10-02-2018, 07:39 AM (This post was last modified: 10-02-2018, 07:41 AM by lapolis.)
Hello everyone! Recently I had the opportunity to put together a 6x 1070gtx mixed brands.
When I use hashcat piping one of the c++ scripts found on here ( https://hashcat.net/forum/thread-6170.html ) the GPU usage is not 100% all the time. I'll explain better with pics.
SW:
-Ubuntu server 16
-nVidia driver 384.90
-hashcat 4.2.1
If I run:
./ua | hashcat64.bin -m2500 -w4 lollo.hccapx
I've notice that the GPUs go from 100% to 0% and vice versa all the time as the pics shows. I guess that is the CPU not able to create fast enough the wordlist?
Any idea on how I can improve the performance of my rig?
Also if I run:
hashcat64.bin --stdout wordlist.txt | hashcat64.bin -m2500 -w4 lollo.hccapx
I get something way worse. But if I run it in straight mode it will keep a nice 100% on all the GPUs at all the time.
If you need more info just ask me, I will reply as soon as possible!!
Thanks in advance!
10-03-2018, 03:03 AM (This post was last modified: 10-03-2018, 03:08 AM by slyexe.)
My biggest concern is why you're only getting 50% hashrate for WPA2 2500. My 1070 easily gets 350K which is double your rate per card. Otherwise, judging by the images your hashrate is still being produced so its most likely just an issue gathering the cards usage info and just drawing it as 0%
On a side note, your RAM should be upgraded as it could be a potential issue. You're Motherboards RAM should match that of the total VRAM of your Video cards. So if you're running 6 1070's with 8GB VRAM you should have 48GB RAM on the MOBO.
(10-03-2018, 03:03 AM)slyexe Wrote: My biggest concern is why you're only getting 50% hashrate for WPA2 2500. My 1070 easily gets 350K which is double your rate per card. Otherwise, judging by the images your hashrate is still being produced so its most likely just an issue gathering the cards usage info and just drawing it as 0%
when I took the screens I think I did not oc them, however I can get 290K - 310k still to improve anyway, I've got mixed 1070s, have to tweak them one by one, I'll get to 350k sooner or later
Also what make me think they are actually at 0% is also the watts from the wall getting lower when 3 gpus are at 100% and rising when 3 of them are.
(10-03-2018, 03:03 AM)slyexe Wrote: On a side note, your RAM should be upgraded as it could be a potential issue. You're Motherboards RAM should match that of the total VRAM of your Video cards. So if you're running 6 1070's with 8GB VRAM you should have 48GB RAM on the MOBO.
I did not see this coming, from htop I can see the ram usage holding at 7.3GB out of 16GB, but it actually make sense what you say. Unfortunately I guess it have to wait.
Your PCIe bandwidth simply cannot keep up with the amount of words being sent to the GPUs. Unless you are running into some hashcat limitations (eg because you're using a very small wordlist with lots of rules) you should avoid piping if possible, especially on more powerful rigs. Straight wordlist attacks will almost always be CPU/bandwidth limited unless you are attacking a very slow hash.
(10-03-2018, 11:53 AM)undeath Wrote: Your PCIe bandwidth simply cannot keep up with the amount of words being sent to the GPUs. Unless you are running into some hashcat limitations (eg because you're using a very small wordlist with lots of rules) you should avoid piping if possible, especially on more powerful rigs. Straight wordlist attacks will almost always be CPU/bandwidth limited unless you are attacking a very slow hash.
ok yeah now it makes lot of sense. Thanks.
Playing a bit further I realized that using a straight wordlist I can get 5gpus at 100% and just one randomly going down at about 80%. At this point I guess I need to use masks even if they will generate a "dirty" list or create my dictionary.