Intensity change on one card affects both cards' hashrates dramatically

[3lw33xi]

Hi all,

I have 2 Gigabyte R9 280x (GV-R928XOC-3GD REV2 R9 280X) cards which should each be pulling ~ 700 kh/s but their combined hash rate shown by cgminer is ~602 kh/s.  For best results, both are currently set at Intensity 20 and they run fine at about 300 kh/s each .  If I change the intensity of GPU0 down to I19, the card with I20 races up towards 400 while the one at I19 drops toward 200 -- effectively keeping the combined hash rate at 600ish.  If I swap intensities, the same thing happens with the numbers reversed.  Any idea what is keeping the combined hash rate around 600?  How can I get them to each to run at 600+ kh/s each? 

Setup:
Motherboard -- ASRock 970 Extreme4
Processor -- AMD Sempron 145
Memory -- 4GB G.SKILL DDR3 SDRAM (2 x 2GB)
Power Supply -- Seasonic 1000w Platinum PSU   
GPUs -- 2 @ Gigabyte R9 280x (GV-R928XOC-3GD REV2 R9 280X)
OS -- Xubuntu -- uname -a says: Linux box1 3.5.0-17-generic #28-Ubuntu SMP Tue Oct 9 19:31:23 UTC 2012 x86_64 x86_64 x86_64 GNU/Linux


Screenscrapes: 

As for the number of HW errors in the scrapes below, note that this is after it's been running for 48+ hours -- don't think that's a high number, but chime in if you think it is.  Also, notice that GPU 0 is at about a third of max RPM while GPU 1 is always at 9999.  Not sure if that makes a difference.



Running with both GPUs at the same I20 Intensity level:
--------------------------------------------------------------------------------
 GPU 0:  72.0C 2755RPM | 334.4K/300.0Kh/s | A:681268 R:3472 HW:17 WU:275.7/m I:20
 GPU 1:  42.0C 9999RPM | 285.3K/301.0Kh/s | A:675573 R:2393 HW:23 WU:282.4/m I:20
--------------------------------------------------------------------------------

If I change GPU Intensity: GPU0 @ 19 and GPU1 @ 20 and let run for 5 minutes:

 GPU 0:  72.0C 2854RPM | 204.2K/299.6Kh/s | A:683744 R:3472 HW:17 WU:275.3/m I:19
 GPU 1:  43.0C 9999RPM | 404.1K/301.3Kh/s | A:678255 R:2393 HW:24 WU:281.9/m I:20


Switch Intensity so: GPU0 @ 20 and GPU1 @ 19 and let run for 5 minutes:

 GPU 0:  72.0C 2657RPM | 397.1K/299.8Kh/s | A:686014 R:3472 HW:17 WU:275.4/m I:20
 GPU 1:  42.0C 9999RPM | 201.3K/301.2Kh/s | A:679700 R:2393 HW:24 WU:281.7/m I:19


Command line:
./cgminer --scrypt -I 20  --shaders 2048 --thread-concurrency 21712 --auto-fan --no-submit-stale  -o stratum+tcp://xxxxxx:3333 -u xxx.1 -p x

I have tried numerous things from several online guides to overclock, over mem, etc, but everything I tried lowered the hash rate from what you see as the first screen scrape.  Something is clearly not right but I cannot figure it out.  Any help greatly appreciated.

Thanks!

3lw33xi

[paaco1981]

You seem like a pretty technically inclined guy so Im not sure this will help, and honestly doesnt sound like your problem anyway since your hashrates are so much lower than they should be, but on my card which is a 7870. To get the best hashrate possible I kind of had to find the sweet spot on my clocks. Which actually meant underclocking the core from 1050 to 970 and then overclocking the shader. At the stock 1050/1200 my card gets about 360. At 970/1250 I get about 396.

Again, this doesn't sound like the solution to your problem at all, which sounds strange to me being so far under what you would expect from a 280x. But I hope it's good info for you or someone else anyway that higher core clocks aren't always best..

EDIT:: Looking at temps alone it doesn't look like GPU1 is doing any work at all with those temps. That looks like idle temps. Maybe try swapping the cards in the slots and see if you get a reverse and GPU1 starts to show higher temps and GPU0 shows the idle like temps. That would suggest something wrong with the card. If there is no change and GPU1 still shows idle temps after the card swap I would think some sort of weird driver/configuration error may be the culprit. or something wrong with that second PCI Express slot.

[tadspoles]

I have read that the 280s are pretty much the same as 7970s.  I get above 740Khs on my 7970s with this:

./cgminer --scrypt --thread-concurrency 8192 --worksize 256 -g 2 -I 13 --shaders 2048 --auto-fan

[3lw33xi]

tadspoles ~ I used your settings and allowed it to settle - numbers were lower.  Increased intensity slowly and maxed out at pretty much the same numbers I posted.

paaco1981 ~ Yes, the temps are curious, but they are both on the motherboard and the higher temp one has fans feeding off the back of the other card.  I will swap them and test that, then add the risers to each which I have but had not installed.  I'll post numbers in a little bit.

Thanks for your help and suggestions!

[3lw33xi]

Oh, and both cards are getting accepted shares at about the same rate, so it's not like that low temp card isn't working.

[paaco1981]

Oh, and both cards are getting accepted shares at about the same rate, so it's not like that low temp card isn't working.

I noticed that also, which had me at a loss, either the temp is being incorrectly displayed for that card or that card isn't working at all and its splitting the data for one card and showing it for both. ( sounds improbable but it wouldn't be the strangest thing I've ever seen ) Easiest way I knew to test what was going on was swap the cards to see if you get a change.

I'm also not a huge linux guru so I cant get too technical to try to help with drivers and things. I messed around with Suse years ago but not enough to be a wiz with it ;p Either way I hope you get things figured out.

