Full Breakdown
HPE to Run First High-Performance Computing Benchmarks on the Moon
By Drooid · · How we work
Lunar HPC Benchmark Mission
Hewlett Packard Enterprise (HPE) will fly its fourth-generation Spaceborne Computer-4 to the Moon’s south-pole region in late 2026. The fan-less edge module will ride Astrolab’s FLIP rover on Voyager Technologies’ Griffin-1 lander under NASA’s CLPS program. It houses a low-power CPU, GPU and two mirrored 1 TB KIOXIA BG6 SSDs in a self-contained enclosure.
During surface operations the system will run the High-Performance LINPACK (HPL) and High-Performance Conjugate Gradient (HPCG) benchmarks—the first such runs off Earth. Additional workloads include large-language-model inference, LiDAR terrain mapping, and NASA’s FLUID experiment, which coordinates neural-network models trained on data from Earth, the Moon and the ISS. Deloitte will field its Silent Shield cyber-anomaly detection payload.
Background & Context
HPE’s Spaceborne program has flown commercial silicon on the ISS since 2017. Chief scientist Mark Fernandez notes the fleet has recorded no SSD failures since 2019, with early issues traced to capacitor defects rather than radiation. The lunar surface temperature swings from roughly 80 °C in sunlight to about -200 °C in shadow, and radiation levels are estimated at nearly three times those on the ISS.
Processing data locally mitigates the ~2.5 s one-way signal delay and limited downlink bandwidth. For a rover navigating polar terrain, waiting for Earth-based analysis is impractical; onboard AI can generate actionable insights and transmit only concise results.
Data & Statistics
- Launch window: late 2026 (planned)
- Landing site: lunar south-pole region
- Storage: 2 × 1 TB KIOXIA BG6 SSDs in RAID-1
- Thermal design: operation from -200 °C to 80 °C
- Communication latency: ~5 s round-trip Earth-Moon
- Mission duration: MIT Technology Review cites a planned 10-day surface stay; HPE says the computer is expected to operate for the “life of the mission.”
- Power budget: rover rated for at least 150 h of lunar night operation
- Economic outlook: Futurum projects a $1 trillion AI-compute cap-ex market by 2030, with a 4.5× premium for orbital deployments versus terrestrial edge installations.
Official Statements & Responses
Fernandez emphasized that the design relies on “software-based hardening,” with the module monitoring hardware metrics and autonomously throttling, rebooting, or powering down when thresholds are exceeded. NASA describes the FLUID experiment as a test of coordination among three neural-network models running on Earth, the Moon and the ISS.
Conflicting Reports & Gaps
Sources differ on the operational length of the lunar stay. MIT Technology Review reports a 10-day surface mission, while HPE suggests the computer will function for the entire mission, leaving uncertainty about performance through a full 14-day lunar night. The specific CPU and GPU models remain undisclosed; speculation ranges from NVIDIA to AMD or smartphone-class silicon, but no vendor has confirmed the components. Benchmark performance expectations (e.g., FLOPS) are also absent, making assessment impossible until post-flight data are released.
What’s Next
- Launch: Griffin-1 lander scheduled for late 2026.
- Surface operations: Execution of HPL and HPCG benchmarks, followed by a TOP500 ranking if successful.
- Scientific experiments: Completion of NASA’s FLUID federated-learning test and Deloitte’s Silent Shield run.
- Post-mission reporting: HPE is expected to disclose benchmark results, hardware specifications and durability findings, including operation through the full lunar night.
These steps will provide the first empirical evidence of commercial silicon’s ability to perform high-performance computing and AI inference in the Moon’s extreme environment, informing future lunar and deep-space missions.