isn't there anyone else that wants to have the 7970 optimized by ck?
Perhaps people are wasted from donating to diablo for the same thing. |
|
|
|
|
Sleep tight, but unfortunately you've picked one of the bad commits. That code never worked as they were attempts at doing it that didn't work properly and evolved like this (first unbroken commit was 05970):
commit 83a836765432dbc88b972588fb678aff17979a35
Linux's ADL uses a busnumber in descending order for devices so enumerate them in the opposite order to windows.
commit 059701272c75bc30236f1c4a3aaf8db5e59bf718
Carry virtual gpu number across.
commit 3bc0583454b09b5c4dac18450c4766572d69644f
Iterate and change virtual device order instead of shuffling ram.
commit 371e5f688a2b6e5446e243ac9edcc490c0243a79
Reorder displayed devices to map to physical locations and initialise according to logical location instead.
commit 5a0b4f62d0966d08bd92142d92d8b97f8e68f1e7
Map GPU devices to virtual devices in their true physical order based on BusNumber.
|
|
|
|
http://ck.kolivas.org/apps/cgminer/temp/Since this may be a recurring theme, I've uploaded a new bleeding edge binary for both windows and linux into that directory which can work around the problem but will give a warning when it happens. I'll try and keep bleeding edge binaries in there if/when required. Starting with --verbose with these binaries should give you a lot more information which will help you debug where devices are going as well if you're having a new ordering issue.
|
|
|
|
|
Try cleaning out your tree with a fresh ./autogen.sh && ./configure && make clean first perhaps.
|
|
|
|
Any system changes needed with Linux to move from 2.1.1 to 2.2.0? I'm observing massive problems with the latest updates. I kept the previous working binary as reference, but was not able to isolate the problem. First obvious thing is ADL fails to initialize: [2012-01-29 19:30:32] ADL Initialisation Error!
The remaining initialization sequence looks ok, at least no obvious differences to what 2.1.1 was reporting: [2012-01-29 19:30:35] Loaded binary image phatk110817Caymanbitalignv2w128long4.bin ...but the hash rates achieved are at about 30% [2012-01-29 20:18:56] GPU2 (5s):95.6 (avg):121.2 Mh/s | A:43 R:0 HW:0 U:1.44/m I: 9
of what 2.1.1 does: [2012-01-29 20:22:51] GPU2 (5s):372.4 (avg):350.6 Mh/s | A:4 R:0 HW:0 U:4.35/m | I: 9
As said, using the same APP and ADL version, same kernel. I also tried with deleting previous kernel binaries with the same result. I am using the current git HEAD for 2.2.0 (82af288e69). Any ideas? Most unusual No system changes should be needed. ADL failing to initialise could be anything, but seems more common with multiple instances of cgminer running. Can you do a git pull and start with --verbose -D and see what it says with the extra debugging I just committed? What does the startup debugging give on the first device if you delete the binaries? 30% sounds like something horribly wrong. In fact it looks like all 3 "GPUs" may just be one device. EDIT: You did do "export DISPLAY=:0" didn't you? |
|
|
|
when using anything above intensity 9 on my 5970 w/ 11.12 drivers. The cpu usage is pegged... 9 and below cpu usage is 0%
Any reason why its higher on I 10+? thanks
just curious what mhash your getting with your 5970 using 11.12 ? I messed around with several different configurations over the last couple of days for mine and the best I came up with is 11.9 w/ cgminer 2.20 - I get a nice 820 mhash - woohoo! btw: ckvolias do you know why Anubis web front end stopped working with the 2.2.0 upgrade? thanks! Changed API with the newer commands I would guess. |
|
|
|
|
Calm down fellas. They're different CPU bugs. There are 3 in all. Either way intensity above 9 is mostly harmful.
|
|
|
|
Do I need to enable engine clock autotune for "temp-cutoff" activation? Or it's active by default?
You need to enable --auto-gpu .However if you dont give it a range for engine clock speeds, it won't adjust any clock speeds, but it will still cut off the device if it hits the --temp-cutoff threshold. |
|
|
|
Latest git tends to die a miserable death, Con.
The previous build i reported testing some 12 hours ago works just fine.
Has been fixed. Dodgy cleanups. |
|
|
|
But why 2.1.2 don't have such issue then?
It stopped after it found the first device. 2.2.0 finds all the devices and then reorders them correctly according to their PCI bus position. As I said you were lucky it wasn't giving you temps and fanspeed for the 4250 instead because it happened to put the 4250 second in the driver order even though you physically have it first in the slot - You can see it has a higher adapterindex in that list despite having a lower busnumber. As for a solution to this problem that is correct as opposed to "lucky"... I'm thinking, but I have no ideas apart from explicitly telling cgminer at this stage. This won't be the correct fix, but this exe should work around the problem for you for now (put into 2.2.0 directory): http://ck.kolivas.org/apps/cgminer/temp/cgminer.exe |
|
|
|
Question: Can your other device (4250) mine?
