Hashcat speed benchmark vs recovery

[LittleKita]

Hi,

i have a Nvidia TITAN Xp.

I did various tests with md5 to compare speeds.

Code:

hashcat --benchmark -D 2 -m 0 -d 2

Speed.#2.........: 39199.0 MH/s (50.93ms) @ Accel:64 Loops:1024 Thr:1024 Vec:8

# 1 hash
hashcat -D 2 -d 2 -O --status --status-timer=1 --potfile-disable --markov-disable --restore-disable -m 0 -a 3 9dc5f758e8e773980f07a95bfc8642f1 "?a?a?a?a?a?a"
Speed.#2.........: 34478.8 MH/s (6.43ms) @ Accel:32 Loops:256 Thr:1024 Vec:8

# 32 hash
hashcat -D 2 -d 2 -O --status --status-timer=1 --potfile-disable --markov-disable --restore-disable -m 0 -a 3 hashlists/32.hash "?a?a?a?a?a?a"
Speed.#2.........: 21318.8 MH/s (10.91ms) @ Accel:32 Loops:256 Thr:1024 Vec:8

# 1k hash
hashcat -D 2 -d 2 -O --status --status-timer=1 --potfile-disable --markov-disable --restore-disable -m 0 -a 3 hashlists/1k.hash "?a?a?a?a?a?a"
Speed.#2.........: 21358.2 MH/s (10.96ms) @ Accel:32 Loops:256 Thr:1024 Vec:8

# 1m hash
hashcat -D 2 -d 2 -O --status --status-timer=1 --potfile-disable --markov-disable --restore-disable -m 0 -a 3 hashlists/1m.hash "?a?a?a?a?a?a"
Speed.#2.........: 15239.9 MH/s (7.67ms) @ Accel:64 Loops:64 Thr:1024 Vec:8

Edit: change "-d 1" to "-d 2".

Why are the numbers so different from the benchmark?
Have I possibly used the wrong parameters?
Or is it the host system?

Regards,
  LittleKita

[Mem5]

You don't use the same devices :

--benchmark -D 2 -m 0 -d 2

# 1 hash
hashcat -D 2 -d 1

And obviously, the more hashes you give, the less speed you get. Benchmark is done with 1 hash.

[LittleKita]

Thanks,

sorry "-d 1/2" was copy/paste problem.

1 vs 32 hashes only the half of the speed?
I thought the hash computation don't increase only the compares.
I don't understand why the speed goes down.

Regards,
  LittleKita

[undeath]

You can gain a litte higher cracking rates by adding -w3. This is what benchmark uses.

As to 1 vs 32: you can apply significant optimizations with only a single target hash that are not possible with multiple hashes.

[LittleKita]

Thanks,

I've tried "-w 3" but got only minimal improvement.

I think the benchmark is only for GPU comparison,
not for realistic rates to restore hash's.
These depend more on the parameters, e.g. hash count/masks/dictionary.

Regards,
  LittleKita