[Bug 615779] New: Lynnfield (Core i5) is underperforming
http://bugzilla.novell.com/show_bug.cgi?id=615779 http://bugzilla.novell.com/show_bug.cgi?id=615779#c0 Summary: Lynnfield (Core i5) is underperforming Classification: openSUSE Product: openSUSE 11.3 Version: Factory Platform: x86-64 OS/Version: openSUSE 11.3 Status: NEW Severity: Normal Priority: P5 - None Component: Kernel AssignedTo: kernel-maintainers@forge.provo.novell.com ReportedBy: vincent.lejeune@institutoptique.fr QAContact: qa@suse.de Found By: --- Blocker: --- User-Agent: Mozilla/5.0 (X11; U; Linux; fr-FR) AppleWebKit/532.4 (KHTML, like Gecko) konqueror/4.4.3 Safari/532.4 Performances are quite bad with core i5 hardware. I have a laptop with a core i5 430M (2,27 Ghz), and my opensuse is slower than ubuntu by about 20% for mathematical operations. It is even worse when comparing with windows, which is 4 times faster for doing the same computation ! The program used to do benchmark is below. The performance are the same when using gcc and when using icc. Reproducible: Always Steps to Reproduce: 1. Download Eigen 3 dev version 2. Build the program below with O3, -msse4.2, with include pointing to Eigen source 3. Look at the results Actual Results: ~8s Expected Results: less than 8s The sample program is : #include <Eigen/QR> #include <ctime> #include <iostream> using namespace Eigen; typedef Matrix<double,Dynamic,Dynamic> MAT; int main() { MAT A=MAT::Random(2048,2048); clock_t t0=clock(); HouseholderQR<MAT> qr(A); clock_t t1=clock(); double T=t1-t0; T/=CLOCKS_PER_SEC; std::cout<<T; return 0; } Obviously you need to set the include to Eigen source with the -I flag to pass to gcc. -- Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.
https://bugzilla.novell.com/show_bug.cgi?id=615779 https://bugzilla.novell.com/show_bug.cgi?id=615779#c Jeff Mahoney <jeffm@novell.com> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |jeffm@novell.com, | |matz@novell.com AssignedTo|kernel-maintainers@forge.pr |mgalbraith@novell.com |ovo.novell.com | -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.
https://bugzilla.novell.com/show_bug.cgi?id=615779 https://bugzilla.novell.com/show_bug.cgi?id=615779#c1 Michael Matz <matz@novell.com> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEW |NEEDINFO InfoProvider| |vljn@ovi.com --- Comment #1 from Michael Matz <matz@novell.com> 2011-04-26 12:53:41 UTC --- Can you get an profile with oprofile of this program, for both systems, opensuse and ubuntu? -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.
https://bugzilla.novell.com/show_bug.cgi?id=615779 https://bugzilla.novell.com/show_bug.cgi?id=615779#c2 --- Comment #2 from Michael Matz <matz@novell.com> 2011-04-26 13:06:08 UTC --- FWIW, my core i7 860 (2.8GHz) needs 1.5 seconds for the HouseholderQR step. It should be noted that this preconditioning of the matrix heavily depends on the actual contents of it (including time until convergence) and hence using random data is not at all a useful testcase, it would be better to initialize the matrix with some known values. I bet that's also the reason why Windows is so much faster here, its random generator is fairly low quality. As this is a mobile processor: does anything change if you disable any powersaving (just so to rule out on obvious reason for slowdowns)? -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.
https://bugzilla.novell.com/show_bug.cgi?id=615779 https://bugzilla.novell.com/show_bug.cgi?id=615779#c3 --- Comment #3 from Mike Galbraith <mgalbraith@novell.com> 2011-04-26 13:52:18 UTC --- my 2.4 GHz E5620 running openSUSE-11.3 with the desktop kernel runs it in 2.28 seconds, and it's nearly all userspace. perf record ./testcase perf report --sort comm,dso # Samples: 5754748817 # # Overhead Command Shared Object # ........ ........ ..................... # 97.44% testcase testcase 1.45% testcase [kernel.kallsyms] 1.11% testcase libc-2.11.2.so I don't have an i5 to try, but perhaps you can try perf record and perf report to see where it's spending it's time. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.
https://bugzilla.novell.com/show_bug.cgi?id=615779 https://bugzilla.novell.com/show_bug.cgi?id=615779#c4 Jeff Mahoney <jeffm@suse.com> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEEDINFO |RESOLVED InfoProvider|vljn@ovi.com | Resolution| |WONTFIX --- Comment #4 from Jeff Mahoney <jeffm@suse.com> 2014-08-08 16:27:28 EDT --- This report is against openSUSE 11.4 which is no longer under maintenance. If you are able to reproduce it with openSUSE 13.1 or openSUSE Factory, please re-open and reset the the "Product" field to the appropriate release. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.
participants (1)
-
bugzilla_noreply＠novell.com