About benchmark and actual speed

[hijack]

大家好，我正在尝试破解我的pdf密码，基准3x1080ti速度达到55021.8 kH/s，但实际运行只有10118.1 kH/s，我使用的是15G字典文件。破解速度受ssd读取速度影响吗？如何提高破解速度？

【hashcat -m 10500 hash.txt password.txt -w3】

会话............：哈希猫
状态…………：退出
哈希模式............：10500（PDF 1.4 - 1.6 (Acrobat 5 - 8)）
哈希.目标......：$pdf$4*4*128*-12*0*16*a47c7b41f77a6d3ba83cd5abbba56...dd76e3
Time.Started.....：2022 年 4 月 21 日星期四 13:09:04（10 秒）
Time.Estimated...: Thu Apr 21 13:11:16 2022 (2 mins, 2 secs)
Kernel.Feature...: Pure Kernel
Guess.Base.......: File (password.txt)
Guess.Queue......: 1/1 (100.00%)
Speed.#1.........:  3275.2 kH/s (35.50ms) @ Accel:1024 Loops:70 Thr:32 Vec:1
Speed.#2.........:  3429.2 kH/s (36.59ms) @ Accel:1024 Loops:70 Thr:32 Vec:1
Speed.#3.........:  3413.7 kH/s (36.32ms) @ Accel:1024 Loops:70 Thr:32 Vec:1
Speed.#*.........: 10118.1 kH/s
Recovered........: 0/1 (0.00%) Digests
Progress.........: 87180206/1328779672 (6.56%)
Rejected.........: 17326/87180206 (0.02%)
Restore.Point....: 85345198/1328779672 (6.42%)
Restore.Sub.#1...: Salt:0 Amplifier:0-1 Iteration:0-70
Restore.Sub.#2...: Salt:0 Amplifier:0-1 Iteration:0-70
Restore.Sub.#3...: Salt:0 Amplifier:0-1 Iteration:0-70
Candidate.Engine.: Device Generator
Candidates.#1....: 47527935 -> 48448809
Candidates.#2....: 48448810 -> 49369423
Candidates.#3....: 46606852 -> 47527934
Hardware.Mon.#1..: Temp: 33c Fan: 29% Util: 34% Core:1479MHz Mem:5005MHz Bus:8
Hardware.Mon.#2..: Temp: 31c Fan: 29% Util: 29% Core:1314MHz Mem:5005MHz Bus:8
Hardware.Mon.#3..: Temp: 35c Fan: 23% Util:  0% Core:1322MHz Mem:5005MHz Bus:16

---------------------------------------------------------------------------------------------------------------------------------
【hashcat -b -m 10500 -w 3】


hashcat (v6.2.5) starting in benchmark mode

CUDA API (CUDA 11.6)
====================
* Device #1: NVIDIA GeForce GTX 1080 Ti, 10227/11263 MB, 28MCU
* Device #2: NVIDIA GeForce GTX 1080 Ti, 10228/11263 MB, 28MCU
* Device #3: NVIDIA GeForce GTX 1080 Ti, 10228/11263 MB, 28MCU

OpenCL API (OpenCL 3.0 CUDA 11.6.127) - Platform #1 [NVIDIA Corporation]
==================================================== =======================
* 设备 #4：NVIDIA GeForce GTX 1080 Ti，跳过
* 设备 #5：NVIDIA GeForce GTX 1080 Ti，跳过
* 设备 #6：NVIDIA GeForce GTX 1080 Ti，跳过

OpenCL API (OpenCL 3.0 CUDA 11.6.127) - 平台 #2 [NVIDIA Corporation]
==================================================== =======================
* 设备 #7：NVIDIA GeForce GTX 1080 Ti，跳过
* 设备 #8：NVIDIA GeForce GTX 1080 Ti，跳过
* 设备 #9：NVIDIA GeForce GTX 1080 Ti，跳过

基准相关选项：
============================
* --工作负载配置文件=3

-------------------------------------------------- ----------------
* Hash-Mode 10500 (PDF 1.4 - 1.6 (Acrobat 5 - 8)) [迭代次数：70]
-------------------------------------------------- ----------------

速度.#1.......: 19076.8 kH/s (32.80ms) @ Accel:1024 Loops:70 Thr:32 Vec:1
速度.#2.......: 19166.2 kH/s (32.61ms) @ Accel:1024 Loops:70 Thr:32 Vec:1
速度.#3.......: 19000.7 kH/s (33.11ms) @ Accel:1024 Loops:70 Thr:32 Vec:1
速度.#*.......: 57243.7 kH/s

开始时间：2022 年 4 月 21 日星期四 13:11:52
停止：2022 年 4 月 21 日星期四 13:11:57

[Snoopy]

try using option -O for optimized kernel

benchmark is "best or maximum" speed expected, but you will mostly never reach this 

it seems you are using a plain dict with passes 8 digits in length only, pure wordlist will never reach full speed due to the way hashcat works, inner outer loop, rules could speed up your cracking run depending on your wordlist

if its really digits only you could use a dict with 6 digits only and a ruleset with append 00 to 99, this should speed up your cracking a little bit

[hijack]

try using option -O for optimized kernel

benchmark is "best or maximum" speed expected, but you will mostly never reach this 

it seems you are using a plain dict with passes 8 digits in length only, pure wordlist will never reach full speed due to the way hashcat works, inner outer loop, rules could speed up your cracking run depending on your wordlist

if its really digits only you could use a dict with 6 digits only and a ruleset with append 00 to 99, this should speed up your cracking a little bit

Thanks, I am using a complex and changeable password dictionary and most of them have no rules. Can I only use rules to improve the cracking speed?

[Snoopy]

well yes,

as mentioned before you can add

Code:

-O

to your command line to use optimized kernel

you can also try adding

Code:

-r best64.rule

, this will add some basic rules to your wordlists, generating your words + modifications

[hijack]

well yes,

as mentioned before you can add

Code:

-O

to your command line to use optimized kernel

you can also try adding

Code:

-r best64.rule

, this will add some basic rules to your wordlists, generating your words + modifications

Thanks, -O seems to have no effect. After using the best64.rule rule, the speed is significantly improved, and the time required for cracking is also significantly improved