Posts by Richard Haselgrove

1) Message boards : Questions and problems : bug? computation error on restarting based on time of day (Message 91867)
Posted 8 days ago by Richard Haselgrove
Post:
It might be worth randomly suspending single tasks and allowing them to restart later in their own time.

GPU tasks are always removed from GPU memory when suspended for any reason (unlike CPU tasks, which can be left in memory if your preferences permit). That means that on restart, they have to be reloaded from the checkpoint file.

Over the years, some projects have from time to time had problems with their checkpointing code. You wouldn't notice any problems when suspending, but you can get a crash when the app tries to read back a bad file. The same thing can happen if your hard disk is flaky.

Start with the project that failed on restart, but test other projects too. Only rely on unattended stops/starts when you're certain that all your active projects have reliable checkpoint/restart code.
2) Message boards : GPUs : Request for Intel iGPU support to be a working feature again. (Message 91861)
Posted 8 days ago by Richard Haselgrove
Post:
As mentioned somewhere in this thread, I updated to version 26.20.100.6890 (Latest) Date: 5/29/2019 (DCH version) from https://downloadcenter.intel.com/download/28816/Intel-Graphics-Windows-10-DCH-Drivers?product=80939, and all my previous problems went away as well.

Raistmer is still looking through the test logs I sent him, but I think it might really have been the older drivers this time (he always blames the drivers, and I always blame his coding. We're fine: last time, it turned out to be a compiler optimisation flag, so we were both wrong).

My test machine is running at All tasks for computer 8670176: looking good so far. The single inconclusive seems to be a dodgy wingmate with several inconclusives and invalids.

Unfortunately, there's no sign of the promised v7.16 yet, but anyone who finds themselves stymied by the DCH extensions list can download the 'artifact' test build from https://ci.appveyor.com/project/BOINC/boinc/builds/25266698/artifacts
3) Message boards : GPUs : Request for Intel iGPU support to be a working feature again. (Message 91846)
Posted 10 days ago by Richard Haselgrove
Post:
I also refer to a SETI test in that thread: like then, today's test with the SETI app has led to an inconclusive result, so I suspect Intel have tweaked the precision or driver optimisation again. Bernd is aware of the Einstein problem, but too busy to do anything about it: I don't feel like waking Raistmer out of his slumbers before I set off on my travels. So I'm inclined to let this one lie until some cold, boring, winter's evening.
Change of plan. The two SETI tasks I ran with the original OEM driver threw up multiple issues, so much so that the first one needed four replications before SETI found a pair which could validate together.

That suggests that there may be something more seriously wrong with the current application mix at SETI, and Raistmer and I are going to run some more substantial offline tests.

In the meantime, that level of replication is obviously wasteful for the project and all concerned: I suggest that - whilst observed testing is very welcome - you don't put these newest iGPUs into full SETI production until we report back.
4) Message boards : Questions and problems : GPU detection failed, error 0xc0000005 (Message 91833)
Posted 11 days ago by Richard Haselgrove
Post:
6/6/2019 4:15:21 PM | | [coproc] clGetDeviceInfo failed to get device extensions for device 0
Update. We have now established that this error is intrinsic to BOINC, and will occur when the new Intel Windows DCH drivers are installed - whether they are sourced from Microsoft or Intel.

BOINC v7.14.x will always fail to detect an iGPU with these drivers, but v7.16.x should work. It is anticipated that testing of the new client should start either over this weekend, or during the week commencing 17 June 2019.
5) Message boards : GPUs : Request for Intel iGPU support to be a working feature again. (Message 91830)
Posted 12 days ago by Richard Haselgrove
Post:
Yes, that was my reaction at first too.

But luckily I had appropriate hardware available, and I plowed through the very limited information provided.

And yes, it was a BOINC problem. Not a deliberate removal of a feature (as the thread title seems to imply), but simply the outside world outgrowing us.

