WARNING: This website is obsolete! Please follow this link to get to the new Albert@Home website!
App for Intel HD graphics? |
Message boards :
Wish List :
App for Intel HD graphics?
Message board moderation
Previous · 1 · 2 · 3 · 4 · Next
Author | Message |
---|---|
Bikeman (Heinz-Bernd Eggenstein) Volunteer moderator Project administrator Project developer Send message Joined: 28 Aug 06 Posts: 1483 Credit: 1,864,017 RAC: 0 |
Thanks for testing!
The BOINC server side code is not yet updated to have the INTEL GPU code in it, and therefore there is no app version installed on the server for Intel GPUs, so yes, those files are still required on the client side. Cheers HB |
Holmis Send message Joined: 4 Jan 05 Posts: 104 Credit: 2,104,736 RAC: 0 |
I did a little more investigation. Drivers installed via Windows Update almost never have the extra parts required for OpenCL, CUDA or what not. It's always recommended to get the drivers from Intel, Nvidia or AMD directly. No problem with the GPU nr. being 0. Both my 660Ti and the HD 4000 is nr. 0 and it still works. |
Claggy Send message Joined: 29 Dec 06 Posts: 78 Credit: 4,040,969 RAC: 0 |
The messages now show Don't worry about different devices being Device 0, they will show up as CAL device 0, 1, 2, etc, or Cuda device 0, 1, 2, etc, or as OpenCL device 0, 1, 2 etc for a particular vendors platform, ie, AMD APP, Nvidia CUDA or Intel, What you shouldn't have is two devices for the AMD APP platform with the same device number, What does CLinfo show? is that right, does Boinc run OpenCL apps on the right AMD GPU, or only on device 0? I'd be very tempted to go down a few driver versions checking if the devices are displayed properly (fully cleaning out the drivers with the AMD Catalyst Un-install Utility each time) and recheck operation. Claggy |
Alex Send message Joined: 1 Mar 05 Posts: 88 Credit: 398,734 RAC: 0 |
ATM I have some ati-wu's in queue, will wait until they are finished. Running on device 0 and device 1. Having gpu-z open twice, both gpu's show a usage of 75% or more. So it's possibly more a ui-problem than a device handling problem. since clinfo is quite long I post a link. https://dl.dropboxusercontent.com/u/50246791/clinfo1.txt I'm using beta-drivers for amd (13.6 beta; 13.5 beta 2 was last clean install). BTW: when googling for intel driver url's I found this: (sorry, german) http://www.pcgameshardware.de/CPU-Hardware-154106/News/Open-CL-Intel-Treiber-AMD-Prozessor-1068152/ Maybe it's nonsense, but maybe it helps making some apps a little bit faster. |
Claggy Send message Joined: 29 Dec 06 Posts: 78 Credit: 4,040,969 RAC: 0 |
since clinfo is quite long I post a link. There is text missing from the beginning, if you do from cmd prompt: clinfo >Output.txt you will get all the text outputed into the Output.txt file. Claggy |
Alex Send message Joined: 1 Mar 05 Posts: 88 Credit: 398,734 RAC: 0 |
https://dl.dropboxusercontent.com/u/50246791/clinfo2.txt Is complete now. Windows Power Shell did it. Alex |
Claggy Send message Joined: 29 Dec 06 Posts: 78 Credit: 4,040,969 RAC: 0 |
I've reported the Boinc 7.1.17 AMD device reporting problem to the Boinc Alpha list. Claggy |
Alex Send message Joined: 1 Mar 05 Posts: 88 Credit: 398,734 RAC: 0 |
I rearranged my systems a bit because the gpu's are needed somwhere else, will get them back during the weekend. Is a non-amd system for the moment. I followed the instructions given earlier and had the app running as 'Lokal'. Finished in a little bit more than 27 min. GPU usage ~98%, temp @ 38° (oversized cooler). Intel i3 3220 @ 3.3GHz Intel HD2500 BM 7.1.17 win7-64 GPU core clock 1050 MHz, GPU mem clock 666.7 MHz Mem usage dedicated 8MB, mem usage dynamic ~400MB GPU power reported as 2.9 W (GPU-Z 0.7.2) https://dl.dropboxusercontent.com/u/50246791/first%20albert-intel-opencl.PNG one can follow the progress here: http://albert.phys.uwm.edu/results.php?hostid=6571 Will it help to keep that running? Requesting new wu's needs babysitting, stays dry for long periods. |
Alex Send message Joined: 1 Mar 05 Posts: 88 Credit: 398,734 RAC: 0 |
I've reported the Boinc 7.1.17 AMD device reporting problem to the Boinc Alpha list. THX; I'm in contact with Charlie Fenton now. One can follow here: http://boinc.berkeley.edu/dev/forum_thread.php?id=8140&postid=49733#49733 |
Alex Send message Joined: 1 Mar 05 Posts: 88 Credit: 398,734 RAC: 0 |
Problem solved. |
zombie67 [MM] Send message Joined: 10 Oct 06 Posts: 130 Credit: 30,924,459 RAC: 0 |
Here are results from my i7-3770S, no OC of the CPU or GPU. It has no additional discrete GPU. This is with BOINC showing 1CPU+1GPU. With 1 additional thread reserved, 708 seconds, GPU load 96% flat: https://albert.phys.uwm.edu/result.php?resultid=988045 With no additional thread reserved, 729 seconds, GPU load was a slightly wobbly ~93%: https://albert.phys.uwm.edu/result.php?resultid=988042 It looks like reserving a thread helps a small bit. And that is odd, because according to task monitor, the task is using only about 5% of a thread (against a full thread reservation by BOINC). So that should mean it has 95% of a thread unused already, so why would manually reserving an additional thread improve anything? Can't wait for the servers to be upgraded to have an intel check box! FWIW, I have a couple boxes with both a nvidia discrete GPU + an intel HD 4000, crunching on both, with no monitors connected to either. Dublin, California Team: SETI.USA |
