Windows SSE 32-bit Computation errors
log in

Advanced search

Message boards : Number crunching : Windows SSE 32-bit Computation errors

Author Message
Dr Who Fan
Send message
Joined: 12 Oct 16
Posts: 4
Credit: 111,788
RAC: 0
United States
Message 1341 - Posted: 6 Aug 2018, 15:32:34 UTC

6 of 6 tasks failed the same way with "Computation errors" after task run a few seconds on the new Windows SSE 32-bit tasks on my Windows XP host ...

Example: https://gen.disi.unitn.it/test/result.php?resultid=34999192

Stderr output
<core_client_version>7.12.1</core_client_version> <![CDATA[ <message> (unknown error) - exit code -1073741795 (0xc000001d)</message> <stderr_txt> Start @ Mon Aug 06 10:10:21 2018 Unhandled Exception Detected... - Unhandled Exception Record - Reason: Illegal Instruction (0xc000001d) at address 0x00402928 Engaging BOINC Windows Runtime Debugger... ********************

____________

Profile valterc
Project administrator
Project tester
Send message
Joined: 30 Oct 13
Posts: 623
Credit: 34,677,535
RAC: 2
Italy
Message 1342 - Posted: 7 Aug 2018, 10:58:24 UTC

It seems that your OS/CPU doesn't fully support SSE2 (that's the meaning of 'illegal instruction'). If I remember correctly XP needs SP3 for this, but it seems that you already have it installed. Please check with some tools (like CPU-Z) if SSE2 is correctly enabled on your host.

Dr Who Fan
Send message
Joined: 12 Oct 16
Posts: 4
Credit: 111,788
RAC: 0
United States
Message 1343 - Posted: 7 Aug 2018, 15:49:21 UTC - in response to Message 1342.

Please check with some tools (like CPU-Z) if SSE2 is correctly enabled on your host.

Highly doubt that is the problem as I can run SSE2 apps at other project with no problems and it is correctly reported by the OS.

I suspect it is the compiler and/or compiler switches you are using. I have seen the same problems at some newer projects or project that have used new(er) MS Visual Studio Compilers that are not Windows XP compatible by default.

See: Configuring Programs for Windows XP


From CPU-Z >>
Specification: Intel(R) Celeron(R) CPU 2.40GHz
Instructions sets: MMX, SSE, SSE2
Windows Version: Microsoft Windows XP Home Edition Service Pack 3 (Build 2600)


From BOINC Startup:
07-Aug-2018 01:11:49 [---] Starting BOINC client version 7.12.1 for windows_intelx86
07-Aug-2018 01:12:16 [---] Processor: 1 GenuineIntel Intel(R) Celeron(R) CPU 2.40GHz [Family 15 Model 2 Stepping 9]
07-Aug-2018 01:12:16 [---] 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 pbe
07-Aug-2018 01:12:16 [---] OS: Microsoft Windows XP: Home x86 Edition, Service Pack 3, (05.01.2600.00)

____________

Profile valterc
Project administrator
Project tester
Send message
Joined: 30 Oct 13
Posts: 623
Credit: 34,677,535
RAC: 2
Italy
Message 1344 - Posted: 7 Aug 2018, 16:46:17 UTC - in response to Message 1343.
Last modified: 7 Aug 2018, 16:46:44 UTC

OK. It's difficult to figure out the problem, there are many XP computers that compute correctly (like hostid=29204). The Windows x32 SSE2 version of the application was actually compiled using one of the latest Cygwin (gcc MingW 6.4.0).

m
Send message
Joined: 27 May 17
Posts: 4
Credit: 8,488,193
RAC: 0
United Kingdom
Message 1401 - Posted: 3 Dec 2018, 12:50:59 UTC
Last modified: 3 Dec 2018, 13:06:06 UTC

There are a few 32bit XP boxes here that I keep to help out with LHC Sixtrack tasks. They suffer from the same problem. They run sse2 Sixtrack tasks OK.

The non sse2 application runs OK but the latest version available is 1.10 which sometimes fails to validate against other Windows hosts. Linux seems OK. As an alternative to getting the sse2 version to work I would very much like to try a "vanilla" (non sse2) 32bit Windows version 1.11 if one could be made available.

Profile valterc
Project administrator
Project tester
Send message
Joined: 30 Oct 13
Posts: 623
Credit: 34,677,535
RAC: 2
Italy
Message 1404 - Posted: 3 Dec 2018, 14:49:01 UTC - in response to Message 1401.
Last modified: 3 Dec 2018, 14:54:25 UTC

There are a few 32bit XP boxes here that I keep to help out with LHC Sixtrack tasks. They suffer from the same problem. They run sse2 Sixtrack tasks OK.

The non sse2 application runs OK but the latest version available is 1.10 which sometimes fails to validate against other Windows hosts. Linux seems OK. As an alternative to getting the sse2 version to work I would very much like to try a "vanilla" (non sse2) 32bit Windows version 1.11 if one could be made available.