Anyway, my patch has been accepted. You will both test version 7.16 (expected to be released for testing this weekend), won't you?
6) Message boards : GPUs : Request for Intel iGPU support to be a working feature again. (Message 91827)
Posted 12 days ago by Richard Haselgrove
Post:
OK, now we're getting somewhere. I've updated to 26.20.100.6890 (Latest) Date: 5/29/2019, and Oblomov still sees OpenCL:

  Platform Name                                   Intel(R) OpenCL
Number of devices                                 2
  Device Name                                     Intel(R) UHD Graphics 620
  Device Vendor                                   Intel(R) Corporation
  Device Vendor ID                                0x8086
  Device Version                                  OpenCL 2.1 NEO
  Driver Version                                  26.20.100.6890
  Device OpenCL C Version                         OpenCL C 2.0
  Device Type                                     GPU
but BOINC doesn't.

13/06/2019 19:38:53 |  | Starting BOINC client version 7.14.2 for windows_x86_64
13/06/2019 19:38:53 |  | log flags: file_xfer, sched_ops, task, cpu_sched, sched_op_debug
13/06/2019 19:38:53 |  | Libraries: libcurl/7.47.1 OpenSSL/1.0.2g zlib/1.2.8
13/06/2019 19:38:53 |  | Data directory: C:\BOINCdata
13/06/2019 19:38:53 |  | Running under account Richard Haselgrove
13/06/2019 19:38:53 |  | OpenCL CPU: Intel(R) Core(TM) i5-8250U CPU @ 1.60GHz (OpenCL driver vendor: Intel(R) Corporation, driver version 7.6.0.0228, device version OpenCL 2.1 (Build 0))
13/06/2019 19:38:53 |  | No usable GPUs found
The GPU detection code returned

<warning>clGetDeviceInfo failed to get device extensions for device 0</warning>


Does that match your diagnostics?

Continuing: that test comes a long way down the list - https://github.com/BOINC/boinc/blob/master/client/gpu_opencl.cpp#L844 - it's already found everything else.

Oblomov found them:

  Device Extensions                               
cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_fp16 cl_khr_depth_images cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_icd cl_khr_image2d_from_buffer cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_intel_subgroups cl_intel_required_subgroup_size cl_intel_subgroups_short cl_khr_spir cl_intel_accelerator cl_intel_media_block_io cl_intel_driver_diagnostics cl_intel_device_side_avc_motion_estimation cl_khr_priority_hints cl_khr_throttle_hints cl_khr_create_command_queue cl_khr_fp64 cl_khr_subgroups cl_khr_il_program cl_intel_spirv_device_side_avc_motion_estimation cl_intel_spirv_media_block_io cl_intel_spirv_subgroups cl_khr_spirv_no_integer_wrap_decoration cl_khr_mipmap_image cl_khr_mipmap_image_writes cl_intel_planar_yuv cl_intel_packed_yuv cl_intel_motion_estimation cl_intel_advanced_motion_estimation cl_khr_gl_sharing cl_khr_gl_depth_images cl_khr_gl_event cl_khr_gl_msaa_sharing cl_intel_dx9_media_sharing cl_khr_dx9_media_sharing cl_khr_d3d10_sharing cl_khr_d3d11_sharing cl_intel_d3d11_nv12_media_sharing cl_intel_simultaneous_sharing
That's a long list - 1.09 KB (1,117 bytes). I think we might need a bigger buffer...

OK, make that https://github.com/BOINC/boinc/blob/master/lib/opencl_boinc.h#L48. But increasing it gives:

13/06/2019 20:31:42 |  | Starting BOINC client version 7.15.0 for windows_x86_64
13/06/2019 20:31:42 |  | This a development version of BOINC and may not function properly
13/06/2019 20:31:42 |  | log flags: file_xfer, sched_ops, task, cpu_sched, sched_op_debug
13/06/2019 20:31:42 |  | Libraries: libcurl/7.47.1 OpenSSL/1.0.2g zlib/1.2.8
13/06/2019 20:31:42 |  | Data directory: C:\BOINCdata
13/06/2019 20:31:42 |  | Running under account Richard Haselgrove
13/06/2019 20:31:42 |  | OpenCL: Intel GPU 0: Intel(R) UHD Graphics 620 (driver version 26.20.100.6890, device version OpenCL 2.1 NEO, 3166MB, 3166MB available, 211 GFLOPS peak)
13/06/2019 20:31:42 |  | OpenCL CPU: Intel(R) Core(TM) i5-8250U CPU @ 1.60GHz (OpenCL driver vendor: Intel(R) Corporation, driver version 7.6.0.0228, device version OpenCL 2.1 (Build 0))
I think that earns me a beer - PR to follow later.

