GLM-4.5-Air on DGX Spark (GB10): Fits (Sep 2026)
GB10 is the superchip named on this record, NVIDIA DGX Spark (GB10, 128 GB unified). 128 GB LPDDR5x unified memory on the GB10 Superchip.
Conditional memory judgement for one configuration. Change context, quant, or engine and the conclusion can change. Configuration used on this page: GGUF Q6_K (Unsloth), 8,192 cached tokens, 1 sequence, engine llama.cpp. Status: Likely fits. Software: Not verified..
Verdict on NVIDIA DGX Spark (GB10, 128 GB unified)
Yes, run Q6_K. The calculator puts it at 93.8 GiB–96.1 GiB at 8,192 tokens, inside 112.0 GiB of usable unified memory (after the OS reserve). Q8_0 is the next step up and does not clear the bar (111.0 GiB–113.6 GiB).
With Q6_K, the longest context that still counts as a fit is about 33,276 tokens of a 131,072-token native window, one sequence, fp16 KV cache.
| GGUF quant | File weights | KV cache | Estimated total | Verdict |
|---|---|---|---|---|
| Q8_0 | 109.4 GiB | 1.4 GiB | 111.0 GiB–113.6 GiB | Tight |
| Q6_K (recommended) | 92.2 GiB | 1.4 GiB | 93.8 GiB–96.1 GiB | Fits |
| Q5_K_M | 77.8 GiB | 1.4 GiB | 79.3 GiB–81.4 GiB | Fits |
| Q4_K_M | 68.0 GiB | 1.4 GiB | 69.5 GiB–71.4 GiB | Comfortable |
Engine llama.cpp, one sequence, fp16 KV. File weights are the published GGUF sizes in the catalog. Totals come from the same calculator that powers the tool below; they are estimates, not a measurement of this setup.
Fixed page configuration
GGUF Q6_K needs 93.8 GiB-96.1 GiB at 8,192 tokens against 112.00 GiB net on NVIDIA DGX Spark (GB10, 128 GB unified). Q8_0 is tight or over budget.
- ModelGLM-4.5-Air
- HardwareNVIDIA DGX Spark (GB10, 128 GB unified)
- QuantizationGGUF Q6_K (Unsloth)
- Context length (cached tokens)8,192
- Engine
llama.cpp - Concurrent sequences1
- KV cache typefp16
Technical identifiers and share data
- Model
glm-4.5-air - Hardware
nvidia-dgx-spark-128gb-unified - Quantization
gguf-q6_k - Share fragment for this identity
v2.eyJzY2hlbWFWZXJzaW9uIjoyLCJjYWxjdWxhdG9yVmVyc2lvbiI6IjAuMy4wIiwiZGF0YXNldFZlcnNpb24iOiIyMDI2LTEwLTAyLjEiLCJjb25maWciOnsibW9kZWxJZCI6ImdsbS00LjUtYWlyIiwiaGFyZHdhcmVJZCI6Im52aWRpYS1kZ3gtc3BhcmstMTI4Z2ItdW5pZmllZCIsImRldmljZUNvdW50IjoxLCJxdWFudGl6YXRpb25JZCI6ImdndWYtcTZfayIsImNvbnRleHRUb2tlbnMiOjgxOTIsImNvbmN1cnJlbnRTZXF1ZW5jZXMiOjEsImVuZ2luZSI6ImxsYW1hLmNwcCIsImt2VHlwZSI6ImZwMTYifX0
Memory breakdown
- Weight artifact
- 92.21 GiB (99,007,528,192 bytes)
- Raw KV cache
- 1.44 GiB (1,543,503,872 bytes)
- Engine extras / workspace
- 0.13 GiB–2.44 GiB
- Total estimated demand
- 93.77 GiB–96.09 GiB
- Net calculator budget
- 112.00 GiB
- OS / display reserve
- 16.00 GiB
- Remaining vs net budget
- 15.91 GiB–18.23 GiB
96.09 GiB high estimate vs 112.00 GiB net
Enough memory does not mean this engine, quant, and OS will run. Support does not mean it will be fast.
Other published files on this device
Same model, device and context (8,192 tokens, 1 sequence, FP16 KV). Each row is one published weight-file set with its usual engine, sorted by memory.
| Quantization | Engine | Estimated memory | Fit |
|---|---|---|---|
| GGUF Q8_0 (Unsloth) | llama.cpp | 110.95 GiB–113.62 GiB | Tight fit |
| BF16 safetensors | vllm | 104.94 GiB–106.86 GiB | Tight fit |
| GGUF Q6_K (Unsloth) This page | llama.cpp | 93.77 GiB–96.09 GiB | Fits |
| GGUF Q5_K_M (Unsloth) | llama.cpp | 79.35 GiB–81.38 GiB | Fits |
| GGUF Q4_K_M (Unsloth) | llama.cpp | 69.53 GiB–71.36 GiB | Runs comfortably |
Assumptions that affect this result
- Requested context is within the recorded native window of 131072 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: Q6_K/GLM-4.5-Air-Q6_K-00001-of-00002.gguf, Q6_K/GLM-4.5-Air-Q6_K-00002-of-00002.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).
Related pages
Sources
- GLM-4.5-Air model card
- config.json — revision a24ceef6ce4f3536971efe9b778bdaa1bab18daa
- safetensors index — metadata.total_size=110468824832
- Artifact repo unsloth/GLM-4.5-Air-GGUF
- NVIDIA DGX Spark product page — 128 GB LPDDR5x coherent unified system memory. This is not dedicated VRAM. Default reserve deducts OS and other processes. Multiple Sparks are not one pool.
- NVIDIA DGX Spark hardware docs — 128 GB LPDDR5x unified memory on the GB10 Superchip.
Calculator
This combo page keeps the route model and device. Customize to change them on the main calculator; the share then keeps that identity.
Loading catalog snapshot…
Search or pick a catalog checkpoint to see memory demand.