hashcat Forum

Full Version: ERROR: cuMemAllocHost() 2
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
Pages: 1 2
Getting error "ERROR: cuMemAllocHost() 2" when running benchmark (-b). Other hashcat tests can complete fine.

Running Windows 10 64-bit and latest Nvidia drivers (355.82). Titan X cards.

Have applied the Timeout registry fix.

Any ideas?
How much available host memory do you have?
(09-08-2015, 07:59 AM)epixoip Wrote: [ -> ]How much available host memory do you have?

32 GB
32GB total, or 32GB free? And how many Titan X do you have?

cuMemAllocHost is returning CUDA_ERROR_OUT_OF_MEMORY so you're definitely running out of host memory at some point.
(09-08-2015, 12:16 PM)epixoip Wrote: [ -> ]32GB total, or 32GB free? And how many Titan X do you have?

cuMemAllocHost is returning CUDA_ERROR_OUT_OF_MEMORY so you're definitely running out of host memory at some point.

We have 4 Titan X cards in this machine.

Is it "normal" to run out of 32 GB's of host-memory with hashcat? Anything I can tweak?
You're not answering the most important question: 32GB total, or 32GB free? Are you observing the memory utilization while running hashcat?

It's not "normal," but it all depends on what else is running on the system and what exactly you're doing. If you have other RAM-hungry processes running, or built the system without swap space, or are working on a large hash list with lots of salts, then yeah, that can certainly happen.

Also remember that the Titan X is a 12GB card, and while 32GB should be fine for four Titan X in a dedicated cracking rig, it could in fact want to allocate 48GB of host memory for some attack. Just depends on what you're doing.
(09-09-2015, 09:49 PM)epixoip Wrote: [ -> ]You're not answering the most important question: 32GB total, or 32GB free? Are you observing the memory utilization while running hashcat?

It's not "normal," but it all depends on what else is running on the system and what exactly you're doing. If you have other RAM-hungry processes running, or built the system without swap space, or are working on a large hash list with lots of salts, then yeah, that can certainly happen.

Also remember that the Titan X is a 12GB card, and while 32GB should be fine for four Titan X in a dedicated cracking rig, it could in fact want to allocate 48GB of host memory for some attack. Just depends on what you're doing.

Once my current burn-in test is done, I can observe the memory consumption, but this system is running nothing else. Clean & default install of W10. Totally dedicated for hashcat and please note, that the error only appears when trying the "hashcat -b/--benchmark" option - within a few seconds.

I don't know what the benchmark option does "behind the scenes", but I'm not specifying any hash lists with/without salts etc.
A lot of people on the forums have reported issues and crashes on Windows 10. Might want to try another OS.
(09-10-2015, 04:29 AM)epixoip Wrote: [ -> ]A lot of people on the forums have reported issues and crashes on Windows 10. Might want to try another OS.

While I appreciate your help, I don't think the OS is the reason. I've been running hashcat on W10 (TP) for months on various systems without any issues, and on this particular system I've succesfully cracked a great number of hashes. It's performened perfectly on my manual request, like NTML/MD5 bruteforce, wordlist and hybrid attacks.

Something in the benchmark function makes this error occur and I'd like to get closer to the actual root cause.

And please don't get me wrong. I'm freaking agnostic - also when it comes to operating systems.
Ok. Well in the meantime I would advise observing memory utilization while the benchmark is running, maybe you can pinpoint which algorithm is causing you to run out of RAM. Or if there's just a memory leak in general.
Pages: 1 2