hashcat Forum

Commandline and driver look good.

Try the following:

1. rm /etc/X11/xorg.conf*
2. aticonfig --initial -f --adapter=all
3. reboot

Bingo. I'm definitely saving this thread for future reference. I'm still not quite up to where your documentation says I should be, but this is around what I was getting in Windows:

Speed........: 3899.0k c/s Real, 4220.5k c/s GPU

Thanks a million for development and quick support responses, atom. You're the best. Let me know if you have any additional ideas to get me up to 5000k c/s, but like I said - this is what I got in Windows, so I'm happy.

add -n 32 --gpu-loops 1000

Speed........: 4799.4k c/s Real, 4934.1k c/s GPU

That actually did help quite a bit. Is there any documentation on exactly what those commands do and how to tweak them? I've looked, but I can't find them. Obviously the wiki (and the tool command line) give a basic description, but I was hoping for something a little more detailed.

well md5crypt is 1000 times iterated. if you choose 1000 gpu-loops you choose an ideal situation for cracking md5crypt. that might be different on different algorithms.

the -n is just an amplifier of workload. the higher you set it, the faster it gets (till you hit a wall at some point) but also your desktop becomes slower. thats why i added this, for example if you play FPS games or watch 1080p movies you want to set this to -n 8 or so you can both at once on the same gpu.

Dearest atom,

I am writing to ask you about your own personal rig. More importantly, I would like some suggestions on some racks that are appropriate for the cracking department. For instance, enough PCI slots as well as room for the cards that go in them. Power is another concern, of course.

I've been looking at a four socket Tyan board that I would install Magny-Cours chips in. I'm not crazy about the latest architecture from AMD. I am a RISC man but when I am forced into x86-64 (AMD64) I generally prefer to go with AMD - although I confess the vast majority of my x86 systems are indeed Intel for various reasons.

So, a four socket Tyan board with a few PCI slots are where I am at, however I would like to be able to scale up to eight graphics cards as needed so that is looking like a dead end. I remember something about a small company that offers like 4U racks with room for eight graphics cards for the design industry. Is that really our only option? I'd prefer to avoid a tower setup as that really doesn't suit my needs.

Lastly, do slower PCI slots cause any bottlenecks? What is the lowest one can go before hitting a bottleneck with the 7970s? Would something like the 7990s require x16 given the dual cards on one channel? I'm shamefully unfamiliar with these type things. It really isn't my department.

Also, I still haven't been able to find any information regarding the usefulness of FirePro cards in cracking. Do they offer anything at all of benefit? Do they not of certain features built in to accommodate processing rigs with multiple cards?

Really looking forward to when you integrate system pooling. Be looking for a donation from my direction.

Sincerest thanks,

Host

(09-11-2012, 05:25 AM)Host Wrote: [ -> ]I am writing to ask you about your own personal rig. More importantly, I would like some suggestions on some racks that are appropriate for the cracking department. For instance, enough PCI slots as well as room for the cards that go in them. Power is another concern, of course.

Things change fast. My current suggestion is go away from expensive boards with many pcie slots. Better buy cheap board with 2 slots. If possible, try to get one which have enough space between them so that you can attach cards with custom coolers that require 3 slots space and then combine the systems using VCL.

https://hashcat.net/wiki/doku.php?id=vcl_cluster_howto

(09-11-2012, 05:25 AM)Host Wrote: [ -> ]I've been looking at a four socket Tyan board that I would install Magny-Cours chips in. I'm not crazy about the latest architecture from AMD. I am a RISC man but when I am forced into x86-64 (AMD64) I generally prefer to go with AMD - although I confess the vast majority of my x86 systems are indeed Intel for various reasons.

For cracking Intel with AVX is currently better suited than AMD with XOP. However, since we are not doing a lot cracking on CPU its not so important.

(09-11-2012, 05:25 AM)Host Wrote: [ -> ]So, a four socket Tyan board with a few PCI slots are where I am at, however I would like to be able to scale up to eight graphics cards as needed so that is looking like a dead end. I remember something about a small company that offers like 4U racks with room for eight graphics cards for the design industry. Is that really our only option? I'd prefer to avoid a tower setup as that really doesn't suit my needs.

You talk about renderstream? AFAIK, yes, it is your only option.

(09-11-2012, 05:25 AM)Host Wrote: [ -> ]Lastly, do slower PCI slots cause any bottlenecks? What is the lowest one can go before hitting a bottleneck with the 7970s? Would something like the 7990s require x16 given the dual cards on one channel? I'm shamefully unfamiliar with these type things. It really isn't my department.

That has never been tested to the max, but from the theory x1 is enough.

(09-11-2012, 05:25 AM)Host Wrote: [ -> ]Also, I still haven't been able to find any information regarding the usefulness of FirePro cards in cracking. Do they offer anything at all of benefit? Do they not of certain features built in to accommodate processing rigs with multiple cards?

I doubt it. oclHashcat-* has integrated control for multiple GPUS. There is no need for a hardware overlay to do the management for a software that does not know it is running on multiple GPUS, that is not made for it. These cards (like TESLA) are made for fast floating point calculation. We totally do not need floating point calculation in crypto. Thats why I think you better go hd7970's.

(09-11-2012, 05:25 AM)Host Wrote: [ -> ]Really looking forward to when you integrate system pooling. Be looking for a donation from my direction.

That is nice of you, but I do not accept donations. Thanks!

(09-11-2012, 05:25 AM)Host Wrote: [ -> ]Lastly, do slower PCI slots cause any bottlenecks? What is the lowest one can go before hitting a bottleneck with the 7970s? Would something like the 7990s require x16 given the dual cards on one channel? I'm shamefully unfamiliar with these type things. It really isn't my department.

I have been using an HD5750 on x1 without any problems. Okay, it's not really a top end card but I don't think that it would be a problem. I strongly recommend that you read the following:

https://blog.zorinaq.com/?e=42

Thank you very much for your replies.

HD6990 runs perfectly on x1 with low end cpu. As a card it still faster than 7970 so at the moment we have plenty of room in pciex1