Posts: 4
Threads: 1
Joined: May 2013
Hi guys, I want to get together a group of PCs with several GPUs each, I've a couple of questions.
1) What's the real world difference between cracking with or without Inifiniband? Which percentage of cracking power do you lose if you use gigabit ethernet cables vs using Infiniband? Is it dependant on the hash cracked?
2) Does the VCL/oclHashcat-plus combo supports computers with different GPUs models or do they need to be the exact same chip?
3) Is the PCI-e port speed important? Some motherboards have the GPUs connected via a cable, that goes to the 1x PCI-e port. For bitcoin mining, that's allright. Would I lose any speed when hash cracking?
Thanks guys!
Posts: 179
Threads: 13
Joined: Dec 2012
1) It's all about latency. Infiniband's latency is 1/100 of gigabit ethernet.
2) Yes, you can use different types of AMD GPUs if they are supported by the latest catalyst.
3) As far as I know you wouldn't.
Posts: 4
Threads: 1
Joined: May 2013
Thanks for the replies, Kuci.
Regarding question number 1, in a real world scenario, would this drop my hash cracking rate to 10%? 50%? 90%?
Anyone had any experiences testing this setup with and without Infiniband?
Posts: 179
Threads: 13
Joined: Dec 2012
Well, let's think about a little bit. Could there occur packet loss on the LAN ? I don't think so...(but not 100% sure). So, if we think about the same, no it wouldn't. But if you're talking about speed, yes, it would be affected.
To be honest, I've no experience with Infiniband, but atom and epixoip may have.
Posts: 4
Threads: 1
Joined: May 2013
OK, I'll wait for them to expand on the subject, if they might.
Thanks!
Posts: 2,936
Threads: 12
Joined: May 2012
Kuci pretty much said it all.
It's difficult to calculate what percentage of speed loss you will see because it depends on a number of factors (number of nodes, number of GPUs, network topology, attack mode, number of hashes, etc), but the biggest problems you will experience with ethernet are very slow starts and erratic performance.
It may take several minutes for hashcat to actually start cracking, and then when it does, your speeds will be all over the place. You might have one or two GPUs cracking at full speed, a few more cracking at half speed, and then you'll probably have several that are barely doing any work at all.
But used Infiniband hardware is no more expensive than ethernet hardware. You can find SDR switches for just a couple hundred bucks, and HCAs for under $50. If you can afford to buy lots of GPUs for a cluster, you can surely afford to build an appropriate network for the cluster. Even if you have to sacrifice buying one or two GPUs to build a solid network, it's well worth it.
Posts: 4
Threads: 1
Joined: May 2013
Thanks for the reply!
So... now the totally off the wall question.
Is there any way to implement hashcat with VCL (or any other system) that will allow distributed hash cracking OVER THE INTERNET?
I want to create a centrally managed system that can distribuite workload among different computers all over the world, which will report to it and get assigned new jobs.
I don't see why it couldn't be done.
Posts: 179
Threads: 13
Joined: Dec 2012
As we have already wrote, it's about about the latency and because of the internet, it's about speed as well. You just can't compare speed of your ethernet and speed of your internet connection. Even if they are both 100 MBit/s.
Posts: 2,936
Threads: 12
Joined: May 2012
third-party wrappers like disthc are currently the only way to distribute hashcat over the internet
