Please help: 2 of the 4 290X are way too slow when performing dictionary attack
#11
gaming is nothing like compute, you cannot compare the two.

no, it is not constantly streaming like video games. if it was, you wouldn't get any gpu acceleration. use stdin mode on a fast hash and you will see what i mean. the host-device transfers add a shitload of overhead, regardless of bus speed. i'd suggest brushing up on your OpenCL.

USB speeds are irrelevant. the USB risers do not actually use USB. they simply make use of the size and flexibility of a USB cable.

no, it's not false. for these purposes, there's not a whole lot of difference in running off a 1x slot than a 16x slot. you can run off of a 1x slot just fine.
#12
I remember some time ago there were "fake" 6970 cards. I don't mean the "unlocked" 6950. That was a 6850 or so with a faked bios so that it showed as a 6970 to the system. Don't know about the details
#13
sure, but why do you mention this?
#14
Just a though, those 290x maybe are not 290x Smile
#15
ahhh, i see what you're saying. well if that were the case, i would suspect that he wouldn't be getting relatively the same speed out of all four gpus with brute force.
#16
Right, I just had that idea when he said that he's using 2 different brands, 2 times ASUS and 2 times Sapphire so I thought maybe the Sapphire (or the ASUS) are fake ones.

I just had an idea. Why not switch the ASUS and the Sapphire, so that the ASUS are connected to the risers and the Sapphire on the board. If it's the riser card or the 1x, the speed should be roughly the same (I mean the drop of card #3 and #4)
#17
(07-23-2014, 09:24 AM)epixoip Wrote: gaming is nothing like compute, you cannot compare the two.

no, it is not constantly streaming like video games. if it was, you wouldn't get any gpu acceleration. use stdin mode on a fast hash and you will see what i mean. the host-device transfers add a shitload of overhead, regardless of bus speed. i'd suggest brushing up on your OpenCL.

USB speeds are irrelevant. the USB risers do not actually use USB. they simply make use of the size and flexibility of a USB cable.

no, it's not false. for these purposes, there's not a whole lot of difference in running off a 1x slot than a 16x slot. you can run off of a 1x slot just fine.

you took my entire post out of context -_-

I was not comparing them to begin with. I was making an example of a process that is limited by bandwidth. I stated i don't know if hashcat is similar or not but something to make sure you dont over look -_-

1)I was just making the point that usb 3.0 is faster then 1x
1a) i never said it was false i just pointed out that 1x and 16x are drastically different in speeds but i also said i dont know if that is relevant in this process.
2)i dont know how hashcat works intimately hence why i said i dont know and i posted what i at least know in hope of helping him. (trying to not let anything get over looked...you know proper trouble shooting)

For all i know the GPU may have to communicate with the CPU or the RAM or the other GPUs constantly but if it is a completely contained process then it is irrelevant, which i guess it is. shesh
#18
I have read all the answers so far.
I am glad this has triggered a lot of talk as I have been desperate to find a solution to the problem.
I will keep my answers short:

1) none of the card is fake. As already stated the ASUS are this model (http://www.asus.com/Graphics_Cards/R9290XDC2OC4GD5/) whereas the Sapphire this (http://www.sapphiretech.com/presentation...2060&lid=1)

2) all 4 cards perform identically when performing brute force attacks. Problem materializes when performing dictionary attacks

3) AMD Overdrive does only see ONE card (the #1 card). This is the primary card connected to the monitor. I dunno if this has anything to do with the problem though I doubt it as card #2 works at its fullest whether in brute force or dictionary attack. Furthermore, since Overdrive cant see the 3 cards it is unable to set the fan speed for cards #3 and #4 BUT is able to do so for cards #1 and #2! Same goes for overclocking (sets fine for cards #1 and #2 but is unable to do so for #3 and #4). To solve this I use Sapphire's Trixx utility which is able to see and set anything on all 4 cards. I have ordered HDMI dummy plugs (special ones for R9 290X cards) and waiting for delivery. I am curious whether this will solve the problem of Overdrive in seeing the rest of the 3 cards and whether this plays any significance to the dictionary attack problem.

4) The dictionary I am using is sorted by length (3-17 length).

5) Over the weekend I will try what atom has suggested: swap positions between the Asus and Sapphire cards. This will give us a lot of useful feedback.

6) I still am waiting for an attack to finish. As soon as it does I will make some tests and post all relevant info requested so that it is clear for everyone what my settings are.

Thanks all for jumping into this!

l8r....
#19
(07-24-2014, 08:42 AM)john2014 Wrote: 2) all 4 cards perform identically when performing brute force attacks. Problem materializes when performing dictionary attacks

OK so on brute force it runs 100% but on a dictionary attack it runs at 85%...brute force shouldn't need to transfer anything but on a dictionary attack is the whole dictionary in RAM or VRAM? if it is in RAM then it could be due to the 1x limit because the only thing different from brute force and dictionary is that a dictionary is needed. How is the dictionary supplied to the GPU? constant? Bulk? or is 100% of it loaded into it (VRAM)?
#20
SomeGuy,

unless you're using stdin mode, everything is transferred to the device upon execution. the x1 slot will only cause minor delays in host-device transfers. there are dozens of users who run gpus on 1x slots without issue.

don't take this personally, but attempting to help someone troubleshoot when you don't understand how the program works ultimately results the blind leading the blind. it's not incredibly constructive, especially when there are already devs and mods on the thread who do know how the program operates. all it does is convolutes the thread. it would be better if you spent some time hanging out and learning and familiarizing yourself with the program before attempting to assist others.

john2014:

do you have crossfire enabled? if so, that would explain why overdrive can only see one gpu.