Cracking at Extreme Scale
Jeremi Gosney discusses building and managing multi-GPU supercomputing clusters engineered for massive-scale password recovery.
Summary
Jeremi Gosney breaks down the engineering challenges and architecture required to build and operate massive GPU password-cracking clusters at extreme scale. Drawing on years of designing high-performance computing clusters, he details the physical, thermal, electrical, and hardware pitfalls encountered when scaling from single multi-GPU rigs to distributed multi-node clusters.
Gosney covers driver configuration headaches and OpenCL kernel execution nuances across heterogeneous GPU architectures at 13:10 and 26:25. He explores the software layer at 20:55, explaining the necessity of distributed job scheduling, hashlist distribution across nodes, workload balancing, and fault tolerance when individual GPU cards crash or overheat. At 32:00, he describes automated remote cluster provisioning, monitoring frameworks, and managing multi-tenant cluster access without compromising security or job integrity.
(AI summary — Gemini Flash)
Related talks
Password Village·24 min
PNI
An introductory guide to setting up and running Hashcat for GPU-accelerated password recovery and hash analysis.
Password Village·24 min
PNI
A walkthrough demonstrating core Hashcat attack modes, rule manipulation, and dictionary optimization.
Password Village·24 min
PNI
Covers essential Hashcat fundamentals, hardware benchmarking, attack flags, and hash cracking workflows.