kano
Legendary
Offline
Activity: 4592
Merit: 1851
Linux since 1997 RedHat 4
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 04:04:24 AM |
|
... Try it before you laugh.
The point he has already stated is that if you are using more than one GPU in your computer - you are wasting the others. The stability issue is the display GPU. Set the other GPU's to higher intensity. You are reducing the performance of the other GPU's by using the same settings on all of them that you have determined are needed to keep the display GPU stable. Kano, You are not hearing me. Here are some visuals: I: d,9,9 E: 1130 M: 1000 http://www.fileswap.com/dl/YWZrbwsLyU/d-9-9-1130-1000.png.htmlI: d,11,11 E: 1130 M: 1000 http://www.fileswap.com/dl/d6Akac5gdI/d-11-11-1130-1000.png.htmlI: 7 E: 1130 M: 1000 http://www.fileswap.com/dl/URJ9zQ5EDs/7-7-7-1130-1000.png.htmlI'm just reporting what your software does on Windows with three 7970s not what it should do. Best RATE and stability is achieved with intensity 7 for ALL GPUs. Maybe it is a bug, but that is what is happening. Try it for yourself. BTW, your "gpu" command is working great. I get status of each gpu by parsing your reply json string. I wrote a little watchdog to monitor and restart cgminer when it crashes or status goes to sick or dead. Heh ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif) I only wrote the API (and a few patches here and there) But yeah I didn't see any comments about using multiple intensities - that's why I said that. Have you tried anything like 7,X,X for any of X=8, 9, 10 or 11? (since d has other effects also) That's very strange that the other cards doing nothing in the system except mining would require a setting of 7 Is there anything unusual about your setup? (obviously not this: but what I mean would be like having 1 display on each card or some other unusual setup) Edit: I'll also add that a few 100 share test is pretty small ... and could be unreliable. Give it an hour at least to be sure of the value.
|
|
|
|
kano
Legendary
Offline
Activity: 4592
Merit: 1851
Linux since 1997 RedHat 4
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 04:50:20 AM |
|
Well I have actually discussed recently what the intensity actually does. It simply divides up the work based on: a higher intensity means longer work for the GPU to do non-stop. ... now where was that post ... https://bitcointalk.org/index.php?topic=28402.msg771425#msg771425So each step up in intensity is making the GPU process twice as long between very short 'rests' (though they are only cgminer GPU rests while the CPU readies the next GPU request, not OS rests)
|
|
|
|
The00Dustin
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 10:39:33 AM |
|
Isn't the 7970 a dual-gpu card? Maybe having the display on the first card means both of those GPUs affect Windows because of this? IOW, instead of maybe or with 1/3rd the 9's if you only have 2 cards.
|
|
|
|
kano
Legendary
Offline
Activity: 4592
Merit: 1851
Linux since 1997 RedHat 4
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 01:49:05 PM |
|
Isn't the 7970 a dual-gpu card? Maybe having the display on the first card means both of those GPUs affect Windows because of this? IOW, instead of maybe or with 1/3rd the 9's if you only have 2 cards. Actually - that's the smartest comment yet on the subject ... Because it could be that GPU0 isn't the display ... i.e. you'd have to make sure which GPU's are the display and set it lower. Then I'm almost certain the rest could be higher ... Edit: though I should point out ... ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif) The 7970 is a single GPU ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif)
|
|
|
|
kano
Legendary
Offline
Activity: 4592
Merit: 1851
Linux since 1997 RedHat 4
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 02:04:17 PM |
|
... and on a different subject ... ZTEX
I wasted an hour or so on it so far and worked out what appears to be the commands (and changed all the cgminer code that needs changing and created a template ztex.c)
So if anyone has one it would be good if they could drop by #cgminer on FreeNode and I'll ask you to run a few commands so I can work out what the commands really are.
