Nvidia's recent acquisition of SchedMD, the company responsible for Slurm - an open-source workload manager that schedules computing tasks on supercomputers and AI clusters - has stirred debate among artificial intelligence specialists and supercomputing practitioners. The deal, announced in December, gives Nvidia ownership of software that SchedMD says helps run about 60% of the world’s supercomputers.
Slurm is widely used to coordinate the intensive computation required to train large language models that underpin chatbots and other AI applications. It also operates on government supercomputers used for weather forecasting and defense-related work. That ubiquity is part of the reason some engineers and executives are closely watching Nvidia’s next moves, sources said.
Five people interviewed for this report expressed unease that Nvidia could subtly tilt Slurm development in ways that advantage its own chips. Specific concerns include the company prioritizing software updates for Nvidia hardware ahead of integrating support for rival processors from companies such as Advanced Micro Devices or Intel. Those worries were voiced by three individuals in the AI industry and two with knowledge of supercomputer operations - all of whom have worked with or developed systems that include non-Nvidia hardware.
One engineer who has extensive experience with Slurm suggested that an early signal will be how quickly Nvidia incorporates forthcoming AMD chips into Slurm’s code base relative to how rapidly it optimizes the software for its own technologies, such as InfiniBand networking products. That person framed these timelines as a possible test of Nvidia’s long-term intentions.
For some users the acquisition is also a potential upside. Observers said Nvidia, as the most valuable publicly traded technology company, might channel resources into modernizing Slurm, which was originally built for government supercomputing environments and has seen broader adoption among leading AI firms. Addison Snell, CEO of Intersect360 Research, noted Nvidia could help SchedMD’s customers - especially government laboratories - adopt newer AI methods alongside conventional supercomputing workloads.
Even so, concern persists that Nvidia might, over time, produce updates or optimizations that make Slurm run noticeably better on Nvidia gear than on competing hardware. That scenario echoes unease tied to Nvidia’s 2022 purchase of Bright Computing. Industry sources said Bright’s software, while usable with non-Nvidia hardware, had been optimized for Nvidia in practice, introducing performance penalties for users of other chips unless additional effort was made. Nvidia has disputed that characterization, saying Bright technology supports "nearly any CPU or GPU-accelerated cluster."
In public statements, Nvidia has stressed its intention to keep Slurm open-source and available to the community. The company said: "Customers everywhere benefit from our open source and free software. Slurm is open-source and we continue to provide enhancements for everyone." When announcing the SchedMD acquisition, Nvidia also stated a commitment to developing and widely distributing what it described as "open-source, vendor-neutral software."
Nvidia added that it encourages others to contribute to the ecosystem of free and open-source software and pointed to a track record of maintaining free and improved offerings after acquiring open-source projects. The company said it will continue to provide open-source software support, training and development for Slurm to SchedMD’s hundreds of customers.
Some organizations that use Slurm include major AI labs and research groups. The software is in use at Meta Platforms, French AI startup Mistral, and Anthropic for certain tasks, including elements of AI training, according to people familiar with deployments. OpenAI uses a different approach based on technology developed by Google, a spokesperson said. Anthropic, Mistral and Meta did not respond to requests for comment.
Not all experts who use SchedMD’s tools expressed alarm. Several said they were aware of concerns and were monitoring Nvidia’s stewardship closely. Many in the supercomputing and AI communities view the acquisition as a test case for whether Nvidia will maintain true vendor neutrality for software that plays a central role in both national lab and commercial AI operations.
Those watching the company said concrete actions - such as the speed and manner of integrating non-Nvidia chips into Slurm, and the balance of development resources toward general improvements versus vendor-specific optimizations - will be key indicators in the months and years ahead.
Summary
Nvidia’s purchase of SchedMD, owner of the widely used Slurm scheduler, has prompted scrutiny from AI and supercomputing users concerned the company might favor its own hardware in future software developments. Others hope Nvidia will invest in needed updates. Nvidia has said Slurm will remain open-source and vendor-neutral.
Key points
- SchedMD’s Slurm software schedules computing tasks on roughly 60% of the world’s supercomputers, and is used in both government and commercial AI environments.
- Some AI and supercomputing professionals fear Nvidia could prioritize support and optimizations for its own chips, potentially disadvantaging AMD, Intel and other competitors.
- Nvidia says it will keep Slurm open-source and continue to provide enhancements and support to SchedMD’s customers, while some users are cautiously optimistic the company may accelerate development.
Risks and uncertainties
- Potential vendor preference - Slurm updates or optimizations might be implemented first for Nvidia hardware, affecting performance and interoperability for users of competing chips; this concerns the AI and data center sectors.
- Integration timelines - How quickly new non-Nvidia chips are integrated into Slurm compared with Nvidia-specific enhancements will serve as an early indicator of Nvidia’s neutrality; this impacts supercomputing and AI infrastructure procurement decisions.
- Historical precedent - Prior Nvidia acquisitions such as Bright Computing prompted similar worries about optimization bias, which could influence how public labs and commercial AI builders approach reliance on Slurm; this affects government computing and commercial AI operations.