Busy indeed. Right now we're struggling with multiple hardware failures and urgent infrastructure upgrades to keep the project running and our volunteers happy :-/ I'll forward this issue again to bring it to the right people's attention...
If I remember correctly (waiting for confirmation from Bernd), Bernd had configured a beta-app version for Intel iGPUs that does not check for a max driver version. The intention was that people can use this (by allowing beta apps in the web prefs) to explore if new drivers have fixed whatever problem existed.
So if we are lucky, the problem doesn't exist anymore with newer drivers.
This would have to be tested with several combinations of models of HD iGPUs and drivers to work out save driver versions to allow again. I guess we from E@H need volunteers' help here, because we have very few Windows PCs with newer HD iGPUs under our control: anyone interested in helping can enable beta apps and see if the Intel HD iGPU tasks are validating against non-beta app version results, and should post both negative and positive results to this thread.
EDIT: for some of the newer models, we might have to edit the scheduler config to correctly detect the Intel iGPU. So if you cannot get work even with beta apps enabled, please point us towards your host so we can see how BOINC reports the GPU's name.
From my experience, the problem with Intel drivers does not exist anymore for Windows 10. I have five Windows 10 hosts with Intel HD 4600 GPUs, they all use the drivers supplied automatically by Windows (driver version 10.18.15.4256) and produce valid results: host1 host2 host3 host4 host5
They use the beta apps though...
I can't say anything about Windows 7 or 8 and their related drivers.
From my experience, the problem with Intel drivers does not exist anymore for Windows 10. I have five Windows 10 hosts with Intel HD 4600 GPUs, they all use the drivers supplied automatically by Windows (driver version 10.18.15.4256) and produce valid results: host1 host2 host3 host4 host5
They use the beta apps though...
I can't say anything about Windows 7 or 8 and their related drivers.
Great news! Do the (internal) Windows 10 drivers have an Intel-like version number such that we could use it as in indicator for working upstream drivers, helping Win7/8 users maybe...?
Great news! Do the (internal) Windows 10 drivers have an Intel-like version number such that we could use it as in indicator for working upstream drivers, helping Win7/8 users maybe...?
Oliver
I posted the version number above: 10.18.15.4256
The (internal) Windows 10 driver is actually an Intel driver:
edit: Ok, Intel has a new numbering style for Windows 10, i.e. 15.36.26.4294 seems to be the same as 10.18.14.4294. So the last 4 digits matter...
edit2: BOINC reports: 3/31/2016 5:36:26 PM | | OpenCL: Intel GPU 0: Intel(R) HD Graphics 4600 (driver version 10.18.15.4256, device version OpenCL 1.2, 1922MB, 1922MB available, 184 GFLOPS peak)
when installing the latest driver (15.40.18.4380) from the Intel website it is reported by BOINC (and Windows) as: 03/31/2016 5:36:44 PM | | OpenCL: Intel GPU 0: Intel(R) HD Graphics 4600 (driver version 20.19.15.4380, device version OpenCL 1.2, 1630MB, 1630MB available, 208 GFLOPS peak)
I don't know yet if that driver works with E@H though.
What a mess...
I posted the version number above: 10.18.15.4256
The (internal) Windows 10 driver is actually an Intel driver:
So if this is the latest official/internal Win10 driver, we see that
* we have at least a
new last known good driver version (as it's newer than 10.18.10.3906): 10.18.15.4256
* Win10 seems to use pretty outdated drivers (compared to 15.40.18.4380)
Quote:
edit: Ok, Intel has a new numbering style for Windows 10, i.e. 15.36.26.4294 seems to be the same as 10.18.14.4294. So the last 4 digits matter...
Why new numbering style? It didn't change to my eyes, they might just be newer. How do you tell those two are the same? FYI, the last digits should be the build number which are an integral part of the whole version number.
Quote:
when installing the latest driver (15.40.18.4380) from the Intel website it is reported by BOINC (and Windows) as: 03/31/2016 5:36:44 PM | | OpenCL: Intel GPU 0: Intel(R) HD Graphics 4600 (driver version 20.19.15.4380, device version OpenCL 1.2, 1630MB, 1630MB available, 208 GFLOPS peak)
I don't know yet if that driver works with E@H though.
What a mess...
I posted the version number above: 10.18.15.4256
The (internal) Windows 10 driver is actually an Intel driver:
So if this is the latest official/internal Win10 driver, we see that
* we have at least a
new last known good driver version (as it's newer than 10.18.10.3906): 10.18.15.4256
* Win10 seems to use pretty outdated drivers (compared to 15.40.18.4380)
No, they don't work that way. From the driver I'm using under Windows 7:
Microsoft Windows* 7 64
Microsoft Windows* 8 64
Microsoft Windows* 8.1 64
Release Date: May 21, 2014
You usually see the 15 series number on Intel's download pages - I wish they'd call that the 'package' number, or something.