Oh, all right. PR #3183. NOW beer. Seeya!
7) Message boards : GPUs : Request for Intel iGPU support to be a working feature again. (Message 91826)
Posted 12 days ago by Richard Haselgrove
Post:
Yes, you're right. "Using current driver" should work. The trouble is that my current driver does work, your current driver doesn't work. So, what's the difference, and is it something for BOINC to fix, Microsoft to fix, or Intel to fix?

I've posted my current driver version (25.20.100.6374), but you've said nothing except "since the shift to Windows DCH drivers". Any chance we could have comparable details?

Until we work out who broke what, we won't know whose job it is to fix it.

Edit - I assume you're somewhere on https://downloadcenter.intel.com/download/28816/Intel-Graphics-Windows-10-DCH-Drivers?product=80939, but that page includes 9 different versions.
8) Message boards : GPUs : Not getting any work for my Anonymous platform at Einstein@home (Message 91825)
Posted 12 days ago by Richard Haselgrove
Post:
Try doing a command line

boinc --help
to see what the start-up options are (I can't find them documented in the Wiki just now)

'--fetch_minimal_work' sounds like what you're looking for.
9) Message boards : GPUs : Request for Intel iGPU support to be a working feature again. (Message 91797)
Posted 14 days ago by Richard Haselgrove
Post:
Have a look at the opening post in another intel gpu opencl thread at Einstein. That's the same UHD 620 that I used in this test. You'll see that I got it to fetch work (needs Beta apps enabled), but not to run it - it stalled after doing the initial data setup, and never reached the first checkpoint.

I also refer to a SETI test in that thread: like then, today's test with the SETI app has led to an inconclusive result, so I suspect Intel have tweaked the precision or driver optimisation again. Bernd is aware of the Einstein problem, but too busy to do anything about it: I don't feel like waking Raistmer out of his slumbers before I set off on my travels. So I'm inclined to let this one lie until some cold, boring, winter's evening.
10) Message boards : The Lounge : The Seti is Down Cafe (Message 91795)
Posted 14 days ago by Richard Haselgrove
Post:
Well, it's now 05:20 PDT, and nothing's happened y.........
11) Message boards : GPUs : Request for Intel iGPU support to be a working feature again. (Message 91793)
Posted 14 days ago by Richard Haselgrove
Post:
As you can see in my previous post (now complete), BOINC can see my OEM driver just fine. So...

running my secondary monitor with the iGPU
Yes, that's normally a prerequisite for iGPU computing. I have 'headless' dummy plugs on the motherboard outputs on all my NVidia-equipped desktops.

the shift to Windows DCH drivers
That's beginning to sound like the nub of the problem. I'll see if I can replicate.

Now downloading driver 25.20.100.6374 dated 18 November 2018 from Microsoft.

And that's got OpenCL support too:

  Platform Name                                   Intel(R) OpenCL
Number of devices                                 2
  Device Name                                     Intel(R) UHD Graphics 620
  Device Vendor                                   Intel(R) Corporation
  Device Vendor ID                                0x8086
  Device Version                                  OpenCL 2.1 NEO
  Driver Version                                  25.20.100.6374
  Device OpenCL C Version                         OpenCL C 2.0
which BOINC can see:

11/06/2019 13:15:21 |  | Starting BOINC client version 7.14.2 for windows_x86_64
11/06/2019 13:15:21 |  | log flags: file_xfer, sched_ops, task, cpu_sched, sched_op_debug
11/06/2019 13:15:21 |  | Libraries: libcurl/7.47.1 OpenSSL/1.0.2g zlib/1.2.8
11/06/2019 13:15:21 |  | Data directory: C:\BOINCdata
11/06/2019 13:15:21 |  | Running under account Richard Haselgrove
11/06/2019 13:15:22 |  | OpenCL: Intel GPU 0: Intel(R) UHD Graphics 620 (driver version 25.20.100.6374, device version OpenCL 2.1 NEO, 3166MB, 3166MB available, 211 GFLOPS peak)
11/06/2019 13:15:22 |  | OpenCL CPU: Intel(R) Core(TM) i5-8250U CPU @ 1.60GHz (OpenCL driver vendor: Intel(R) Corporation, driver version 7.6.0.716, device version OpenCL 2.1 (Build 716))
and use:

