Automated password cracking with AWS GPU instances and CUDA hashcat
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.

Automated GPU-accelerated hashcat cracking on AWS


Automate using GPU-accelerated hashcat in the cloud, for fast and inexpensive cracking


One day, poking around on AWS, I saw someone had uploaded a CUDA Hashcat AMI, I started up an instance and did a quick test. I confirmed that the hashing speed corresponded with the benchmarks I found here.

The problem with GPU instances is that they're expensive. However on the spot market, they tend to be reasonably inexpensive (< $.07 / hr)

Initially I just set out to see how far I could get automating a spin up of a node to crack a password. Eventually I'd like to enable parallelism for even faster cracking.


  • Completely hands off on the AWS instance, with flexibility in hashcat options
  • Hashcat runs in screen so if necessary you can connect to the instance and see the status/progress
  • Runs all single Sha512 (unix crypt) hash against rockyou in ~21 minutes (~12000 H/s)


There's two critical scripts involved.

  • - Launches a spot instance, at this point, this is where most of the settings you care about are

  • - This gets pulled down and run by the instance after it spins up, this is where most of the actual automation/cracking takes place

However, there's some other components at play:


  • We start a spot instance request
  • We associate the runtime parameters with the spot instance request via Tags
  • If/when the instance is launched successfully, it will need to reference these tags


  • We store the results of cracked passwords in an s3 bucket (/incoming/)
  • We also store a copy of rockyou.txt and pull it down from S3 in (/assets/) [optional, but you'll want to look at]
  • We enable versioning on the bucket, so that overwrites (unlikely) are preserved [optional]
  • We enable a lifecycle policy on the bucket, so objects are automatically removed from /incoming/ after 2 days [optional]


  • We create an IAM policy for the hashcat instances. They grant the following 4 permissions:
    • Describe EC2 Instances (to find the spot instance request id)
    • Get tags (to retrieve the transient data from the spot instance request tags)
    • PutObject (on the /incoming/ folder of the S3 bucket)
    • GetObject (on the /assets/ folder of the S3 bucket)


  • We use the AWS CLI on the instnace to access the various AWS components

To Do

  • Automating parallelization
  • Bake AWS CLI into the AMI
  • Bake rockyou.txt into the AMI (?)
  • Provide some scripts to configure the IAM policy and S3 bucket
  • Maybe move more of the hardcoded configuration from to spot instance request tags (e.g., s3bucket, s3folder)
  • Migrate to SQS (or something) to reduce run-time configuration
  • Take better advantage of partial hours (if we only run for 20 minutes, use the other 40 minutes for something useful)
  • Progress reporting