 GiB / 249.94 GB) First shard: Q8_0/Qwen3-235B-A22B-Instruct-2507-Q8_0-00001-of-00006.gguf Last shard: Q8_0/Qwen3-235B-A22B-Instruct-2507-Q8_0-00006-of-00006.gguf View all 6 shard files `Q8_0/Qwen3-235B-A22B-Instruct-2507-Q8_0-00001-of-00006.gguf` `Q8_0/Qwen3-235B-A22B-Instruct-2507-Q8_0-00002-of-00006.gguf` `Q8_0/Qwen3-235B-A22B-Instruct-2507-Q8_0-00003-of-00006.gguf` `Q8_0/Qwen3-235B-A22B-Instruct-2507-Q8_0-00004-of-00006.gguf` `Q8_0/Qwen3-235B-A22B-Instruct-2507-Q8_0-00005-of-00006.gguf` `Q8_0/Qwen3-235B-A22B-Instruct-2507-Q8_0-00006-of-00006.gguf`	249,940,106,336 (232.77 GiB / 249.94 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
INT4 theoretical	No published file	117,546,817,280 (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
BF16 safetensors	438.89 GiB–440.77 GiB	439.99 GiB–441.91 GiB	444.40 GiB–446.45 GiB
GGUF Q4_K_M (Unsloth)	132.88 GiB–135.91 GiB	133.99 GiB–137.11 GiB	138.39 GiB–141.83 GiB
GGUF Q5_K_M (Unsloth)	155.85 GiB–159.34 GiB	156.95 GiB–160.54 GiB	161.36 GiB–165.25 GiB
GGUF Q6_K (Unsloth)	180.25 GiB–184.23 GiB	181.35 GiB–185.43 GiB	185.76 GiB–190.14 GiB
GGUF Q8_0 (Unsloth)	233.27 GiB–238.30 GiB	234.37 GiB–239.50 GiB	238.77 GiB–244.22 GiB
INT4 theoretical Theoretical rows are not downloadable artifacts.	109.97 GiB–142.01 GiB	111.07 GiB–143.38 GiB	115.47 GiB–148.81 GiB

Raw KV contribution

Standard GQA/MHA raw KV is 2 × layers × kvHeads × headDim × tokens × sequences × bytes per element. This page uses KV heads (4), not query heads (64). Using query heads would overstate this cache by 16×.

8,192 tokens, 1 sequence, FP16: 1,577,058,304 bytes (1.4688 GiB)
32,768 tokens: 6,308,233,216 bytes (5.8750 GiB)

Open the calculator with this model

Sources

Qwen3-235B-A22B-Instruct-2507 model card
config.json — revision ac9c66cc9b46af7306746a9250f23d47083d689e
safetensors index — metadata.total_size=470187269120
Artifact repo unsloth/Qwen3-235B-A22B-Instruct-2507-GGUF

Assumptions that affect this result

Requested context is within the recorded native window of 262144 tokens.
Full-resident weight estimates use total parameters, including inactive MoE experts.
This MoE lists 235093634560 total parameters and 22000000000 active parameters. Active-parameter compute is not used as the weight-residency figure.
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/Qwen3-235B-A22B-Instruct-2507-Q4_K_M-00001-of-00003.gguf, Q4_K_M/Qwen3-235B-A22B-Instruct-2507-Q4_K_M-00002-of-00003.gguf, Q4_K_M/Qwen3-235B-A22B-Instruct-2507-Q4_K_M-00003-of-00003.gguf.
Context budget is the number of cached tokens per sequence (prompt plus reserved generation).
Concurrent sequences share one model replica; weights are not multiplied by sequence count.

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.