Beehive Cluster User Guide

Welcome to the Beehive cluster, a high-performance computing environment with extensive GPU resources designed for machine learning and AI research at TTIC.

Cluster Overview

Beehive features a diverse collection of computing resources, with a focus on modern GPUs, including Blackwell, Ada and Ampere architecture cards (plus older Turing cards). The cluster has both general-access nodes and privately-owned resources.

Partitions

• low-gpu - the same nodes and 8-hour limit as gpu, but with a lower priority (5 vs 500); use it for work that can wait behind regular gpu jobs. The full partition list is on the Jobs page.

Key Specifications

• Slurm Version: 22.05.8 (Documentation)
• Operating System: Debian 12 (bookworm), Linux 6.1 kernel
• CUDA Version: 13.0 Update 2
• cuDNN Version: 9.14.0
• NVIDIA Driver Version: 595.104.02 rollout in progress; nodes not yet upgraded still run 580.95.05 - both support CUDA 13.0 Update 2

Node Reservations

• g22 - reserved for whc-own-gpu through 2026-10-14
• g2 - reserved for shiry-gpu through 2026-10-31

Maintenance Notes - October 2026

• Compute-node updates completed on all rollout nodes except g22: OS updates and NVIDIA driver 595.104.02 are installed on the updated GPU nodes. g22 will be updated after its current jobs finish.
• Head-node maintenance completed October 13. Scheduling, accounting, the dashboard, and mesh services are restored. Login and compute nodes were not rebooted.
• Login-node update - Monday morning, October 19 (CDT): cluster.ttic.edu will be updated and rebooted. SSH, tmux/screen sessions, and transfers will be interrupted; login-launched interactive jobs may also end. Save your work beforehand.