Hello Everyone,
Before diving into the issue at hand, let me provide some context about my setup:
Cloud Platform: AWS
Instance Type: p2.16xlarge
Operating System: Ubuntu Server 22.04 LTS (HVM), SSD Volume Type (See attached image: config.png)
GPU: Nvidia Tesla K80 (16 GPUs)
Nvidia Driver: Version 470.199.02
CUDA Version: 11.4 (See attached image: my_drivers.png)
Hashcat Version: 6.2.4 (See attached image: output.png)
After considerable effort, I successfully installed both the Nvidia driver and the CUDA toolkit. However, I've been unable to get Hashcat to run properly on this setup. Despite multiple reboots and re-installations of the CUDA toolkit—both manually and via
apt
—I consistently encounter the following error:
Code:
hashcat nvrtc: error: invalid value for --gpu-architecture (-arch)
I also attempted to use an older version of Hashcat (6.2.3) but faced the same issue (See attached image:error.png ).
So far, I've tried various fixes, including:
Running apt remove libnvrtc*
Manually uninstalling all Nvidia drivers and reinstalling CUDA directly using the .run file
Unfortunately, none of these solutions have resolved the issue.
I would greatly appreciate any advice or suggestions you may have. Thank you in advance for your assistance.
The GPUs are you trying to use are very old and I'm honestly not sure you can even use them at all anymore as no driver new enough for Hashcat's minimum spec will support them, at least to my knowledge. Further, the Hashcat version(s) you are using is(are) also very old and you should ensure you are using the latest release of hashcat. Unfortunately, those GPUs are terrible even if you do manage to get them working, their architecture lacks support for instructions that we make heavy use of and that makes them extremely slow, even compared to similarly powerful GPUs from around the same time. I would strongly recommend using the latest hashcat release(or building from github) as well as using a more modern GPU. AWS supports many other GPUs and all of them are significantly faster and/or better supported, while not necessarily being more expensive.
(09-21-2023, 09:06 PM)Chick3nman Wrote: The GPUs are you trying to use are very old and I'm honestly not sure you can even use them at all anymore as no driver new enough for Hashcat's minimum spec will support them, at least to my knowledge. Further, the Hashcat version(s) you are using is(are) also very old and you should ensure you are using the latest release of hashcat. Unfortunately, those GPUs are terrible even if you do manage to get them working, their architecture lacks support for instructions that we make heavy use of and that makes them extremely slow, even compared to similarly powerful GPUs from around the same time. I would strongly recommend using the latest hashcat release(or building from github) as well as using a more modern GPU. AWS supports many other GPUs and all of them are significantly faster and/or better supported, while not necessarily being more expensive.
Thank you for your prompt response. Could you recommend the most suitable AWS machine types that are compatible with the latest Hashcat release? While browsing through AWS, I wasn't able to easily identify the GPU options they offer.
P5, G5, P4, P3, and maybe G3/G4dn/G5g instances, roughly in that order. You can see the types of instances and their hardware configurations here: https://aws.amazon.com/ec2/instance-types/
(09-21-2023, 09:29 PM)Chick3nman Wrote: P5, G5, P4, P3, and maybe G3/G4dn/G5g instances, roughly in that order. You can see the types of instances and their hardware configurations here: https://aws.amazon.com/ec2/instance-types/
