Qwen3.8-Flash-Next on DGX Spark (GB10): Tight fit, Rough estimate
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 Q4_K_M (bartowski), 8,192 cached tokens, 1 sequence, engine llama.cpp. Status: Tight. Software: Not verified..
Verdict on NVIDIA DGX Spark (GB10, 128 GB unified)
Only just: Q4_K_M is the one file that gets in. It needs 111.8 GiB–131.5 GiB at 8,192 tokens against 112.0 GiB of usable unified memory (after the OS reserve). Close other GPU apps first, and expect to cut context if the upper estimate is hit.
With Q4_K_M, the longest context that still fits at all (tight, no headroom) is about 19,092 tokens of a 262,144-token native window, one sequence, fp16 KV cache.
| GGUF quant | File weights | KV cache | Estimated total | Verdict |
|---|---|---|---|---|
| Q8_0 | 175.3 GiB | 0.2 GiB | 175.7 GiB–206.5 GiB | Does not fit |
| Q6_K | 156.6 GiB | 0.2 GiB | 156.9 GiB–184.5 GiB | Does not fit |
| Q5_K_M | 125.4 GiB | 0.2 GiB | 125.8 GiB–148.0 GiB | Does not fit |
| Q4_K_M (recommended) | 111.4 GiB | 0.2 GiB | 111.8 GiB–131.5 GiB | Tight |
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
Qwen3.8-Flash-Next GGUF Q4_K_M on DGX Spark unified memory after the catalog system reserve. Unified memory is not dedicated VRAM.
- ModelQwen3.8-Flash-Next
- HardwareNVIDIA DGX Spark (GB10, 128 GB unified)
- QuantizationGGUF Q4_K_M (bartowski)
- Context length (cached tokens)8,192
- Engine
llama.cpp - Concurrent sequences1
- KV cache typefp16
Technical identifiers and share data
- Model
qwen3.8-flash-next - Hardware
nvidia-dgx-spark-128gb-unified - Quantization
gguf-q4_k_m - Share fragment for this identity
v2.eyJzY2hlbWFWZXJzaW9uIjoyLCJjYWxjdWxhdG9yVmVyc2lvbiI6IjAuMy4wIiwiZGF0YXNldFZlcnNpb24iOiIyMDI2LTEwLTAyLjEiLCJjb25maWciOnsibW9kZWxJZCI6InF3ZW4zLjgtZmxhc2gtbmV4dCIsImhhcmR3YXJlSWQiOiJudmlkaWEtZGd4LXNwYXJrLTEyOGdiLXVuaWZpZWQiLCJkZXZpY2VDb3VudCI6MSwicXVhbnRpemF0aW9uSWQiOiJnZ3VmLXE0X2tfbSIsImNvbnRleHRUb2tlbnMiOjgxOTIsImNvbmN1cnJlbnRTZXF1ZW5jZXMiOjEsImVuZ2luZSI6ImxsYW1hLmNwcCIsImt2VHlwZSI6ImZwMTYifX0
Memory breakdown
- Weight artifact
- 111.39 GiB (119,599,030,080 bytes)
- Raw KV cache
- 0.19 GiB (201,326,592 bytes)
- Engine extras / workspace
- 0.18 GiB–19.95 GiB
- Total estimated demand
- 111.75 GiB–131.53 GiB
- Net calculator budget
- 112.00 GiB
- OS / display reserve
- 16.00 GiB
- Remaining vs net budget
- -19.53 GiB–0.25 GiB
131.53 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 |
|---|---|---|---|
| BF16 safetensors | vllm | 336.14 GiB–388.73 GiB | Does not fit · Rough estimate |
| GGUF Q8_0 (Unsloth) | llama.cpp | 175.66 GiB–206.50 GiB | Does not fit · Rough estimate |
| GGUF Q6_K (bartowski) | llama.cpp | 156.93 GiB–184.53 GiB | Does not fit · Rough estimate |
| GGUF Q5_K_M (bartowski) | llama.cpp | 125.79 GiB–147.99 GiB | Does not fit · Rough estimate |
| GGUF Q4_K_M (bartowski) This page | llama.cpp | 111.75 GiB–131.53 GiB | Tight fit · Rough estimate |
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.
- 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: Qwen3.8-Flash-Next-Q4_K_M/Qwen3.8-Flash-Next-Q4_K_M-00001-of-00004.gguf, Qwen3.8-Flash-Next-Q4_K_M/Qwen3.8-Flash-Next-Q4_K_M-00002-of-00004.gguf, Qwen3.8-Flash-Next-Q4_K_M/Qwen3.8-Flash-Next-Q4_K_M-00003-of-00004.gguf, Qwen3.8-Flash-Next-Q4_K_M/Qwen3.8-Flash-Next-Q4_K_M-00004-of-00004.gguf.
- This checkpoint has a vision component. Text-only inference does not include vision encoder or projector bytes.
- Multi-token prediction / draft weights exist and are not included unless enabled.
Related pages
Sources
- Qwen3.8-Flash-Next model card — Hybrid MoE (36 linear + 12 full). Expert offload unspecified; active params are not resident weights. File size is the packed checkpoint.
- config.json — revision de4b8e4d43b917e7706784d8bb445c9af86a3540
- safetensors index — metadata.total_size=359999963128
- Artifact repo unsloth/Qwen3.8-Flash-Next-GGUF
- Artifact repo bartowski/Qwen3.8-Flash-Next-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.