02-12-2024, 10:11 AM
(This post was last modified: 02-12-2024, 10:29 AM by arkansawdave74.)
I've been using hashcat for a few years now and I usually use it with older hardware; usually so old that it's hard to find the correct cuda version to run....so, I usually don't even install cuda. Now, though, I finally have a GPU that's just barely EOL...lol. a Quadro m2000m. I'm testing things out and I'm so happy with it's speed!! (speed 2 is my Core i7-6820HQ)
```
Speed.#1.........: 50022 H/s (10.49ms) @ Accel:32 Loops:128 Thr:128 Vec:1
Speed.#2.........: 8199 H/s (6.92ms) @ Accel:64 Loops:512 Thr:1 Vec:8
Speed.#*.........: 58221 H/s
```
But I keep seeing this warning
```
Failed to initialize NVIDIA RTC library.
* Device #1: CUDA SDK Toolkit not installed or incorrectly installed.
CUDA SDK Toolkit required for proper device support and utilization.
Falling back to OpenCL runtime.
```
So, since I have a more current GPU, I decide to do it right and find out how much faster it can go with cuda installed. I installed cuda from the arch extra repo.
```
╭─user@zbook in ~ as 慄 took 110ms
╰─λ pacman -Q cuda
cuda 12.3.1-3
```
I then run hashcat, eager to see how much faster it will crack.....it initializes and I don't see the warning....instead I see that it's going to run with the Cuda API!!
```
CUDA API (CUDA 12.3)
* Device #1: Quadro M2000M, 3196/4042 MB, 5MCU
```
AWESOME!! I must've got the right package, I'm thinking....Then it starts cracking...
```
Speed.#1.........: 27894 H/s (8.25ms) @ Accel:8 Loops:128 Thr:256 Vec:1
Speed.#2.........: 8148 H/s (6.90ms) @ Accel:32 Loops:1024 Thr:1 Vec:8
Speed.#*.........: 36042 H/s
```
What the heck happened!?? It cut my nvidia cracking speed in half!! I've been around and seen enough recommendations by hashcat to KNOW that cuda will be faster than opencl. But it's not. Why all the pushing people to get the HUGE cuda when it only hurts your performance. Not hurts....DESTROYS your performance. I'm so surprised by that. Does it do that to everyone's performance? I bet a lot of people have no idea they could crack much faster (twice as fast) without cuda installed. Honestly....why does it warn and recommend so strongly something that kills your speed so badly?
I sure hope that all I have to do is uninstall cuda to regain my speed....
*****************************************************************************************************************
EDIT: Ooops. This is embarassing...I read thew cuda install, which said I needed to logout for the binaries to be in $PATH. So, I logged out and also installed the optional packages and then tried again.
```
Speed.#1.........: 50004 H/s (9.58ms) @ Accel:16 Loops:128 Thr:256 Vec:1
Speed.#2.........: 10319 H/s (11.39ms) @ Accel:128 Loops:512 Thr:1 Vec:8
Speed.#*.........: 60322 H/s
```
All I can do now is shutup, I guess....and say Thank You!!
But, really, that's just a hair faster. I czn achieve 49,500h/s without that huge 5gigs of cuda installed. Is there something I'm missing?
```
Speed.#1.........: 50022 H/s (10.49ms) @ Accel:32 Loops:128 Thr:128 Vec:1
Speed.#2.........: 8199 H/s (6.92ms) @ Accel:64 Loops:512 Thr:1 Vec:8
Speed.#*.........: 58221 H/s
```
But I keep seeing this warning
```
Failed to initialize NVIDIA RTC library.
* Device #1: CUDA SDK Toolkit not installed or incorrectly installed.
CUDA SDK Toolkit required for proper device support and utilization.
Falling back to OpenCL runtime.
```
So, since I have a more current GPU, I decide to do it right and find out how much faster it can go with cuda installed. I installed cuda from the arch extra repo.
```
╭─user@zbook in ~ as 慄 took 110ms
╰─λ pacman -Q cuda
cuda 12.3.1-3
```
I then run hashcat, eager to see how much faster it will crack.....it initializes and I don't see the warning....instead I see that it's going to run with the Cuda API!!
```
CUDA API (CUDA 12.3)
* Device #1: Quadro M2000M, 3196/4042 MB, 5MCU
```
AWESOME!! I must've got the right package, I'm thinking....Then it starts cracking...
```
Speed.#1.........: 27894 H/s (8.25ms) @ Accel:8 Loops:128 Thr:256 Vec:1
Speed.#2.........: 8148 H/s (6.90ms) @ Accel:32 Loops:1024 Thr:1 Vec:8
Speed.#*.........: 36042 H/s
```
What the heck happened!?? It cut my nvidia cracking speed in half!! I've been around and seen enough recommendations by hashcat to KNOW that cuda will be faster than opencl. But it's not. Why all the pushing people to get the HUGE cuda when it only hurts your performance. Not hurts....DESTROYS your performance. I'm so surprised by that. Does it do that to everyone's performance? I bet a lot of people have no idea they could crack much faster (twice as fast) without cuda installed. Honestly....why does it warn and recommend so strongly something that kills your speed so badly?
I sure hope that all I have to do is uninstall cuda to regain my speed....
*****************************************************************************************************************
EDIT: Ooops. This is embarassing...I read thew cuda install, which said I needed to logout for the binaries to be in $PATH. So, I logged out and also installed the optional packages and then tried again.
```
Speed.#1.........: 50004 H/s (9.58ms) @ Accel:16 Loops:128 Thr:256 Vec:1
Speed.#2.........: 10319 H/s (11.39ms) @ Accel:128 Loops:512 Thr:1 Vec:8
Speed.#*.........: 60322 H/s
```
All I can do now is shutup, I guess....and say Thank You!!
But, really, that's just a hair faster. I czn achieve 49,500h/s without that huge 5gigs of cuda installed. Is there something I'm missing?