11/06/2019 13:21:57 | SETI@home | [cpu_sched] Starting task blc25_2bit_guppi_58406_23898_HIP20842_0094.28629.818.21.44.208.vlar_1 using setiathome_v8 version 820 (opencl_intel_gpu_sah) in slot 4
12) Message boards : GPUs : Request for Intel iGPU support to be a working feature again. (Message 91791)
Posted 14 days ago by Richard Haselgrove
Post:
OK, it sounds like it's time to do some testing again. I've got a Windows 10 i5 on the desk in front of me, and BOINC saw the Intel GPU just fine when I bought it - running build 1809. It's recently updated itself to 1903, so worth another look. It's an ultraportable, so I don't usually use the iGPU for crunching (not enough cooling), and I'll have to switch BOINC out of service mode when I get to that stage: but I'll update this post as I go along.

Hardware: i5 8250U CPU, Intel UHD 620
Software: Windows 10 Pro, version 1903, build 18362.145
Graphics driver: 23.20.16.4973 dated 28 Feb 2018 (OEM supplied)

Well, Oblomov's CLinfo sees the right sort of detail - I'll switch to that machine to post.

Number of platforms                               1
  Platform Name                                   Intel(R) OpenCL
  Platform Vendor                                 Intel(R) Corporation
  Platform Version                                OpenCL 2.1
  Platform Profile                                FULL_PROFILE
  Platform Extensions                             cl_intel_dx9_media_sharing cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_d3d11_sharing cl_khr_depth_images cl_khr_dx9_media_sharing cl_khr_fp64 cl_khr_gl_sharing cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_icd cl_khr_image2d_from_buffer cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_spir
  Platform Host timer resolution                  100ns
  Platform Extensions function suffix             INTEL

  Platform Name                                   Intel(R) OpenCL
Number of devices                                 2
  Device Name                                     Intel(R) UHD Graphics 620
  Device Vendor                                   Intel(R) Corporation
  Device Vendor ID                                0x8086
  Device Version                                  OpenCL 2.1 NEO
  Driver Version                                  23.20.16.4973
  Device OpenCL C Version                         OpenCL C 2.1
  Device Type                                     GPU
...
  Device Name                                     Intel(R) Core(TM) i5-8250U CPU @ 1.60GHz
  Device Vendor                                   Intel(R) Corporation
  Device Vendor ID                                0x8086
  Device Version                                  OpenCL 2.1 (Build 611)
  Driver Version                                  7.6.0.611
  Device OpenCL C Version                         OpenCL C 2.0
  Device Type                                     CPU
And BOINC sees

11/06/2019 12:49:08 |  | Starting BOINC client version 7.14.2 for windows_x86_64
11/06/2019 12:49:08 |  | log flags: file_xfer, sched_ops, task, cpu_sched, sched_op_debug
11/06/2019 12:49:08 |  | Libraries: libcurl/7.47.1 OpenSSL/1.0.2g zlib/1.2.8
11/06/2019 12:49:08 |  | Data directory: C:\BOINCdata
11/06/2019 12:49:08 |  | Running under account Richard Haselgrove
11/06/2019 12:49:08 |  | OpenCL: Intel GPU 0: Intel(R) UHD Graphics 620 (driver version 23.20.16.4973, device version OpenCL 2.1 NEO, 3166MB, 3166MB available, 211 GFLOPS peak)
11/06/2019 12:49:08 |  | OpenCL CPU: Intel(R) Core(TM) i5-8250U CPU @ 1.60GHz (OpenCL driver vendor: Intel(R) Corporation, driver version 7.6.0.611, device version OpenCL 2.1 (Build 611))
11/06/2019 12:49:08 |  | Processor: 8 GenuineIntel Intel(R) Core(TM) i5-8250U CPU @ 1.60GHz [Family 6 Model 142 Stepping 10]
11/06/2019 12:49:08 |  | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 fma cx16 sse4_1 sse4_2 movebe popcnt aes f16c rdrandsyscall nx lm avx avx2 vmx tm2 pbe fsgsbase bmi1 smep bmi2
11/06/2019 12:49:08 |  | OS: Microsoft Windows 10: Professional x64 Edition, (10.00.18362.00)
13) Message boards : Questions and problems : Boinc Manager impossible to uninstall (Message 91781)
Posted 16 days ago by Richard Haselgrove
Post:
Even if you didn't keep the original downloaded installer executable, you can download it again - look at any project you attached the laptop to, if you can't remember which version you had installed. Your account/host list will still display that information. All released version are kept in https://boinc.berkeley.edu/dl/, not least for this purpose.
14) Message boards : Questions and problems : Boinc Manager impossible to uninstall (Message 91779)
Posted 16 days ago by Richard Haselgrove
Post:
The Windows installer is trying to find the instructions for UNinstalling the previous installation. They are included in the boinc.msi file used to INSTALL that previous version.

