oclHashcat and memory
#1
First, I just want to thank Atom for is answers in the other posts I made. I would have replied specifically but the threads are closed now. Updating the Cuda driver solved my problem for the missing OpenCl.dll and this computer also has a GTX260 and it runs 2 GPUs without problems. My new problem is that I am trying to run a combinator attack using a 427 Mb dictionary. When I start the attack, after a while, oclHashcat says that it has insufficient memory. In Windows task manager I have 2845 Mb of "Available" memory but of that 1077 is "Free" and 1828 is "Cached". When I start the attack, I see that it gives the error once it depleted the "Free" memory. It does not use the "Cached" one at all. Is it something that can be changed in the way oclHashcat interact with memory or something I have to change in Windows. I have a total of 4 Gb of RAM under Windows 7 x64. Thanks.
#2
Howdy, Mastercracker.
I dont think something might be done in Windows to overcome this issue.
Atom plans to add wordlist segmentation in the future.
Right now, the workaround is simple: split the wordlist(In half for instance), and run 2 attacks on each halves of it.It will be at most 30 sec. slower than attacking on one whole wordlist Big Grin
#3
(05-07-2010, 09:46 AM)Rolf Wrote: Howdy, Mastercracker.
I dont think something might be done in Windows to overcome this issue.
Atom plans to add wordlist segmentation in the future.
Right now, the workaround is simple: split the wordlist(In half for instance), and run 2 attacks on each halves of it.It will be at most 30 sec. slower than attacking on one whole wordlist Big Grin
That makes sense. I will do that in the mean time.
#4
Thanks to the author for the nice tool !
I have a question: What is the maximum supported dictionary size on VGA with 512MB RAM ?
#5
(05-14-2010, 09:53 AM)vector Wrote: Thanks to the author for the nice tool !
I have a question: What is the maximum supported dictionary size on VGA with 512MB RAM ?

the size depends on the vendor. if you're on ATI the OpenCL runtime is limited to 128MB size (without tuning environment parameters). if you're on NV it is nearly the whole available memory of your card.

but another very important thing: to optimize the speed it is required to convert each word of a wordlist to a structure the gpu kernel can work with more easily. the size of this structure is 20 bytes.

so it is hard to say what the maximum size of a wordlist is. it is better to know the number of words in the wordlist. if you have 1m words in your wordlist, this requires about 20mb gpu memory.
#6
next version v0.20 will have wordlist segmentation. this allows wordlists to be of arbitrary size. limited memory will no longer be an issue.
#7
Great news !
#8
done in v0.20, thread closed