- SLURM is a key cluster management tool used by AI researchers
- CoreWeave is making it easier to deploy SLURM on Kubernetes-based cloud infrastructure
- The move reflects where CoreWeave thinks it should play in the services realm
It’s hard to take something called SLURM seriously. But the capabilities this silly acronym represents could be critical to realizing AI’s potential – and neocloud CoreWeave wants to be the place the SLURMber party happens.
SLURM stands for Simple Linux Utility for Resource Management. Basically, SLURM is a cluster management and job coordination platform for Linux systems, and it has become the de facto standard for AI researchers. But getting SLURM to work on top of modern AI compute infrastructure is tricky.
That’s why CoreWeave created SUNK or SLURM on Kubernetes. As the name implies, it allows SLURM workloads to more easily run in Kubernetes-based cloud environments.
There are two versions of CoreWeave’s platform: SUNK Self-Service, which makes it easier to set up a SLURM workspace in a simple, templatized way; and SUNK Anywhere, which allows workloads to be run even beyond CoreWeave’s infrastructure.
“As AI training footprints expand across clouds and customer-owned infrastructure, teams need speed to deploy without losing governance or creating operational fragmentation,” IDC VP Dave McCarthy said in a statement. That’s where SUNK can help.
Why though?
If you’re wondering why CoreWeave bothered to build what feels like a niche platform offering, we had the same question. Is this a go at moving up the stack from bare metal to new kinds of services? Is it something else entirely?
Corey Sanders, SVP of Product at CoreWeave, told Fierce the company is skating to where the puck is heading.
“As we’re seeing AI become much more democratized, I think there’s going to be AI researchers at almost every company in the world,” Sanders said.
While big AI labs (think the likes of OpenAI, Anthropic and Periodic Labs) are leading the charge today, Sanders noted “we are seeing an increasing demand from what you’d consider maybe classic enterprise or AI enterprise, where they have researchers and they’re doing training within their environment to be able to fine tune a model or get their own custom model deployed.”
Much ado has been made about the need for neoclouds to move up the stack to offer more services as a way to differentiate themselves and grow revenue going forward. CoreWeave’s decision to cater to AI researchers reflects exactly where it wants to play in the services realm.
“We are very focused on being AI first. We prioritize being the best [at] delivering AI capabilities and even when capabilities are more generic, we enable them in AI-specific ways,” Sanders explained.
“So when I think about where we will continue to deliver and grow and expand our services, it will be in the realm of delivering what’s needed for AI solutions and scenarios to be accomplished,” he concluded. “The broader hyperscale cloud set of services are not areas that we’re going to delve into unless they are critical for AI workloads.”