zombie67 [MM] Send message Joined: 10 Oct 06 Posts: 130 Credit: 30,924,459 RAC: 0 |
Update: I think that one task at 709 seconds was an anomaly. All the rest have been 720-730 seconds, regardless of reserving an additional thread. Dublin, California Team: SETI.USA |
Bernd Machenschalk Volunteer moderator Project administrator Project developer Send message Joined: 15 Oct 04 Posts: 1956 Credit: 6,218,130 RAC: 0 |
|
Alex Send message Joined: 1 Mar 05 Posts: 88 Credit: 398,734 RAC: 0 |
We released a BRP4 app version for Intel HD GPUs over at Einstein. THX, running a nice mix now. https://dl.dropboxusercontent.com/u/50246791/nice%20mix%201.PNG Alex |
Alex Send message Joined: 1 Mar 05 Posts: 88 Credit: 398,734 RAC: 0 |
@ BM: The settings needs rework; there is no tab to enable Intel-gpu like it is for AMD and nVidia and the GPU utilization factor of BRP apps has no effect. GPU-Z says gpu usage between 86 and 98%, Mem usage dedicated 6MB, Mem usage dynamic 375MB. There is headroom left ... Cheers Alex |
Holmis Send message Joined: 4 Jan 05 Posts: 104 Credit: 2,104,736 RAC: 0 |
That's great news! As there's been no announcement of the release on the Einstein boards I'll give some feedback here instead. I've run some tasks on my HD4000 over at Einstein and so far all looks good, 3 have validated. Link to the host The only small problem is that the task are sent with instructions to use 0.5CPU+1GPU, this has the effect on my system that the GPU load as per GPU-Z is less than 20%. If I free one core by setting my prefs to use 87.5% of the CPUs then the GPU load goes up to about 85-90%. For now I'm running with an app_config.xml and using 0.5CPU+0.5GPU to run x2 thus reserving one core when there is task available, GPU load is 90-95%. |
zombie67 [MM] Send message Joined: 10 Oct 06 Posts: 130 Credit: 30,924,459 RAC: 0 |
@ BM: +1 Not sure why, but it is clear that updating the server to know intel GPU, does not also add the preferences selector. This happened at both SETI Beta and Collatz. So then the admin has to make a second change. Annoying for everybody. Edit: I forgot to say thanks..... THANKS! Edit 2: Hmm. Something seems wrong. Can't get work (on einstein): 79164 Einstein@Home 7/9/2013 3:49:47 PM update requested by user 79165 Einstein@Home 7/9/2013 3:49:51 PM [sched_op] Starting scheduler request 79166 Einstein@Home 7/9/2013 3:49:51 PM Sending scheduler request: Requested by user. 79167 Einstein@Home 7/9/2013 3:49:51 PM Requesting new tasks for intel_gpu 79168 Einstein@Home 7/9/2013 3:49:51 PM [sched_op] CPU work request: 0.00 seconds; 0.00 devices 79169 Einstein@Home 7/9/2013 3:49:51 PM [sched_op] intel_gpu work request: 108000.00 seconds; 1.00 devices 79170 Einstein@Home 7/9/2013 3:49:53 PM Scheduler request completed: got 0 new tasks 79171 Einstein@Home 7/9/2013 3:49:53 PM [sched_op] Server version 611 79172 Einstein@Home 7/9/2013 3:49:53 PM No work sent 79173 Einstein@Home 7/9/2013 3:49:53 PM Project has no jobs available 79174 Einstein@Home 7/9/2013 3:49:53 PM Project requested delay of 60 seconds 79175 Einstein@Home 7/9/2013 3:49:53 PM [sched_op] Deferring communication for 00:01:00 79176 Einstein@Home 7/9/2013 3:49:53 PM [sched_op] Reason: requested by project I have intel GPU selected (and nothing else), and I have all 4 apps selected. According to the server status, all the queues are full of tasks ready to send. Dublin, California Team: SETI.USA |
Bernd Machenschalk Volunteer moderator Project administrator Project developer Send message Joined: 15 Oct 04 Posts: 1956 Credit: 6,218,130 RAC: 0 |
@ BM: Setting to disable INTEL GPU was added last night. I honestly don't expect anyone to stress these poor little Intel GPUs with more than one task at a time, so the "GPU utilization factor" was disabled there on purpose. BM |
Bernd Machenschalk Volunteer moderator Project administrator Project developer Send message Joined: 15 Oct 04 Posts: 1956 Credit: 6,218,130 RAC: 0 |
Edit 2: Hmm. Something seems wrong. Can't get work (on einstein): What does the scheduler log say? Can you give me a hostid that I can look up? BM |
Bernd Machenschalk Volunteer moderator Project administrator Project developer Send message Joined: 15 Oct 04 Posts: 1956 Credit: 6,218,130 RAC: 0 |
In general, sorry for the somewhat rushed issuing of the Intel GPUs on Einstein. We needed a bit more computing power to help the CPUs crunching through the new Arecibo data, so we decided to just have a try with the Intel GPUs, as the damage, if any, would be very limited. BM |