I asked the ZTEX people and got 2 useless replies in one email: 1) Read the java code, the mining commands for the bitstreams aren't documented (yeah I had already done that that's why I asked to confirm what I had worked out) 2) "You cannot do this within a day." lol - well I've only spend a couple of hours on it so far ... (and then decided after than comment ... stuff it I'll do it tomorrow ... or whenever ...)
|
|
|
|
The00Dustin
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 02:11:03 PM |
|
Actually - that's the smartest comment yet on the subject ... Edit: though I should point out ... ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif) The 7970 is a single GPU ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif) Being smart without actually being smart is awesome... I forgot about the gpu ordering issue, so basically each of these needs tried if there's not an easier way to work out which is the display (since board order and driver order don't match and even differ between OS): -I 9,7,9,9 -I 9,9,7,9 -I 9,9,9,7 Good thing it isn't dual-GPU, that would make it a real PITA. EDIT: Even assuming there's no good way to determine which is which short of such testing, wouldn't display performance have been noticeably worse in the previously attempted if it wasn't making the display GPU dynamic?
|
|
|
|
kano
Legendary
Offline
Activity: 4592
Merit: 1851
Linux since 1997 RedHat 4
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 02:21:01 PM |
|
... EDIT: Even assuming there's no good way to determine which is which short of such testing, wouldn't display performance have been noticeably worse in the previously attempted if it wasn't making the display GPU dynamic? Actually he said: When I enable dynamic intensity, other gpus get most of the action, and the harder they go, the slower the dynamic (monitor gpu) gets. Which sounds like it could be the wrong GPU is 'd' i.e. as you increase the intensity on the display GPU it should of course get less responsive
|
|
|
|
DeathAndTaxes
Donator
Legendary
Offline
Activity: 1218
Merit: 1079
Gerald Davis
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 02:22:48 PM Last edit: March 06, 2012, 03:55:16 PM by DeathAndTaxes |
|
The #0 GPU is rarely the one driving the display. It is up to windows crazy device number scheme. Only only Windows rig (3x watercooled 5970s) the "display GPU" was #4. When one 5970s died and I RMAed it (4 GPUs installed) he "display GPU" become #1. When I got the RMA back (6 GPUs) the "display GPU" after a reinstall become #2. So it varied and it was never #0 in this limited sample set.
So windows users should identify the display GPU. If you have multiple physical displays as long as they are all connected to a single card one GPU will handle them all even on dual GPU cards like 5970s. If you have multiple displays connected to multiple cards well change that it is sub optimal.
To identify the display GPU: 1) stop mining 2) open GPU-Z sensor tab (be sure record data when not shown is checked), AB, or CCC. 3) All GPU should be ~0% GPU load. 4) Do something which requires the GPU acceleration flash accelerated video, (although you should turn it back off when mining), Afterburner test tool, play some 3D game (in windowed mode would be best). 5) Look which GPU load increases. Record this GPU number.
Now you think you might be done but you aren't (because windows & ADL library used by cgminer order cards differently sometimes). Leave whatever tool you are using open. 5) Start cgminer (you can use low intensity and stock clocks for stability) 6) You should see all GPU go to 99% load 7) Disable each GPU one at a time until you find the one which causes load on the GPU identified in #4 (in GPU-Z / AB) to go to zero. Wait a few seconds between each GPU disable as sometimes it take a couple seconds for the tools to show a drop in GPU load.
That is the GPU in cgminer that is driving the display. Yes windows is asinine. You likely want to write it down.
On edit: clarified a few steps.
|
|
|
|
The00Dustin
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 02:31:51 PM |
|
Actually he said: When I enable dynamic intensity, other gpus get most of the action, and the harder they go, the slower the dynamic (monitor gpu) gets. Which sounds like it could be the wrong GPU is 'd' i.e. as you increase the intensity on the display GPU it should of course get less responsive I took that to mean that the MH/s gets slower on the GPU with 'd' but you could be right. Regardless, the advice from DAT after your last post (and just before this post) is exactly what Windows miners need. (and there's still no thumbs up emoticon)
|
|
|
|
nelisky
Legendary
Offline
Activity: 1540
Merit: 1002
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 03:13:05 PM |
|
... and on a different subject ... ZTEX
I wasted an hour or so on it so far and worked out what appears to be the commands (and changed all the cgminer code that needs changing and created a template ztex.c)
So if anyone has one it would be good if they could drop by #cgminer on FreeNode and I'll ask you to run a few commands so I can work out what the commands really are.
