Skip to content
Quantum.world

The Signal

Pasqal brings neutral-atom QPUs into Slurm’s HRES scheduler

Pasqal says Slurm 26.11 can manage its neutral-atom quantum processors as hierarchical resources alongside CPUs and GPUs. The company says the integration has been tested in a simulated on-premises setup and Pasqal Cloud, while CINECA is running QPU jobs through an earlier Slurm route.

Pasqal brings neutral-atom QPUs into Slurm’s HRES scheduler

QPU scheduling moves into Slurm’s own framework

Pasqal says Slurm 26.11 can treat a quantum processing unit (QPU) as a resource it schedules and manages. In a technical post dated November 16, the company describes how Slurm’s new plugin interface for hierarchical resources (HRES), announced as an expansion by NVIDIA, works with the Quantum Resource Management Interface (QRMI) so that Slurm manages access to Pasqal QPUs around each job.

What changes for a computing site

Pasqal says early integrations of quantum hardware with Slurm used licences to track consumption and SPANK plugins to run code and modify a job’s environment, because licences execute nothing and SPANK plugins give the QPU no native consumable representation in Slurm. The HRES features let the company move that logic into Slurm itself.

According to the post, the scheduler knows whether the QPU is available, so jobs wait in the queue while it is down for calibration or maintenance rather than starting and failing. Access is set up before user code runs and removed afterwards: for an on-premises QPU, the plugin opens a session for each job and ends it when the job finishes or is cancelled. Administrators set in one configuration file how many jobs may use each QPU at once.

The HRES–QRMI integration was developed by IBM and Pasqal and works through QRMI, which Pasqal describes as one small interface across quantum backends, so the same plugin serves both IBM and Pasqal QPUs. The new HRES features require Slurm 26.11 or later.

Tests, and what runs in production today

Pasqal ran Pulser and NVIDIA CUDA-Q versions of a small Rydberg-blockade test through Slurm with HRES and QRMI, against Pasqal Cloud’s EMU_FREE emulator and against a simulated on-premises QPU deployment; all four combinations passed. It also checked the job lifecycle a shared QPU needs, including that competing jobs waited their turn and that cancelling a job released its access. Beyond those tests, the company says it ran the workflow on a physical QPU, FRESNEL_CAN1, accessed through Pasqal Cloud.

The post is explicit about the limit: the QRMI SPANK plugin is the production route today. CINECA hosts Sol, a Pasqal QPU integrated with its Slurm cluster through QRMI and Pasqal’s Warden service, and has run jobs on it with the SPANK plugin; Pasqal says CINECA’s experience is from that deployment, not from HRES. Sites can start with the plugin and move to HRES as it matures.

From GPU emulation to the QPU

Developers can describe neutral-atom sequences with Pasqal’s software such as QoolQit, Pulser or its application libraries, or write analog programs directly in CUDA-Q. Pasqal says CUDA-Q 0.17 adds GPU emulation of its devices, so one batch workflow can run a program on a GPU first and then, only if that succeeds, on the QPU. In the company’s test on its own cluster, the GPU emulation and the QPU runs agreed within the statistical spread at every one of six atom distances.

Source

  1. Pasqal

Companies in this story

Related stories

Explore

More articles