GLM-5.3 needs 876.98 GiB

GLM-5.3 (Zhipu AI, zai-org/GLM-5.3) lists 753,329,940,480 parameters in this catalog. Hardware needed starts from the published files: auto-selected GGUF Q8_0 (Unsloth) is 801,357,677,216 bytes, and at 8,192 tokens, one stream, the calculator estimates 747.13 GiB–876.98 GiB (Q8_0). Rough estimate, not a measurement. Official FP8 safetensors are 755,617,140,416 bytes, estimated at 705.03 GiB–813.09 GiB. None of the 32 catalog devices fit GGUF Q8_0 (Unsloth). The largest unified devices are NVIDIA DGX Spark (GB10, 128 GB unified) and Apple M4 Max 128GB unified (MacBook Pro) and Ryzen AI Max+ 395. The largest discrete devices are NVIDIA A100 80GB SXM and NVIDIA H100 80GB SXM. Discrete GPUs are not summed, and unified memory is not added on top of discrete VRAM. INT4 theoretical is 351.61 GiB–453.37 GiB. It is not a published file. No catalog device fits that theoretical row. No can-i-run page is allowlisted for GLM-5.3 or GLM-5.3-Flash.

GLM-5.3-Flash. GLM-5.3-Flash on 2× DGX Spark. DGX Spark 128 GB unified. MiaAI-Lab TensorFold 0.6 TP2. LimeChain TensorFold C8 comparison.

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

Publisher and provenance

PublisherZhipu AI
Hugging Facezai-org/GLM-5.3
Revisionaca966e4e02791568aa6a4ced368624b3d897f42
Total parameters753,329,940,480
Active parametersUnknown / not independently sourced
Layers78
Query heads64
KV heads64 (mla)
Head dimension192
Recorded context capability1,048,576
Completenesspartial · partial

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: 746.32 GiB (801,357,677,216 bytes)
Raw KV cache: 0.69 GiB (736,100,352 bytes)
Engine extras / workspace: 0.13 GiB–129.97 GiB
Total estimated demand: 747.13 GiB–876.98 GiB

