• 10TB/s+ read/write, 100M+ IOPS • 700TB+ capacity, 100B+ files • 20k+ clients, 100k+ GPUs • Scale-out, fully parallel data + metadata • Robust for largest HPC & GPU clusters Efficient Utilization of Compute Nodes • Optimized for large multi-CPU, multi-NIC clients • 2TB+ RAM, 100+ cores, NUMA-aware I/O • Saturate multiple 400Gbps NICs, LNet Multi-rail • Minimal CPU Overhead, RDMA, GPU Direct Flexible Storage Configuration • Storage type aware - TLC/QLC NVMe, HDD • Direct client access - no need for tiering • Utilize client NVMe for local-caching • Re-export via NFS, SMB, S3 protocols • Archive data to external S3, GCS, Tape, … Secure Isolation and Encryption • Secure sub-directory isolation • Proven Data Encryption via AES256/fscrypt • Strong node/user authentication Kerberos, Shared Secret-Key • Fine-grained client administrator controls • Nodemap isolation of clients, even with root