Intel gpu not seen by BOINC

Message boards : Number crunching : Intel gpu not seen by BOINC
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19012
Credit: 40,757,560
RAC: 67
United Kingdom
Message 1627800 - Posted: 15 Jan 2015, 4:57:42 UTC

New MB and CPU on sons computer, it has intel gpu, is seen in device manager (win7 pro) which can and has driven a monitor, but gets these lines in BOINC Manager

15/01/2015 04:44:45 | | OpenCL CPU: Intel(R) Pentium(R) CPU G3220 @ 3.00GHz (OpenCL driver vendor: Intel(R) Corporation, driver version 4.2.0.99, device version OpenCL 1.2 (Build 99))
15/01/2015 04:44:45 | SETI@home | Found app_info.xml; using anonymous platform
15/01/2015 04:44:45 | | app version refers to missing GPU type intel_gpu
15/01/2015 04:44:45 | | app version refers to missing GPU type intel_gpu
15/01/2015 04:44:45 | | app version refers to missing GPU type intel_gpu
15/01/2015 04:44:45 | | app version refers to missing GPU type intel_gpu
15/01/2015 04:44:45 | | app version refers to missing GPU type intel_gpu
15/01/2015 04:44:45 | | app version refers to missing GPU type intel_gpu
15/01/2015 04:44:45 | | app version refers to missing GPU type intel_gpu
15/01/2015 04:44:45 | | app version refers to missing GPU type intel_gpu


So what have "I" done wrong, it's always my fault everbody seems to say so.
ID: 1627800 · Report as offensive
Profile arkayn
Volunteer tester
Avatar

Send message
Joined: 14 May 99
Posts: 4438
Credit: 55,006,323
RAC: 0
United States
Message 1627838 - Posted: 15 Jan 2015, 7:09:31 UTC

Probably either missing the OpenCL drivers for the video or have an unsupported version for BOINC

ID: 1627838 · Report as offensive
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19012
Credit: 40,757,560
RAC: 67
United Kingdom
Message 1627845 - Posted: 15 Jan 2015, 7:50:34 UTC - in response to Message 1627838.  

Probably either missing the OpenCL drivers for the video or have an unsupported version for BOINC

Thanks,
Latest version of BOINC, so will look to see if OpenCL drivers have been loaded and get them if not.
ID: 1627845 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1627860 - Posted: 15 Jan 2015, 8:56:52 UTC - in response to Message 1627845.  
Last modified: 15 Jan 2015, 9:02:20 UTC

Probably either missing the OpenCL drivers for the video or have an unsupported version for BOINC

Thanks,
Latest version of BOINC, so will look to see if OpenCL drivers have been loaded and get them if not.

Be careful with Intel drivers - many of the current ones don't work properley. Hold on while I find a link.

Here we go: http://downloadmirror.intel.com/23885/a08/win64_153322.zip

That should get you the 10.18.10.3621 driver for 64bit Windows, which is generally regarded as the most reliable. There's a lot of useful information in http://einstein.phys.uwm.edu/forum_thread.php?id=10214, where I see someone is about to guinea-pig a newly-released 4080 driver.
ID: 1627860 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1627951 - Posted: 15 Jan 2015, 12:56:04 UTC - in response to Message 1627838.  

Probably either missing the OpenCL drivers for the video or have an unsupported version for BOINC

Or it needs a Monitor connected to it's output, or the desktop need extending onto it.

Claggy
ID: 1627951 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1627972 - Posted: 15 Jan 2015, 14:33:42 UTC

Hmmmm, OpenCL CPU: Intel(R) Pentium(R) CPU G3220.
If you search for CPU G3220 you will not find a single other hit for that CPU.
Has anyone tried that CPU with SETI tasks before? All I've ever seen were i5 & i7 CPUs running the iGPU tasks.
ID: 1627972 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1627983 - Posted: 15 Jan 2015, 15:08:00 UTC - in response to Message 1627972.  
Last modified: 15 Jan 2015, 15:22:32 UTC

Hmmmm, OpenCL CPU: Intel(R) Pentium(R) CPU G3220.
If you search for CPU G3220 you will not find a single other hit for that CPU.
Has anyone tried that CPU with SETI tasks before? All I've ever seen were i5 & i7 CPUs running the iGPU tasks.