Published artifacts

Quantization	Selected files	Bytes	Engine	Evidence
FP8 safetensors fp8 Calculator: FP8 safetensors	141 safetensors shards Total selected bytes: 755,617,140,416 (703.72 GiB / 755.62 GB) First shard: model-00001-of-00141.safetensors Last shard: model-00141-of-00141.safetensors View all 141 shard files `model-00001-of-00141.safetensors` `model-00002-of-00141.safetensors` `model-00003-of-00141.safetensors` `model-00004-of-00141.safetensors` `model-00005-of-00141.safetensors` `model-00006-of-00141.safetensors` `model-00007-of-00141.safetensors` `model-00008-of-00141.safetensors` `model-00009-of-00141.safetensors` `model-00010-of-00141.safetensors` `model-00011-of-00141.safetensors` `model-00012-of-00141.safetensors` `model-00013-of-00141.safetensors` `model-00014-of-00141.safetensors` `model-00015-of-00141.safetensors` `model-00016-of-00141.safetensors` `model-00017-of-00141.safetensors` `model-00018-of-00141.safetensors` `model-00019-of-00141.safetensors` `model-00020-of-00141.safetensors` `model-00021-of-00141.safetensors` `model-00022-of-00141.safetensors` `model-00023-of-00141.safetensors` `model-00024-of-00141.safetensors` `model-00025-of-00141.safetensors` `model-00026-of-00141.safetensors` `model-00027-of-00141.safetensors` `model-00028-of-00141.safetensors` `model-00029-of-00141.safetensors` `model-00030-of-00141.safetensors` `model-00031-of-00141.safetensors` `model-00032-of-00141.safetensors` `model-00033-of-00141.safetensors` `model-00034-of-00141.safetensors` `model-00035-of-00141.safetensors` `model-00036-of-00141.safetensors` `model-00037-of-00141.safetensors` `model-00038-of-00141.safetensors` `model-00039-of-00141.safetensors` `model-00040-of-00141.safetensors` `model-00041-of-00141.safetensors` `model-00042-of-00141.safetensors` `model-00043-of-00141.safetensors` `model-00044-of-00141.safetensors` `model-00045-of-00141.safetensors` `model-00046-of-00141.safetensors` `model-00047-of-00141.safetensors` `model-00048-of-00141.safetensors` `model-00049-of-00141.safetensors` `model-00050-of-00141.safetensors` `model-00051-of-00141.safetensors` `model-00052-of-00141.safetensors` `model-00053-of-00141.safetensors` `model-00054-of-00141.safetensors` `model-00055-of-00141.safetensors` `model-00056-of-00141.safetensors` `model-00057-of-00141.safetensors` `model-00058-of-00141.safetensors` `model-00059-of-00141.safetensors` `model-00060-of-00141.safetensors` `model-00061-of-00141.safetensors` `model-00062-of-00141.safetensors` `model-00063-of-00141.safetensors` `model-00064-of-00141.safetensors` `model-00065-of-00141.safetensors` `model-00066-of-00141.safetensors` `model-00067-of-00141.safetensors` `model-00068-of-00141.safetensors` `model-00069-of-00141.safetensors` `model-00070-of-00141.safetensors` `model-00071-of-00141.safetensors` `model-00072-of-00141.safetensors` `model-00073-of-00141.safetensors` `model-00074-of-00141.safetensors` `model-00075-of-00141.safetensors` `model-00076-of-00141.safetensors` `model-00077-of-00141.safetensors` `model-00078-of-00141.safetensors` `model-00079-of-00141.safetensors` `model-00080-of-00141.safetensors` `model-00081-of-00141.safetensors` `model-00082-of-00141.safetensors` `model-00083-of-00141.safetensors` `model-00084-of-00141.safetensors` `model-00085-of-00141.safetensors` `model-00086-of-00141.safetensors` `model-00087-of-00141.safetensors` `model-00088-of-00141.safetensors` `model-00089-of-00141.safetensors` `model-00090-of-00141.safetensors` `model-00091-of-00141.safetensors` `model-00092-of-00141.safetensors` `model-00093-of-00141.safetensors` `model-00094-of-00141.safetensors` `model-00095-of-00141.safetensors` `model-00096-of-00141.safetensors` `model-00097-of-00141.safetensors` `model-00098-of-00141.safetensors` `model-00099-of-00141.safetensors` `model-00100-of-00141.safetensors` `model-00101-of-00141.safetensors` `model-00102-of-00141.safetensors` `model-00103-of-00141.safetensors` `model-00104-of-00141.safetensors` `model-00105-of-00141.safetensors` `model-00106-of-00141.safetensors` `model-00107-of-00141.safetensors` `model-00108-of-00141.safetensors` `model-00109-of-00141.safetensors` `model-00110-of-00141.safetensors` `model-00111-of-00141.safetensors` `model-00112-of-00141.safetensors` `model-00113-of-00141.safetensors` `model-00114-of-00141.safetensors` `model-00115-of-00141.safetensors` `model-00116-of-00141.safetensors` `model-00117-of-00141.safetensors` `model-00118-of-00141.safetensors` `model-00119-of-00141.safetensors` `model-00120-of-00141.safetensors` `model-00121-of-00141.safetensors` `model-00122-of-00141.safetensors` `model-00123-of-00141.safetensors` `model-00124-of-00141.safetensors` `model-00125-of-00141.safetensors` `model-00126-of-00141.safetensors` `model-00127-of-00141.safetensors` `model-00128-of-00141.safetensors` `model-00129-of-00141.safetensors` `model-00130-of-00141.safetensors` `model-00131-of-00141.safetensors` `model-00132-of-00141.safetensors` `model-00133-of-00141.safetensors` `model-00134-of-00141.safetensors` `model-00135-of-00141.safetensors` `model-00136-of-00141.safetensors` `model-00137-of-00141.safetensors` `model-00138-of-00141.safetensors` `model-00139-of-00141.safetensors` `model-00140-of-00141.safetensors` `model-00141-of-00141.safetensors`	755,617,140,416 (703.72 GiB / 755.62 GB)	`vllm, sglang, unspecified`	`file-metadata`
GGUF Q8_0 (Unsloth) gguf-q8_0 Calculator: GGUF Q8_0 (Unsloth)	17 safetensors shards Total selected bytes: 801,357,677,216 (746.32 GiB / 801.36 GB) First shard: Q8_0/GLM-5.3-Q8_0-00001-of-00017.gguf Last shard: Q8_0/GLM-5.3-Q8_0-00017-of-00017.gguf View all 17 shard files `Q8_0/GLM-5.3-Q8_0-00001-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00002-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00003-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00004-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00005-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00006-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00007-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00008-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00009-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00010-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00011-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00012-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00013-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00014-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00015-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00016-of-00017.gguf` `Q8_0/GLM-5.3-Q8_0-00017-of-00017.gguf`	801,357,677,216 (746.32 GiB / 801.36 GB)	`llama.cpp, ollama, unspecified`	`file-metadata`
INT4 theoretical	No published file	376,664,970,240 (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	704.52 GiB–812.39 GiB	705.03 GiB–813.09 GiB	707.09 GiB–815.89 GiB
GGUF Q8_0 (Unsloth)	746.62 GiB–876.24 GiB	747.13 GiB–876.98 GiB	749.19 GiB–879.89 GiB
INT4 theoretical Theoretical rows are not downloadable artifacts.	351.09 GiB–452.63 GiB	351.61 GiB–453.37 GiB	353.66 GiB–456.28 GiB

Raw KV contribution

Open the calculator with this model

Sources

GLM-5.3 model card
config.json — revision aca966e4e02791568aa6a4ced368624b3d897f42
safetensors index — metadata.total_size=755617140416
Artifact repo unsloth/GLM-5.3-GGUF

Assumptions that affect this result

Requested context is within the recorded native window of 1048576 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.
Weight bytes come from 17 published safetensors shards totaling 801357677216 bytes.
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.

GLM-5.3 needs 876.98 GiB

Publisher and provenance

Memory layers in this estimate

Published artifacts

Memory by quantization and context

Raw KV contribution

Related pages

Sources

Assumptions that affect this result

Calculator