3D-Printed Cooling Mod Tackles NVIDIA Jetson Thor Edge AI Thermal Throttling
A developer has released an open-source, 3D-printable cooling system designed to resolve severe thermal throttling on the NVIDIA Jetson AGX Thor Developer Kit. Powered by NVIDIA's Tegra264 Blackwell architecture, the edge AI unit features an asymmetric internal layout where the lower processing module receives active fan cooling, but the upper compartment—housing the high-power voltage regulator…
Key points
- NVIDIA's Jetson AGX Thor Developer Kit experiences thermal throttling in its passively cooled upper chamber during heavy local LLM inference.
- A custom 3D-printed dock uses dual 120mm fans and aerodynamic deflectors to cool PCIe Gen5 NVMe drives and VRMs.
- The open-source OpenSCAD design stabilizes idle GPU temperatures to 36.9°C and NVMe temperatures to 32.0°C under MIT licensing.
During sustained high-concurrency LLM inference and speculative decoding workloads using frameworks like vLLM and models such as Qwen 2.5, the upper components routinely experience heat-soaking. Temperatures on high-speed Gen5 NVMe drives frequently climb past 60°C to 70°C, triggering drive throttling and carrier board thermal warnings under continuous 76W to 80W power draw.
To address this without invasive hardware modifications or loud server fans, the creator designed a parametric OpenSCAD dual-zone shroud utilizing standard, quiet 120mm Noctua fans. The system channels focused laminar airflow through the internal heatsink and deploys an aerodynamic air-knife deflector over the chassis roof, bringing idle and light-load temperatures down to 36.9°C on the GPU core and 32.0°C on the NVMe drive.
Made a 3D print model to cool down my NVIDIA Thor Dev Kit with two 12 inch fans
github.com · 14 September 2026Loading the full article…
This text was published by github.com. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
1source- Reddit discussionreddit.com
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Hardware & Compute
All →- Applied Materials and Intel expand AI chipmaking collaboration at epic Center · 1 src
- Xeal to deploy 100,000 Nvidia GPUs in EV charging pods across US · 1 src
- Meta’s Muse boosts AMD stock as AI agents shift workload to CPUs · 1 src
- Nokia CEO says data center construction could double without supply constraints · 1 src
- Ghost launches $3,499 AI agent computer with $11M seed · 1 src
Comments
via GitHub Discussions