4× DGX Spark
65.4tok/sEveryday
1 request, prose prompt, with MTP + DFlash2 (per recipe) Source
Intelligence42with thinking
- Custom kernel required
- Non-commercial
- License unclear
Local AI · Recipe · 4× DGX Spark
GLM-5.3-Flash NVFP4/MXFP8 CSF with SparkRing on 4× DGX Spark
GLM-5.3-Flash NVFP4/MXFP8 CSF with SparkRing on 4× DGX Spark: 68 tok/s according to github.com (dataset as of Nov 3, 2026).
68tok/sEveryday
1 request, generation at 32K context (speed table; measured only as a path of four Sparks inside a ring of eight, not as a ring of four), thinking always on (measured before the profile default high, i.e. template default max), with MTP3 (per recipe)
Source: github.comEngine
- Engine
- vLLM (local-inference-lab Fork) via SparkRing (SIRCL-Collectives, switchless RDMA ueber ConnectX-7), Image Release 2026.10.2
- Quantization
- NVFP4/MXFP8 CSF QAD (Local Inference Lab)
- Model family
- GLM-5.3-Flash
- Context
- 1,048,576
- Parameters
- 45 Layer, 288 Experten (MoE, 8 aktiv); 177,7 GB Safetensors
- Creator
- fujitsupolycom
- GitHub stars
- 102
- Repo updated
- Nov 2, 2026
Measurements
Every sourced value of this recipe, each with its condition and source. Bars relative to the largest value in the group.
Decode · 1 request
Decode · parallel requests
Sums over all streams are marked; per request it is correspondingly less.
Prefill by context
What you need
- Hardware
- 4 × NVIDIA DGX Spark (GB10)
- Weights
- Main weightslocal-inference-lab/GLM-5.3-Flash-NVFP4-MXFP8-CSF-QAD other
- Engine
- vLLM (local-inference-lab Fork) via SparkRing (SIRCL-Collectives, switchless RDMA ueber ConnectX-7), Image Release 2026.10.2
- Context
- 1,048,576 tokens
Notes
What matters before you rebuild it.
- Custom kernel required
- License unclear
Sources
- github.com/FujitsuPolycom/sparkring#speed
- github.com/FujitsuPolycom/sparkring/blob/main/performance/records/images/release-2026.10.2-speed-32k-20261012.md
- github.com/FujitsuPolycom/sparkring/blob/main/performance/records/images/dev-20261010-kraken-csf-sircl032-libsircl060cd-plugins-status036-release-2026.10.3-plain-tp4-20261013.md
- github.com/FujitsuPolycom/sparkring#profiles
- github.com/FujitsuPolycom/sparkring/blob/main/profiles/glm53-checkpoints.md
- github.com/FujitsuPolycom/sparkring/blob/main/profiles/glm53-nvfp4-tp8/README.md#request-defaults
Related recipes
2× DGX Spark
60.4tok/sEveryday
Prose decode, 1 request; quality (FP8 KV, 4-bit dense): GSM8K 98.8% / HumanEval 95.7% / HumanEval+ & MBPP+ 86.5% Source
Intelligence42with thinking
- ≤ 3 bit: quantization may cost quality
- Custom kernel required
- Non-commercial
- License unclear
2× DGX Spark
44tok/sEveryday
1 request, prose prompt, with MTP + DFlash2 (per recipe) Source
Intelligence42with thinking
- Custom kernel required
- Non-commercial
- License unclear
4× DGX Spark
40.8tok/sEveryday
1 request, prose prompt, with DFlash2 (per recipe) Source
Intelligence42with thinking
- Custom kernel required
- Non-commercial
3× DGX Spark
40.1tok/sEveryday
1 request, prose prompt, with DFlash2, TR3 weights (per recipe) Source
Intelligence42with thinking
- ≤ 3 bit: quantization may cost quality
- Custom kernel required
- Non-commercial
- License unclear
2× DGX Spark
38tok/sEveryday
1 request, generation at 32K context (speed table, pair), thinking always on (measured before the profile default high, i.e. template default max), with MTP3 (per recipe) Source
Intelligence42with thinking
- Custom kernel required
- License unclear
Local AI in your company?
In a workshop we work out which models and which hardware fit your tasks, and build the first agent on your infrastructure.