Yep, the 1.10 versions are not 100% compatible with 1.11. If you would like to try (using the anonymous platform app_info.xml) this is the link for a Win 32bit no SIMD version: https://gene.disi.unitn.it/test/download/gene_pcim_v1.11_win32.exe

m
Send message
Joined: 27 May 17
Posts: 4
Credit: 8,488,193
RAC: 0
United Kingdom
Message 1407 - Posted: 3 Dec 2018, 16:04:26 UTC

Many thanks. The first couple of tasks are in progress, but it will be a good while before we know if all is well. Thanks again.

Yuriy Naydenov
Send message
Joined: 22 Jun 19
Posts: 6
Credit: 5,284,099
RAC: 0
Russia
Message 2351 - Posted: 5 Aug 2021, 6:59:49 UTC

Faced a similar problem.
http://gene.disi.unitn.it/test/show_host_detail.php?hostid=73254

Yuriy Naydenov
Send message
Joined: 22 Jun 19
Posts: 6
Credit: 5,284,099
RAC: 0
Russia
Message 2358 - Posted: 14 Aug 2021, 8:44:38 UTC
Last modified: 14 Aug 2021, 8:48:24 UTC

What can be done in this case?
http://gene.disi.unitn.it/test/results.php?hostid=73254

World Community Grid works without errors.

Falconet
Send message
Joined: 21 Dec 16
Posts: 105
Credit: 3,092,711
RAC: 0
Portugal
Message 2359 - Posted: 14 Aug 2021, 10:43:57 UTC - in response to Message 2358.

If I had to guess, I'd say that CPU is very old and the app may not be compatible with it. Probably would need a new, compiled version to work.

I'd either retire the CPU or find some other project where it actually works. Do keep in mind it is an old, power hungry and inneficient CPU.
____________

Yuriy Naydenov
Send message
Joined: 22 Jun 19
Posts: 6
Credit: 5,284,099
RAC: 0
Russia
Message 2360 - Posted: 14 Aug 2021, 13:19:32 UTC - in response to Message 2359.

If I had to guess, I'd say that CPU is very old and the app may not be compatible with it. Probably would need a new, compiled version to work.

I'd either retire the CPU or find some other project where it actually works. Do keep in mind it is an old, power hungry and inneficient CPU.

This is kind of an experiment, I know this CPU is old. If we talk about energy efficiency, I am from Russia and electricity is relatively cheap here, in my case 0.031 €.
I have several of these old computer hardware, they are working, and I am a little sorry that they are gathering dust idle.
At the moment, WCG (Mapping Cancer Markers) is running on this CPU with a runtime of about 8 hours.

It's a bit of a shame that older CPUs don't have support compared to the same mobile ARM chips that probably have similar performance. But I don’t insist.

Best regards.

manalog
Send message
Joined: 5 Oct 15
Posts: 33
Credit: 1,098,442
RAC: 0
Italy
Message 2361 - Posted: 14 Aug 2021, 14:02:11 UTC

If these machines are intended to run boinc full time, then install Linux on them. I assure that tn-grid without optimizations runs on P4s (like yours), then if I find it I could give you an optimized app I made for P4 and SSE2 last year.

Falconet
Send message
Joined: 21 Dec 16
Posts: 105
Credit: 3,092,711
RAC: 0
Portugal
Message 2362 - Posted: 14 Aug 2021, 14:45:27 UTC - in response to Message 2360.
Last modified: 14 Aug 2021, 14:48:03 UTC

If I had to guess, I'd say that CPU is very old and the app may not be compatible with it. Probably would need a new, compiled version to work.

I'd either retire the CPU or find some other project where it actually works. Do keep in mind it is an old, power hungry and inneficient CPU.

This is kind of an experiment, I know this CPU is old. If we talk about energy efficiency, I am from Russia and electricity is relatively cheap here, in my case 0.031 €.
I have several of these old computer hardware, they are working, and I am a little sorry that they are gathering dust idle.
At the moment, WCG (Mapping Cancer Markers) is running on this CPU with a runtime of about 8 hours.

It's a bit of a shame that older CPUs don't have support compared to the same mobile ARM chips that probably have similar performance. But I don’t insist.

Best regards.



But it is environmentally not very friendly.

ARM is a growing architecture. There are many more ARM devices than there are old CPUs like the Celeron you have or Pentium 4s or old Athlons. Not to mention a 4-core ARM device will put out a lot more work than the Celeron for a mere fraction of the power consumption.

The last computer I owned with a Pentium 4 used up around 130 watts at 100% usage. I have a 2013 tablet with an Intel Z3740 SOC that runs MCM in 7-8 hours but uses just 9 watts - not an ARM-based tablet but you get the point.

If it's running MCM successfully, that's good. You could also try Linux and see if it runs TN-Grid. If it does, you also get the added benefit of a performance improvement since TN-Grid is known to run quite a quicker under Linux than under Windows - although I'm not sure that applies to 32-bit architectures.
____________


Post to thread

Message boards : Number crunching : Windows SSE 32-bit Computation errors


Main page · Your account · Message boards


Copyright © 2024 CNR-TN & UniTN