[3lw33xi]

I swapped card slots and the numbers are still the same:

 GPU 0:  69.0C 3320RPM | 309.3K/300.2Kh/s | A:1092 R:0 HW:0 WU: 289.1/m I:19
 GPU 1:  30.0C 9999RPM | 291.7K/297.8Kh/s | A: 862 R:0 HW:0 WU: 254.8/m I:19

Maybe it's a bad card slot or mobo issue.  Going to try just one card in each (of 3) slots and see what that provides.

--paaco1981 - your response just popped up while writing this - no real change, I'm going to try one card at a time in all 3 slots (sounds dirtier than it sadly is.......)

[tadspoles]

I had something odd like that happen.  I was running cgminer in screen and had to kill all instances of screen and only have one going.  I also had to delete all the *.bin files.  I dunno.  Good luck man.

[paaco1981]

I have no clue if this is useful info or not. Read this that I copy and pasted from another forum talking about plugging a dummy plug into the cards not directly hooked to a monitor. Maybe ry this guys steps and see if you get a change.

First make sure you have "export DISPLAY=:0"  (no quotes) at the top of your start up script or be sure to enter it every time before you mine.  This is a MUST if you have more than 1 card.  Try this first, if it doesn't do the trick do the following steps.

I agree I am running several rigs on Ubuntu with 2 or more cards and a dummy plug is not needed.  There are a couple of different options to get the additional cards to work.  The easiest if you have direct keyboard, mouse and monitor connected to the rig you go to the Admin Catalyst Control Panel, under Display Manager or Display Options( i cant remember which one and am not around it to check) it will list all of your cards.  The main card will show as active and "single monitor display" however the additional cards will show up but will not be active.  There will be a drop down menu to enable "multi-monitor display," select this option on the 2nd and each additional card.  Click "OK" it will tell you that you have to reboot.  After you restart the computer cgminer will happily mine on all cards without a dummy plus.  The second option is fairly easy too and great if you can only SSH like myself.

Pull up a shell and enter the following commands, it will back up your Xorg automatically so no worries their either.

sudo aticonfig –initial
sudo aticonfig –adapter=ALL –initial              (you may need to use amdconfig instead of aticonfig depending on your config)
sudo reboot
 
Regardless of which option you use be sure to add "export DISPLAY=:0" in your script or make it a habit to enter it before mining.



I hope this helped, if it didn't please paste the output of the following command here or pastebin.com the output and link it below.  I will be happy to help more if i can.

export DISPLAY=:0
fglrxinfo
cat /etc/X11/xorg.conf
cgminer -n


Also please past your config file and start up script.  Remove username and password if you are worried about people seeing it

[3lw33xi]

Ok,

Mixing and matching, all results after letting hash rates settle for a couple minutes, here goes my journey of discovery:

Yanked one card and running only one (in slot 3) gave these results:
GPU 0:  66.0C 2978RPM | 576.9K/582.3Kh/s | A:3516 R:0 HW:2 WU: 537.7/m I:19

So, lets call it G1 is 576 by itself..

Other card (G2) by itself, slot 3:
 GPU 0:  66.0C 2812RPM | 579.3K/579.9Kh/s | A:1174 R:0 HW:1 WU:509.1/m I:19

So they both can do ~575 kh/s each by themselves......

I plug the other card in, this time in slot 2 (to avoid slot 1) so both cards are in and in different slots than before, and results are the same:

 GPU 0:  67.0C 2784RPM | 303.6K/295.0Kh/s | A:772 R:0 HW:0 WU: 286.6/m I:19
 GPU 1:  29.0C 9999RPM | 288.1K/295.0Kh/s | A:794 R:0 HW:1 WU: 286.6/m I:19


So, what, I need a Mobo/memory/chip for each card?  That is ridiculous, something isn't right.

Any ideas?

[paaco1981]

Trying to do a little reading on if you need dummy plugs with linux and Im getting mixed answers. It seems to depend how you set things up and what drivers and linux distro you use. Maybe look into how to make linux think that both cards are plugged into a monitor. Either it worked perfectly for people without them, or they had weird things happening. Again I'm no linux genius but Id do some google searching on Dummy plug when mining with multiple GPU in linux and see if you get anything helpful.

[tadspoles]

You're using the AMD driver downloaded from here right?

http://support.amd.com/en-us/download/desktop?os=Linux+x86

[3lw33xi]

Resolution!!!

Thank you for the suggestions which led me down the path to discovery.  Turns out the guides are incomplete (imagine that!)  Here are some findings:

1. Just because a rig will run and cgminer shows both cards as being used, it may be a lie (like cake).

2. If you have two cards (possibly more, someone else will have to verify this) which seem to balance the expected max hashrate for one, and their hash rates flip flop when Intensity is adjusted, it probably means only one card is being used, but the stats are for both.  This is a tip-off that something is astray.

3. If one card has high temp like 70C while the other is low like 38C, that's another tip-off that one card is not being used.

4. If you plug a monitor into both cards and then fire up cgminer, both cards get recognized and used, my current numbers with the settings as above without any tweaking are now 544 kh/s for GP0 and 570 kh/s for GPU1.  Point here though, is once cgminer is cranking along, you can remove the monitor cables from the rig and both cards still get used as they should.  I will get dummy plugs, but for now the cards only need to be plugged in to start cgminer, so it's not really a big deal.

I hope this helps someone else down the road and thanks again for the help from the GoldCoin community!

3lw33xi / gaz 

[paaco1981]

Glad you got it worked out. I was on the right track after all

[3lw33xi]

Yeah, you nailed it - I just couldn't see it until the numbers hit me in the face.  Thanks!