The register › perception

Neural networks benchmarked on phone-class chips aboard the International Space Station

Ran deep learning inference on two commercial processors aboard the station, to measure how they perform there.

OperatorHewlett Packard Enterprise (the host computer)
DeveloperNASA Jet Propulsion Laboratory, with Ubotica
HostHPE Spaceborne Computer-2, International Space Station
Wherespace station
Classperception
HardwareQualcomm Snapdragon 855 and Intel Movidius Myriad X, hosted by HPE Spaceborne Computer-2
Model or softwareDeep learning models for Mars rover imagery, spectral unmixing and ship detection, by the paper's title
Launchednot applicable
Ranby 2022

What was said, and by whom

  1. “We benchmark deep learning inference on Movidius Myriad X and Snapdragon processors onboard the ISS.”

    Dunkel et al., JPL and Ubotica, the experiment's team: the experimenter. Testing Mars Rover, Spectral Unmixing, and Ship Detection Neural Networks, and Memory Checkers on Embedded Systems Onboard the ISS, Dunkel and co-authors, JPL and Ubotica (the experiment's team). Published 2022 (year in the file's address; no date read in the text). Read 4 Oct 2026, in part: four sentences of 34,900 characters.

  2. “The Snapdragon 855 and Movidius are onboard the International Space Station supported by the Spaceborne Computing-2 by Hewlett Packard Enterprise.”

    JPL Artificial Intelligence Group: the developer. Artificial Intelligence and Advanced Flight Computing on the International Space Station, JPL Artificial Intelligence Group. Published undated (no date on the page). Read 4 Oct 2026, in part: four sentences of 4,100 characters.

  3. “We benchmark a suite of flight software (FSW) applications on the Qualcomm Snapdragon 855 handheld Development Kit (HDK), a high performance embedded processor used in mobile phones.”

    JPL Artificial Intelligence Group: the developer. Artificial Intelligence and Advanced Flight Computing on the International Space Station, JPL Artificial Intelligence Group. Published undated (no date on the page). Read 4 Oct 2026, in part: four sentences of 4,100 characters.

Notes

The models were trained on the ground and their inputs were prepared in advance; this is a test of processors, not of a model at work. The release about Dynamic Targeting says the same team ran similar algorithms on the station in 2022, which this appears to be.

Changed

Added on 4 October 2026, in the second round of reading.

Cite or share

Neural networks benchmarked on phone-class chips aboard the International Space Station. Every AI that has run above the Kármán line, graded by evidence. Reviewed 4 Oct 2026. https://karman-si.pages.dev/r/iss-ml-benchmark/

Address: https://karman-si.pages.dev/r/iss-ml-benchmark/

Kármán register: R2 A badge with this record's marks, at /badge/iss-ml-benchmark.svg. This record as data.