I asked the ZTEX people and got 2 useless replies in one email: 1) Read the java code, the mining commands for the bitstreams aren't documented (yeah I had already done that that's why I asked to confirm what I had worked out) 2) "You cannot do this within a day." lol - well I've only spend a couple of hours on it so far ... (and then decided after than comment ... stuff it I'll do it tomorrow ... or whenever ...)
You may want to read my code instead, as python is much more readable than java and you have everything I needed on mpbm in one single source. https://github.com/nelisky/Modular-Python-Bitcoin-Miner/blob/ztex/worker/fpgamining/ztexdev.pyI will try to find you on irc once I get some work things out of the way.
|
|
|
|
DeathAndTaxes
Donator
Legendary
Offline
Activity: 1218
Merit: 1079
Gerald Davis
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 03:51:44 PM |
|
The way I identified display was to start miner with 1120/1000 for gpu0, 1121/1000 for gpu1, 1122/1000 for gpu2 and then opened AMD Control Center to look at what display adapter it was using in the "Performance Center". The core 1120 set by cgminer was displayed in the Performance Center so I figured that must have been the display gpu. There is only one shown, the other ones are displayed as "disabled" on Information/Hardware screen. (Control Panel shows all enabled, of course)
That is weird. Makes me think something larger is going on with your system. AMD drivers can be weird sometimes but I don't think any GPU should show "disabled" while under load.
|
|
|
|
twmz
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 06:08:10 PM |
|
Has anyone written a script that checks if any GPUs are DEAD/OFF via the API? I'd like to create something that checks periodically and restarts cgminer if that happens (since a restart usually fixes the issue). If someone already has a simple script to talk to cgminer, I'd love to use it as a starting point.
|
Was I helpful? 1 TwmzX1wBxNF2qtAJRhdKmi2WyLZ5VHRs WoT, GPGBitrated user: ewal.
|
|
|
fred0
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 06:52:48 PM Last edit: March 06, 2012, 07:04:44 PM by fred0 |
|
Has anyone written a script that checks if any GPUs are DEAD/OFF via the API? I'd like to create something that checks periodically and restarts cgminer if that happens (since a restart usually fixes the issue). If someone already has a simple script to talk to cgminer, I'd love to use it as a starting point.
https://bitcointalk.org/index.php?topic=66779.0An example would be line 60, however all it does is pop up a non modal window. You can modify it to email you with echo "$node gpu$i reports $STATUS" | mail -s "Device Failure" emailaddress@whatever.comor whatever else you want.
|
|
|
|
twmz
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 08:26:16 PM |
|
Has anyone written a script that checks if any GPUs are DEAD/OFF via the API? I'd like to create something that checks periodically and restarts cgminer if that happens (since a restart usually fixes the issue). If someone already has a simple script to talk to cgminer, I'd love to use it as a starting point.
https://bitcointalk.org/index.php?topic=66779.0An example would be line 60, however all it does is pop up a non modal window. You can modify it to email you with echo "$node gpu$i reports $STATUS" | mail -s "Device Failure" emailaddress@whatever.comor whatever else you want. Outstanding. Thank you.
|
Was I helpful? 1 TwmzX1wBxNF2qtAJRhdKmi2WyLZ5VHRs WoT, GPGBitrated user: ewal.
|
|
|
fred0
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 08:36:22 PM Last edit: March 06, 2012, 09:36:09 PM by fred0 |
|
Has anyone written a script that checks if any GPUs are DEAD/OFF via the API? I'd like to create something that checks periodically and restarts cgminer if that happens (since a restart usually fixes the issue). If someone already has a simple script to talk to cgminer, I'd love to use it as a starting point.
https://bitcointalk.org/index.php?topic=66779.0An example would be line 60, however all it does is pop up a non modal window. You can modify it to email you with echo "$node gpu$i reports $STATUS" | mail -s "Device Failure" emailaddress@whatever.comor whatever else you want. Outstanding. Thank you. Forgive me, I just updated the scripts, the one to modify is the cgm-dev-parse. The new corresponding number is line 45. I think you can just modify that to you liking. I'm sorry I don't already have a device restart function.