The G3220 is listed as OpenCL 1.2 capable on the Intel HD Graphics driver download page:

https://downloadcenter.intel.com/Detail_Desc.aspx?DwnldID=24596&lang=eng

BUT: The driver version on that page (10.18.14.4080) has not yet been tested for compatibility with BOINC projects' OpenCL applications, like SETI and Einstein - it was only released three days ago.

Proceed at caution, with a small cache and checking frequently for acceptable validation rates - and please report back. Best of luck.

Edit: Do not use the 10.18.10.3960 driver shown as 'previously released' on the page retrieved by an Intel driver search for G3220 - that driver has definitely been reported as producing invalid results with current BOINC project science applications.
ID: 1627983 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1628062 - Posted: 15 Jan 2015, 18:33:46 UTC - in response to Message 1627983.  

Hmmmm, OpenCL CPU: Intel(R) Pentium(R) CPU G3220.
If you search for CPU G3220 you will not find a single other hit for that CPU.
Has anyone tried that CPU with SETI tasks before? All I've ever seen were i5 & i7 CPUs running the iGPU tasks.

The G3220 is listed as OpenCL 1.2 capable on the Intel HD Graphics driver download page:

https://downloadcenter.intel.com/Detail_Desc.aspx?DwnldID=24596&lang=eng

BUT: The driver version on that page (10.18.14.4080) has not yet been tested for compatibility with BOINC projects' OpenCL applications, like SETI and Einstein - it was only released three days ago.

Proceed at caution, with a small cache and checking frequently for acceptable validation rates - and please report back. Best of luck.

Edit: Do not use the 10.18.10.3960 driver shown as 'previously released' on the page retrieved by an Intel driver search for G3220 - that driver has definitely been reported as producing invalid results with current BOINC project science applications.

It looks like the Pentium G3xxx was added in the 3907 release. I have the x64 package I can toss up on my server if anyone needs it.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1628062 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1628197 - Posted: 15 Jan 2015, 23:59:54 UTC - in response to Message 1628062.  

Report from Einstein user 'boinc127' that driver 4080 fails like the other recent ones - though he has the more traditional 4th. gen i5 with HD 4600.
ID: 1628197 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1628202 - Posted: 16 Jan 2015, 0:03:03 UTC - in response to Message 1628197.  

Report from Einstein user 'boinc127' that driver 4080 fails like the other recent ones - though he has the more traditional 4th. gen i5 with HD 4600.

Does anyone know if these problems have been reported to Microsoft? Raistmer or Einstein?

Claggy
ID: 1628202 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1628254 - Posted: 16 Jan 2015, 3:51:57 UTC
Last modified: 16 Jan 2015, 4:05:18 UTC

I happen to stumble upon release 15.33.30.64.3958. Which looks to be the last combined release for Ivy Bridge, Haswell, & Bay Trail in one package. With the current Haswell/Broadwell releases 15.36.14.64.4080 & I'm not sure if the 33 vs 36 designation is only a reference to hardware supported or if there are other differences. So the 3958 release may not work any better than the 3960 release.

Also the supported hardware listed on the drivers is not complete as compared to what is listed in the .inf for the driver. In release 15.33.30.64.3958 the .inf includes
; HSW Classic
iHSWGT1D   	= "Intel(R) HD Graphics"
iHSWGT1M    	= "Intel(R) HD Graphics"
iHSWGT15D    	= "Intel(R) HD Graphics 4400"
iHSWGT2D    	= "Intel(R) HD Graphics 4600"
iHSWGT2M    	= "Intel(R) HD Graphics 4600"


