Amplifier

-n Manual workload tuning, set outerloop step size to X

-m Manual workload tuning, set innerloop step size to X

both are outdated

so -w that gordon mentioned before is the latest parameter to control wordload

Straight wordlist attack against MD5 will be faster on CPU than GPU - 2016 02 29 - hashcat forum

Does the PCI-Express speed have any influence on cracking speed ?

Actually, YES!

But in reality, NO !

Two types of hashes

• fast hashes
  • md5 SHA1
• slow hashes
  • typically design to run slow
  • md5crypt bcrypt scrypt

Slow hashes

doing iterations of the same function again and again and using the input of the previous computing to compute a new one

"This is just like amplifier"

Remembered that pregenerate dictionary is faster ?

Makes sense

Pre generate GPU workload：

• PCI-E is not bottle neck for slow hashes
• GPU has to run slow hash function which is just like amplifier

Dynamic generate GPU workload：

• PCI-E is also not bottle neck thank to the amplifier
• Amplifier run the generate function - slower
• Run slow hash functions which is jusk like amplifier

Hashcat CPU version

Install

Need some other dependencies

• glibc-devel.i686
• gmp-devel.i686

Execute command

Basically the same as GPU version.

One additional option:

--threads=N

Hashcat CPU threads

The relation between speed and threads is linear

• Good for time scheduling
• Easy to run in spare machines

Compare to GPU

CPU: 190 kH/s

GPU: 978 kH/s

Pre-built dictionary

Results

Wordlist: 978 kH/s

Combination: 585.7 kH/s
Rules: 106.3 Kh/s