---
type: "Article"
title: "GLM-5.3 NVFP4 with SparkRing on 8× DGX Spark"
description: "GLM-5.3 NVFP4 with SparkRing on 8× DGX Spark: recipe by fujitsupolycom with speed, condition and source for every number, weights and license."
resource: "https://www.contextstudios.ai/local-ai/fujitsupolycom--glm-5.3-nvfp4-tp8-8x-dgx-spark"
language: "en"
generated:
  by: "process:contextstudios-md/1"
  at: "2026-11-11T03:20:23.381Z"
status: "stable"
---

# GLM-5.3 NVFP4 with SparkRing on 8× DGX Spark

GLM-5.3 NVFP4 with SparkRing on 8× DGX Spark: 46 tok/s according to github.com (dataset as of Nov 3, 2026).

- Creator: [fujitsupolycom](https://www.contextstudios.ai/local-ai/creators/fujitsupolycom.md)
- Engine: vLLM (local-inference-lab Fork) via SparkRing (SIRCL-Collectives, switchless RDMA ueber ConnectX-7), Image Release 2026.10.2
- Quantization: NVFP4-Experten/BF16 (Local Inference Lab; Dense-Linears beim Laden MXFP8)
- Model family: GLM-5.3
- Hardware: dgx-spark x8
- Context: 1048576
- Artificial Analysis Intelligence Index: 44.8 ([source](https://artificialanalysis.ai/models/glm-5-3))

## All numbers

- 46 tok/s (Everyday; Everyday, 1 request, generation at 32K context (speed table, ring of eight, TP8/DCP4), thinking always on (measured before the profile default high, i.e. template default max), with MTP2 (per recipe); experimental profile) — [source](https://github.com/FujitsuPolycom/sparkring#speed)
- decode_tps: 46 (1 request) — [source](https://github.com/FujitsuPolycom/sparkring#speed)
- decode_tps: 129 (8 parallel requests) — [source](https://github.com/FujitsuPolycom/sparkring#speed)
- prefill_tps: 1387 (1 request) — [source](https://github.com/FujitsuPolycom/sparkring#speed)

## Weights

- [local-inference-lab/GLM-5.3-NVFP4](https://huggingface.co/local-inference-lab/GLM-5.3-NVFP4) (main, mit)

[https://github.com/FujitsuPolycom/sparkring](https://github.com/FujitsuPolycom/sparkring)

## Related

- [Local AI on your hardware.](https://www.contextstudios.ai/local-ai.md)