"The setup we use consists of two rigs with 8 GPU each (16 total), and a management console. The rigs themselves run multiple open source containers where we can bring up 2, 4, or 8 GPU and assign them tasks from the open source work distribution console. Using GPU 2 and 3 generations behind currently available products, we achieved pre-fieldwork NTLM combined benchmarks of 240GHs testing NTLM via 12 character masks, and 25.6GHs via 10GB dictionary and a 3MB rules file. Actual speeds varied across multiple test configurations during the engagement."