I guess you don't need to answer that, I checked. So this is the problem. One device that can mine and one that can't. There's already nothing to correlate openCL devices with devices that appear in the ATI display library apart from "how many there are" and "in what order they appear". Put a non-opencl device in there (like the 4250) and it's a mess... GAH. Nothing short of some kind of massive whitelist will fix this. If it worked previously you were lucky it wasn't monitoring the 1st card instead. EDIT: This recurring problem of having no way of correlating opencl and ATI devices is getting seriously long in the tooth. Can we get a re-post of those facepalms please? But why 2.1.2 don't have such issue then? It stopped after it found the first device. 2.2.0 finds all the devices and then reorders them correctly according to their PCI bus position. As I said you were lucky it wasn't giving you temps and fanspeed for the 4250 instead because it happened to put the 4250 second in the driver order even though you physically have it first in the slot - You can see it has a higher adapterindex in that list despite having a lower busnumber. As for a solution to this problem that is correct as opposed to "lucky"... I'm thinking, but I have no ideas apart from explicitly telling cgminer at this stage. |
|
|
|
Question: Can your other device (4250) mine?
I guess you don't need to answer that, I checked. So this is the problem. One device that can mine and one that can't. There's already nothing to correlate openCL devices with devices that appear in the ATI display library apart from "how many there are" and "in what order they appear". Put a non-opencl device in there (like the 4250) and it's a mess... GAH. Nothing short of some kind of massive whitelist will fix this. If it worked previously you were lucky it wasn't monitoring the 1st card instead. EDIT: This recurring problem of having no way of correlating opencl and ATI devices is getting seriously long in the tooth. Can we get a re-post of those facepalms please? |
|
|
|
|
Question: Can your other device (4250) mine?
|
|
|
|
2 gpus. 5770 and internal 4250 (set as primary). And no problems with this setup and v2.1.2. Driver installed in XP is 11.12 and 2.5 sdk, nothing else.  Okay so what's happening here is that the ATI display library is finding both GPUs but opencl is only finding one GPU... Now look at this post: I just had an annoying multi-hour debugging session, trying to figure out why cgminer thinks there's only one GPU in my (Debian Testing) system. $ aticonfig --list-ad * 0. 06:00.0 ATI Radeon HD 5800 Series 1. 07:00.0 ATI Radeon HD 5800 Series 2. 08:00.0 ATI Radeon HD 5450 Turns out that AMD is being stupid. Again. OpenCL 1.1 AMD-APP-SDK-v2.4 (595.10) (don't remember where I downloaded it from): library works. OpenCL 1.2 AMD-APP (831.4) (available from http://developer.amd.com/sdks/AMDAPPSDK/Pages/default.aspx ) does NOT see more than one GPU. You can extract the build number etc. from the library with $ strings -a /pth/to/the/library.so | grep "^OpenCL.*AMD-APP" Yes I realise that's a linux report but it could be related. Next, download this bleeding edge build http://ck.kolivas.org/apps/cgminer/temp/cgminer.exeand try this command: cgminer.exe --ndevs |
|
|
|
Well that's interesting, you're getting the "adl found more devices than opencl" problem. How many GPUs do you have? Just one? Did you change anything in your driver setup at all? What output do you get if you use this http://ck.kolivas.org/apps/cgminer/temp/cgdebug.exe in the cgminer directory the same way you would use cgminer.exe and start with -T ? It will give more information about the devices with lines that look like this: [2012-01-22 17:31:58] lpAdapterID 20016752 iBusNumber 1 iDeviceNumber 0 iFunctionNumber 0 iVendorID 4098 strAdapterName AMD Radeon HD 6900 Series strDisplayName :0.0 lpInfo.strUDID 256:26393:4098:12583:5762
|
|
|
|
Err, hardware monitoring in 2.2.0 isn't working even with 2.5 sdk. Just checked with XP.
Huh wot?  Do you get any useful error messages when you start it with -D -T --verbose (as well as your usual parameters) ? |
|
|
|
No idea then  |
|
|
|
Plus I can't use any sdk higher than 2.1, because of cpu bug.
By the way, "cpu bug" has nothing to do with SDK, only the ATI driver does. Others saying that cpu bug is in SDK, not in drivers. Plus, I've 11.12 installed (tested 11.6 as well) and I don't have this bug with 2.1 sdk, but with just switching to 2.3/2.4/2.5/2.6 the bug "suddenly" appear on W7 x64. Can you fix 2.2.0 version so I can use hardware monitoring with 2.1 sdk? It's working in 2.1.2, so why not in 2.2.0? Btw can I change for example phatk110817.cl on phatk 2.2 kernel from Phoenix, by just renaming it's kernel.cl to phatk110817.cl and replace the default one? It will work that way? Mystery. That does not make any sense about the SDK using CPU. I suspect you are actually having conflict issues with the files on your installation rather than anything to do with the SDK itself. I hear that putting the .dlls into the cgminer directory on windows will allow you to run it with whatever sdk you want. No I cannot fix it to work with 2.1 sdk because I suspect the extra features in cgminer 2.2.0 are breaking with that older sdk. No you cannot use a .cl from any other project in cgminer. The Application Programming Interface is different on all of them. I suggest you treat any "reported" speeds from phoenix and guiminer with caution as they're allegedly not reliable... |
|
|
|
Plus I can't use any sdk higher than 2.1, because of cpu bug.
By the way, "cpu bug" has nothing to do with SDK, only the ATI driver does. |
|
|
|
|
Show Posts
