Using CPU and GPU in a cloud instance..
#1
Hi folks,

been using hashcat with good results on my home gaming rig (decently fast quad core Intel CPU and a GTX 1080), however, concerns about proper handling of customer data means I need to move away from the current BYOD model.

As a solution, I'm using hashcat 5.1.0 on an Azure NV24 series VM instance (24 CPU vCores, 4 Tesla M60 GPUs).  It's Ubuntu 18.04 with the Microsoft/Nvidia drivers installed. (accoridng to https://docs.microsoft.com/en-us/azure/v...iver-setup ) all nicely recognised by nvidia-smi

The GPU performance is bearable - I can get just slightly more out of four Tesla vGPUs as I can out of my physical 1080..

OpenCL Platform #1: NVIDIA Corporation
======================================
* Device #1: Tesla M60, 2032/8129 MB allocatable, 16MCU
* Device #2: Tesla M60, 2032/8129 MB allocatable, 16MCU
* Device #3: Tesla M60, 2032/8129 MB allocatable, 16MCU
* Device #4: Tesla M60, 2032/8129 MB allocatable, 16MCU

Benchmark relevant options:
===========================
* --optimized-kernel-enable

Hashmode: 1000 - NTLM

Speed.#1.........: 18779.7 MH/s (27.20ms) @ Accel:64 Loops:512 Thr:1024 Vec:2
Speed.#2.........: 18357.1 MH/s (27.88ms) @ Accel:64 Loops:512 Thr:1024 Vec:2
Speed.#3.........: 18856.1 MH/s (27.08ms) @ Accel:64 Loops:512 Thr:1024 Vec:2
Speed.#4.........: 19772.7 MH/s (26.82ms) @ Accel:64 Loops:512 Thr:1024 Vec:2
Speed.#*.........: 75765.5 MH/s



My concern is - I can't get hashcat to recognise the CPU cores. tried installing an OpenCL sdk for Intel (from here; https://launchpad.net/~intel-opencl/+arc...tel-opencl ), without joy.

The GPU performance is okay - but given the pricing model in Azure, it seems wrong to leave 24 cores sat doing nothing and paying for them.

any ideas folks?
Reply