The bit we're interested in is the Graphics Driver, which always seems to be a 10 series number. That's what BOINC detects at startup:
Quote:
28-Mar-2016 09:23:02 [---] OpenCL: Intel GPU 0: Intel(R) HD Graphics 4600 (driver version 10.18.10.3621, device version OpenCL 1.2, 1298MB, 1298MB available, 192 GFLOPS peak)
though, unlike other GPUs, the driver version doesn't make it through to the computer details page on the server (SETI displays OpenCL version, Einstein doesn't even display that).
I'll try and find some time over the weekend to explore the arcane labyrinth of Intel's download centre, and see what I can find.
Busy indeed. Right now we're
)
Busy indeed. Right now we're struggling with multiple hardware failures and urgent infrastructure upgrades to keep the project running and our volunteers happy :-/ I'll forward this issue again to bring it to the right people's attention...
Please bear with us,
Oliver
Einstein@Home Project
Thanks for reminding us about
)
Thanks for reminding us about this issue.
If I remember correctly (waiting for confirmation from Bernd), Bernd had configured a beta-app version for Intel iGPUs that does not check for a max driver version. The intention was that people can use this (by allowing beta apps in the web prefs) to explore if new drivers have fixed whatever problem existed.
So if we are lucky, the problem doesn't exist anymore with newer drivers.
This would have to be tested with several combinations of models of HD iGPUs and drivers to work out save driver versions to allow again. I guess we from E@H need volunteers' help here, because we have very few Windows PCs with newer HD iGPUs under our control: anyone interested in helping can enable beta apps and see if the Intel HD iGPU tasks are validating against non-beta app version results, and should post both negative and positive results to this thread.
EDIT: for some of the newer models, we might have to edit the scheduler config to correctly detect the Intel iGPU. So if you cannot get work even with beta apps enabled, please point us towards your host so we can see how BOINC reports the GPU's name.
Cheers
HB
From my experience, the
)
From my experience, the problem with Intel drivers does not exist anymore for Windows 10. I have five Windows 10 hosts with Intel HD 4600 GPUs, they all use the drivers supplied automatically by Windows (driver version 10.18.15.4256) and produce valid results:
host1
host2
host3
host4
host5
They use the beta apps though...
I can't say anything about Windows 7 or 8 and their related drivers.
RE: From my experience, the
)
I guess I should try out the beta apps then
Yup beta app seems to be
)
Yup beta app seems to be working fine on the w10 sweet
Great news! Do the (internal)
)
Great news! Do the (internal) Windows 10 drivers have an Intel-like version number such that we could use it as in indicator for working upstream drivers, helping Win7/8 users maybe...?
Oliver
Einstein@Home Project
RE: Great news! Do the
)
I posted the version number above: 10.18.15.4256
The (internal) Windows 10 driver is actually an Intel driver:
edit: Ok, Intel has a new numbering style for Windows 10, i.e. 15.36.26.4294 seems to be the same as 10.18.14.4294. So the last 4 digits matter...
edit2: BOINC reports:
3/31/2016 5:36:26 PM | | OpenCL: Intel GPU 0: Intel(R) HD Graphics 4600 (driver version 10.18.15.4256, device version OpenCL 1.2, 1922MB, 1922MB available, 184 GFLOPS peak)when installing the latest driver (15.40.18.4380) from the Intel website it is reported by BOINC (and Windows) as:
03/31/2016 5:36:44 PM | | OpenCL: Intel GPU 0: Intel(R) HD Graphics 4600 (driver version 20.19.15.4380, device version OpenCL 1.2, 1630MB, 1630MB available, 208 GFLOPS peak)I don't know yet if that driver works with E@H though.
What a mess...
RE: I posted the version
)
So if this is the latest official/internal Win10 driver, we see that
new last known good driver version (as it's newer than 10.18.10.3906): 10.18.15.4256
* Win10 seems to use pretty outdated drivers (compared to 15.40.18.4380)
Why new numbering style? It didn't change to my eyes, they might just be newer. How do you tell those two are the same? FYI, the last digits should be the build number which are an integral part of the whole version number.
Indeed!
Thanks for the details
Einstein@Home Project
RE: RE: I posted the
)
No, they don't work that way. From the driver I'm using under Windows 7:
You usually see the 15 series number on Intel's download pages - I wish they'd call that the 'package' number, or something.
The bit we're interested in is the Graphics Driver, which always seems to be a 10 series number. That's what BOINC detects at startup:
though, unlike other GPUs, the driver version doesn't make it through to the computer details page on the server (SETI displays OpenCL version, Einstein doesn't even display that).
I'll try and find some time over the weekend to explore the arcane labyrinth of Intel's download centre, and see what I can find.
Sigh... Thanks for clearing
)
Sigh... Thanks for clearing this up!
Einstein@Home Project