|
|
|
|
kano
Legendary
Offline
Activity: 4592
Merit: 1851
Linux since 1997 RedHat 4
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 06, 2012, 09:24:50 PM |
|
OK, as I just said to you in IRC - yep I give in, I failed - I'm not gonna do it ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif) As you said, ZTEX requires USB direct control - it won't work via serial - so yeah there's no way I could write that in a day without the hardware ![Tongue](https://bitcointalk.org/Smileys/default/tongue.gif) (If I had the hardware ... well yeah that wouldn't be an issue) I've little experience with USB directly (and no idea why it wont work via serial like Bitforce and Icarus) but if that's the case, yep I certainly can't get the USB interface to it right without the hardware to try stuff Oh and what I was thinking about the Frequency was to do it a little different: Basically have a cgminer option to set the start frequency (comma list for ZTEX devices), and use a simple lookup table based on the hardware (for unspecified on the command line) with a low value for unknown, and step it up N (1 or 2?) every successful nonce then as soon as it gets a hardware error start stepping it down one every time you get a HW error (that would always be the case from then on, never go back up) It would take a while to get to optimal frequency if it started too low but since that will vary from board to board, you would be setting cgminer to start close to the optimal ferquency for each board once you see what it is after an extended run Then also add a new API option to set it to a new value (and report the current value) ... since you may want to be able to push it back up rather than restarting cgminer if conditions change (e.g. a hot day vs a cool night) Thinking about pushing it back up automatically - I don't think that's a good idea since I'd expect in the long term that would be bad for the life of the FPGA - i.e. better to let it settle a couple of Hz below than always going up and down near HW failure (and also losing work due to every time you get a HW failure)
|
|
|
|
twmz
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 07, 2012, 01:50:08 AM |
|
Forgive me, I just updated the scripts, the one to modify is the cgm-dev-parse. The new corresponding number is line 45. I think you can just modify that to you liking. I'm sorry I don't already have a device restart function.
Thanks for the help. The 'nc' command was the trick I needed and wasn't familiar with. I already have a script (init.sh) that kills all running miners and starts new ones in background screen sessions, so I just needed a way to detect that it was necessary. This is pretty simple script I ended up with (I run it every 15 minutes via cron): #!/bin/bash
totaldevs=$(cgminer --ndevs | grep '] GPU' | wc -l) alivedevs=$(echo "{\"command\":\"devs\"}" | nc localhost 4028 | sed 's/\x0$//' | python -mjson.tool | grep 'Alive' | wc -l)
if [ "$totaldevs" != "$alivedevs" ] then echo "At least one GPU is not alive. Restarting miner." /home/username/init.sh fi
|
Was I helpful? 1 TwmzX1wBxNF2qtAJRhdKmi2WyLZ5VHRs WoT, GPGBitrated user: ewal.
|
|
|
fred0
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 07, 2012, 03:36:31 AM Last edit: March 07, 2012, 12:56:46 PM by fred0 |
|
Forgive me, I just updated the scripts, the one to modify is the cgm-dev-parse. The new corresponding number is line 45. I think you can just modify that to you liking. I'm sorry I don't already have a device restart function.
Thanks for the help. The 'nc' command was the trick I needed and wasn't familiar with. I already have a script (init.sh) that kills all running miners and starts new ones in background screen sessions, so I just needed a way to detect that it was necessary. This is pretty simple script I ended up with (I run it every 15 minutes via cron): The first version I had out there was just bad...this is much cleaner. One of these days I might look into email, but I really don't think that I'll need it. Glad someone benefited from the script.
|
|
|
|
twmz
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 07, 2012, 04:49:11 AM |
|
The first version I had out there was just bad...this is much cleaner.
One of these days I might look into email, but I really don't think that I'll need it. Glad someone benefited from the script.
I didn't do emails or anything. The script just quietly fixes the problem by restarting everything. I may add a pause followed by a re-check of the statuses to which I might resort to a full scale automated reboot of the rig if the miner restart didn't fix the issue. I don't really want to just be notified that something is wrong and have to manually fix it. I want the script to fix the problem itself.
|
Was I helpful? 1 TwmzX1wBxNF2qtAJRhdKmi2WyLZ5VHRs WoT, GPGBitrated user: ewal.
|
|
|
cablepair
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
March 07, 2012, 11:59:33 AM |
|
can anyone compile the latest cgminer from git for windows for me?
thanks!
|
|
|
|
|