EDIT:
In picking the drivers apart release 15.33.30.64.3958 contains Intel_OpenCL_ICD64.dll v1.2.11.0 & 15.36.7.64.3960 contains Intel_OpenCL_ICD64.dll v2.0.0.0. The 2.0.0.0 version is also in 15.36.14.64.4080. This seems to be the primary OpenCL dll loaded for the GPU. So I am going to guess until Intel updates it we will continue to have problems with newer driver releases.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1628254 · Report as offensive
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19012
Credit: 40,757,560
RAC: 67
United Kingdom
Message 1628303 - Posted: 16 Jan 2015, 7:55:03 UTC

Thanks for all the help, I haven't tried anything yet as computer is back with son, but will probably see if it is capable sometime over the w/end, if I can get him to stop WOW for a short period. More recycled DDR3 Ram, at least 8 GByte possibly 16, will be on it's way next week hopefully.
ID: 1628303 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1628382 - Posted: 16 Jan 2015, 12:13:56 UTC - in response to Message 1628254.  

In release 15.33.30.64.3958 the .inf includes ...

Comparative information from the working 10.18.10.3621 driver:

DiskId = "Intel(R) Graphics Media Accelerator Driver for Windows"
Intel = "Intel Corporation"
iIVBGD0 = "Intel(R) HD Graphics 4000"
iIVBGM0 = "Intel(R) HD Graphics 4000"
iIVBGD0GT1 = "Intel(R) HD Graphics"
iIVBGM0GT1 = "Intel(R) HD Graphics"
iIVBGD0SRV = "Intel(R) HD Graphics"
iVLVGMT0 = "Intel(R) HD Graphics"
; HSW Classic
iHSWGT1D = "Intel(R) HD Graphics"
iHSWGT1M = "Intel(R) HD Graphics"
iHSWGT15D = "Intel(R) HD Graphics 4400"
iHSWGT2D = "Intel(R) HD Graphics 4600"
iHSWGT2M = "Intel(R) HD Graphics 4600"
; HSW ULT
iHSWGT1UT = "Intel(R) HD Graphics"
iHSWGT2UT = "Intel(R) HD Graphics Family"
iHSWGT3UT = "Intel(R) HD Graphics 5000"
iHSWGT3UT28W = "Intel(R) Iris(TM) Graphics 5100"
iHSWGT2UX = "Intel(R) HD Graphics Family"
iHSWGT1ULX = "Intel(R) HD Graphics"
; HSW CRW
iHSWGT3CW = "Intel(R) Iris(TM) Pro Graphics 5200"
iHSWGT3CWDT = "Intel(R) Iris(TM) Pro Graphics 5200"
; HSW SRVR
iHSWSVGT2 = "Intel(R) HD Graphics P4600/P4700"

and for the DLL



So, it may not be right to blame any of Intel, Microsoft, or indeed NVidia for recent OpenCL problems. It may even be that OpenCL.dll (original name) v2.0.0.0 requires that different programming techniques (or an updated SDK) be used by our esteemed developers.
ID: 1628382 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1628403 - Posted: 16 Jan 2015, 13:12:41 UTC - in response to Message 1628382.  

1) Nope, OpenCL dll is just a common stub to call correct vendor function nothing more.
2) I will try to report issues to Intel but need concise summary of issue with clear test case. Weakly similar will not go.
ID: 1628403 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1628411 - Posted: 16 Jan 2015, 13:52:40 UTC - in response to Message 1628403.  

1) Nope, OpenCL dll is just a common stub to call correct vendor function nothing more.
2) I will try to report issues to Intel but need concise summary of issue with clear test case. Weakly similar will not go.

Unfortunately, 'weakly similar' (SETI) or 'high - but below 100% - invalid rate' (Einstein) is all we've got to go on. One approach for SETI would be to use our standard bench tools, generating reference results using the good driver, and comparing with test results under the newer driver versions. If that threw up, for example, something like the AP result for pre_Fermi (the whole class of 'single pulses' going AWOL), we might get a clearer handle on the nature of the breakage, and be able to supply a set of reproducible test apps - much less complicated than the full SETI/Einstein applications - wheich would be more likely to get Intel's attention.

