Building on the work of Dan_Ee, I have compiled and deployed new Windows x64 apps. These are 50% to 100% faster than the 32-bit apps depending on your processor. Please let me know if you encounter problems.

Do people have to abort current windows wu's to download the new application or is it automatic?

Carlos

Already assigned WU's will use the old app. New assignments will usually get the x64 app. I say "usually" since BOINC will experiment with the old app for a few assignments to see which is faster. My recommendation is to simply let BOINC figure it out over time.

Can you quantify the number of windows and linux cores on NFS@Home universe?... just to have an idea of the jump in progress and credit NFS@Home will have.

Carlos

In the last week, 292 Linux computers, 555 Windows x86 computers, and 656 Windows x64 computers have communicated with the project. This doesn't count the number of cores in each, though.

Hi.

One of my computers seems to be aborting every lasievef 1.08 x64 job exactly at the 1 minute mark. It is computer ID 29759.
Any details that would help you trouble shoot?

It is a Windows 2008 R2 server.
Processor: AMD Phenom II 1090T. Slight overclock to 3.4 GHz. Additional cooling provided by over-sized heat sink.
Ram: 16 GB DDR3 1600 running at 1333.

Only running 1 job at a time on Windows.

Any other technical info that you require?
The machine has shown no stability issues and other work units complete just fine, even the lasievee 1.08 x64.

Thanks.

Enrique,

You can try to update BOINC to version 7.0.64 to eliminate one variable but I think the binaries weren't tested with Windows Server 2008 "R2"....but I can be wrong. You will have to work this out with Greg.

Carlos

Thanks Carlos.

I just updated the version of BOINC. Still failing on the lasievef 1.08 x64 work units I'm afraid.

Greg, anything you need that could assist?

Thank you.

Enrique

Enrique,

From http://boinc.berkeley.edu/wiki/Release_Notes

Known issues with BOINC 7.0.64/.65 and Seti when updating from BOINC 7.0.28 or before and getting 197 (0xc5) EXIT_TIME_LIMIT_EXCEEDED errors on anonymous platform GPU

When you've updated to BOINC 7.0.64/.65, coming from BOINC 7.0.28 or before, AND when crunching work on an anonymous platform GPU application (such as the ones from Lunatics), you have a good chance that all your work errs with a 197 (0xc5) EXIT_TIME_LIMIT_EXCEEDED error.

This is because of an addition to the code in BOINC 7.0.33, where when you run the anonymous platform and you have not supplied a flops value in your app_info.xml file, that the flops value used for GPU tasks is increased by a factor of 10.

It will only affect old work still in cache from before the upgrade. Any newer work your BOINC downloads, will not be affected.

Two ways to work around this:
1. Before you upgrade, set No New Tasks and run your cache empty. Report all. Then upgrade to the new version and set Allow New Tasks.
2. Done the upgrade with a full cache? Reset the project. This will throw away all present work in cache, but since the project has resend lost work available, your tasks will be resent to you and seen as new work.

Anyway, I will contact Greg by email.

Carlos

Actually that's an issue that baffles me! It's a problem when running on AMD Phenom(tm) II X6 CPU's. Intel x86_64 processors work fine. So do all AMD x86_64 processors it has been tried on except AMD Phenom(tm) II X6 processors. Even the AMD Phenom(tm) II X4 works fine.

Anyone have any idea what is different about the AMD Phenom(tm) II X6 that would cause an issue? If necessary, I will look into how to force computers using this processor to use the 32-bit binary.

Maybe you are looking at more info than what is available/visible to me.
But I'm not sure that the article applies.

Looking at the results for task 26650845, I see that the error is:

- Unhandled Exception Record -
Reason: Access Violation (0xc0000005) at address 0x0000000000429314 read attempt to address 0x0076206A

I don't see EXIT_TIME_LIMIT_EXCEEDED in the debugger output. I also don't see any messages logged in the Event Manager. All other failed work units show the exact same type of error.

I'll still follow the instructions, can't hurt to try. I'm going to detach the project and re-attach.

Reset the project and it still throws the same exception.

Greg's right, it must be something to do with these AMD Phenom IIs with 6 cores. All my other Windows machines are running the lasievef x64 units without a problem (or minimal amount of errors).

Perhaps it is the operating system too. In this case Windows 2008 R2.

This particular machine is also running an instance of Ubuntu under Hyper-V. No problems with any type of work units there.
At any given time, 4 cores are running lasievef x64 and lasievee x64 units under Linux; and one core will be running the same but under Windows.
The only units that are consistently failing are the lasievef x64 in Windows.

Perhaps the issue is the combination of this particular CPU and the OS.

Thanks.

Enrique

Trying an experiment here.

Changed the properties for the lasievef_1.08_windows_x86_64 binary for all users. Set the compatibility mode to Windows 7.

So far I've got two work units running for over 5 minutes without exceptions (all used to consistently fail at the 1 minute mark).

Will keep you posted. Thanks.

Enrique

Let me know how that goes. Yours is not the only Phenom II X6 that's throwing this error. There are at least 6 of them on the project right now running various versions of Windows x64 all with the same issue. Occasional work units run fine, but most throw this error. And it doesn't happen in Linux or on any other Intel or AMD processor that volunteers have tried. Very strange.

In any case, this weekend I can implement a check that sends the 32-bit binary to Phenom II X6's running Windows x64 and the 64-bit binary to all other processors running Windows x64.

Sounds good to me.

So far so good. It has successfully completed 5 work units with the lasievef 1.08 x64 binary since changing the compatibility mode. It started working on the 6th and just went past 5 minutes (again, it was consistently failing after exactly 1 minute).
Keeping my fingers crossed. Let's see how it does over the weekend.

Regards,

Enrique

Computers with Phenom II X6's running Windows should now get the 32-bit app. Please verify when you have a chance. Thanks!

FYI. After 24 hours of running the lasievef 64 bit client on a Phenom II x6 there are no more errors to report.

It looks like by changing the compatibility mode to Windows 7 for all users the problem went away.

I'm not sure what others reported, but if there's a workaround it would be great to be able to run the 64 bit clients on these machines.

Thanks.

Enrique

Funny and very disappointing, I didn't see any kind of speed improvement when running on my laptop (Win 7 64 bits). You can see the wu's I made over night. When I get the chance I will install Ubuntu on a VM.

Carlos

Building on the work of Dan_Ee, I have compiled and deployed new Windows x64 apps. These are 50% to 100% faster than the 32-bit apps depending on your processor. Please let me know if you encounter problems.

Is the source code of the Boinc app available somewhere ?

The non-BOINC code is available at the link above. The source modified for BOINC can be downloaded from https://www.dropbox.com/s/eyqy3ljo9gpymk6/lasieve4_win64_boinc.zip.