Windows should keep that version of the file lying around for reference, but if it's gone missing - install the exact same version as you installed before, and then uninstall it again. You (as administrator) are installing it, and you (as the same administrator) are uninstalling it: the missing boinc account shouldn't make any difference.
15) Message boards : Questions and problems : blue of screen death when using my GPU (Message 91757)
Posted 18 days ago by Richard Haselgrove
Post:
From Wikipedia:

Scalable Link Interface (SLI) is a brand name for a multi-GPU technology developed by Nvidia for linking two or more video cards together to produce a single output.
If you have two (or sometimes even more) identical NVidia GPUs, you can hook them together to form the equivalent of one big GPU.

That doesn't help with the work we do in BOINC: BOINC will still see you as having multiple smaller cards, and will use them independently. It's probably better not to have the SLI link present for our sort of work.
16) Message boards : Questions and problems : blue of screen death when using my GPU (Message 91752)
Posted 18 days ago by Richard Haselgrove
Post:
Why that is wedged in there I'm not sure.
Advertising. Usually the advertiser's (interpolated) download button is larger, more prominent and more eye-catching than the download product that brought you to the page in the first place.

They cause vastly more problems than they ever cure, and you need to exercise care only to download the product you really want.

Has anyone successfully used adblockers against this curse?
17) Message boards : GPUs : GPU usage utility available? (Message 91735)
Posted 20 days ago by Richard Haselgrove
Post:

(this message board uses BBcode, rather than HTML, and PNG format preserves sharp detail better than JPG)
18) Message boards : GPUs : GPU usage utility available? (Message 91733)
Posted 20 days ago by Richard Haselgrove
Post:
Thanks - that helps. We can look at Valid tasks for computer 8710401 and see that your GPU is, in fact, returning valid work - most recently at 5 Jun 2019, 20:49:10 UTC (less than an hour ago).

It is possible that your machine is working fine when you're not looking at it, but switching itself off (for SETI, at least) when you look at it, simply because it thinks you want to do something more important with it.

Your description of 'stop signs' and green arrows confirms that you are using the 'Simple View' of the BOINC Manager program: if you are interested and want to get more involved involved in the SETI project, you can switch to the 'Advanced View' and find more information about what's going on and how you can manage it. But that's not important: you're fine as you are.
19) Message boards : GPUs : GPU usage utility available? (Message 91729)
Posted 20 days ago by Richard Haselgrove
Post:
Also, look in your event log for a message like the following
Please post the whole of the first 30 lines, not just selected highlights. You might leave out the parts that seem unimportant you, but are meaningful to us.
20) Message boards : GPUs : GPU usage utility available? (Message 91723)
Posted 20 days ago by Richard Haselgrove
Post:
How do you know that the GPUs aren't being used? What exactly are you seeing, or not seeing, that tells you "no go"?

The most useful information to give potential helpers is the first 30 lines or so of the Event Log, after a fresh system restart. A lot to get our collective teeth into in there.


Next 20

Copyright © 2019 University of California. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.