Likewise, I'd like to see collaboration with the Einstein developers, who I feel should also be paying more attention to what tends to be regarded as a 'minor partner' platform in the GPU stakes. We got close with http://einstein.phys.uwm.edu/forum_thread.php?id=10214&nowrap=true#136427 just before the holidays, but since then the staff have been incredibly busy with emergency server rebuilds: Bernd and Oliver deserve a weekend off, but I'd be prepared to re-open negotiations around Tuesday next week if nothing else has blown up in the meantime.

And finally, if OpenCL.dll is just a stub, can you suggest what other libraries and components we should be scrutinising for differences? I suppose, for completeness, we should open our minds to both possible compiler changes, and runtime library support changes. Which is a very big haystack to search for needles.
ID: 1628411 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1628425 - Posted: 16 Jan 2015, 14:32:03 UTC - in response to Message 1628411.  

Are any users with Ivy Bridge GPUs reporting issues with the current release 15.33.32.4061 or is it only confined to Haswell with release 15.36.7.64.3960/15.36.14.64.4080 so far? I have a feeling this may only be an issue with the drivers that include OpenCL 2.0 support.

I have 15.33.32.4061, The latest for Ivy Bridge & Ba Trail GPUs, on my Bay Trail GPU system, but only put it on last night. The GPU take ~22 hours for a task. So it will be some time before I will know if there are any issues & need to go back to an older driver, but I have a feeling it will be OK.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1628425 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1628434 - Posted: 16 Jan 2015, 14:49:00 UTC - in response to Message 1628425.  
Last modified: 16 Jan 2015, 15:14:15 UTC

Are any users with Ivy Bridge GPUs reporting issues with the current release 15.33.32.4061 or is it only confined to Haswell with release 15.36.7.64.3960/15.36.14.64.4080 so far? I have a feeling this may only be an issue with the drivers that include OpenCL 2.0 support.

I have 15.33.32.4061, The latest for Ivy Bridge & Ba Trail GPUs, on my Bay Trail GPU system, but only put it on last night. The GPU take ~22 hours for a task. So it will be some time before I will know if there are any issues & need to go back to an older driver, but I have a feeling it will be OK.

I suspect that most of these issues arise from people running BOINC with whatever OEM drivers happen to have been current when their machine's factory software image was assembled. There isn't the same gaming pressure on consumers to upgrade drivers every month, needed or not, with Intel GPUs compared to the competition. Waiting for these users to 'report' details of their setups and any related problems is likely to be even more frustrating that is usual for BOINC-land: most reporting is done by wingmates more observant and knowledgeable about issues like validation.

I was lucky: my Dell-hosted Haswell (~September 2013) arrived with OEM drivers new enough to support OpenCL 1.2, and hence I bypassed the initial OpenCL 1.1 problems with that machine. Since then, it's always appeared to me that the 3rd. generation Core i5/i7 drivers (Ivy Bridge) have coalesced with the 4th. generation (Haswell) in a single combination package - though I've never explored whether the deployed contents have been identical for both installation paths. But I do have an Ivy Bridge i7 running as well, and I can test-run your suggested package on that.

Edit - I've downloaded 4061, in the form of win64_153332.zip from https://downloadcenter.intel.com/Detail_Desc.aspx?DwnldID=24593&lang=eng&ProdId=3711 - but that page, like all the others I've seen, claims joint compatibility with both Ivy Bridge and Haswell. I'm not convinced there's any differentiation.

I'll flush down my Einstein cache a bit, and give it a try either later this afternoon or tomorrow.
ID: 1628434 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1628439 - Posted: 16 Jan 2015, 15:17:17 UTC - in response to Message 1628411.  


And finally, if OpenCL.dll is just a stub, can you suggest what other libraries and components we should be scrutinising for differences? I suppose, for completeness, we should open our minds to both possible compiler changes, and runtime library support changes. Which is a very big haystack to search for needles.

1) Look Intel driver in device manager - there is list of files it consists of.
Also, some md5 of system directory approach can be used to see what particular file changes between particular driver versions.

2) What about OpenCL MultiBeam? No errors under all drivers or what?

3) Case with invalid signals under particular driver version and good ones under another (but for same device, not CPU ref) will go too.
ID: 1628439 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1628465 - Posted: 16 Jan 2015, 16:02:40 UTC - in response to Message 1628434.  

I suspect that most of these issues arise from people running BOINC with whatever OEM drivers happen to have been current when their machine's factory software image was assembled. There isn't the same gaming pressure on consumers to upgrade drivers every month, needed or not, with Intel GPUs compared to the competition. Waiting for these users to 'report' details of their setups and any related problems is likely to be even more frustrating that is usual for BOINC-land: most reporting is done by wingmates more observant and knowledgeable about issues like validation.

I was lucky: my Dell-hosted Haswell (~September 2013) arrived with OEM drivers new enough to support OpenCL 1.2, and hence I bypassed the initial OpenCL 1.1 problems with that machine. Since then, it's always appeared to me that the 3rd. generation Core i5/i7 drivers (Ivy Bridge) have coalesced with the 4th. generation (Haswell) in a single combination package - though I've never explored whether the deployed contents have been identical for both installation paths. But I do have an Ivy Bridge i7 running as well, and I can test-run your suggested package on that.

Edit - I've downloaded 4061, in the form of win64_153332.zip from https://downloadcenter.intel.com/Detail_Desc.aspx?DwnldID=24593&lang=eng&ProdId=3711 - but that page, like all the others I've seen, claims joint compatibility with both Ivy Bridge and Haswell. I'm not convinced there's any differentiation.

I'll flush down my Einstein cache a bit, and give it a try either later this afternoon or tomorrow.

I looked at the inf for 3958 and saw it didn't include Haswell. Despite 4th gen core CPUs being listed in the "supported hardware" list. So I had assumed 4061 did not include Haswell either. However the inf for 4061 does in fact include Haswell.
It feels like they have two groups working on things & not talking to each other at this point.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1628465 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1630573 - Posted: 21 Jan 2015, 12:05:14 UTC

OK, time for a report-back. I've been slowly plodding my way through the standard Lunatics tests for MB tasks - full FG set, 15 tasks at a variety of angle ranges - on both HD 4000 (i7 'Ivy Bridge') and HD 4600 (i5 'Haswell').

Specific test runs have been:

HD 4000 with drivers 10.18.10.3621 and 10.18.10.4061
HD 4600 with drivers 10.18.10.3621 and 10.18.14.4080

- in both cases, running the 64-bit version under 64-bit Windows 7.

First, the good news: for SETI@Home, all results were good, with Q values at or above 99% when compared with both CPU and CUDA reference results. I believe HAL is performing similar tests for AP, but on my evidence, if you only run SETI there's no need to worry about driver versions.

The story is different for Einstein. The 3621 driver produces good results on both iGPUs: the 4061 driver is also satisfactory on the HD 4000 Ivy Bridge chip. But the HD 4600 / 4080 driver combination produces a lot of validation errors and invalid results. You should avoid the 4080 driver on Haswell chips if you run Einstein@Home, even occasionally as a backup project.

The difference appears to be in the .cl compiler phase. For all three combinations which work at both SETI and Einstein, the compiled file (e.g. MultiBeam_Kernels_r2489.cl_IntelRHDGraphics4000.bin_V7_1018103621) is approximately 680 KB - sizes vary slightly. But the fourth combination produces a much smaller compiled file - MultiBeam_Kernels_r2489.cl_IntelRHDGraphics4600.bin_V7_1018144080 is only 230 KB.

Similarly, the complete set of 16 MB_clFFTplan files is over 4 MB in the three 'working' cases, but less than 1 MB for the suspect HD 4600 / 4080 driver. I'll report that back to Einstein, who appear to pre-compile their kernels before distribution - that might be a clue.
ID: 1630573 · Report as offensive
1 · 2 · Next

Message boards : Number crunching : Intel gpu not seen by BOINC


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.