Re: strange CPU speedups with SMP on Athlon 64 X2
Nathan Becker wrote: I would be happy to post my exact C source that I use to do the benchmark, but I wanted to get some feedback first in case I'm just doing something stupid. Also, since I'm not subscribed to this list, please cc me directly regarding this topic. Hi Nathan, Cache issues may explain this. When 2 processes are allocating memory in parallel they'll be given different interleavings of pages which could explain the speedup. Start one process, get it to memset all its memory, then pause it and do the same thing. Then set them both running at the same time (ie. after they've each touched their memory in turn), what do you see? (By memset()ing the memory, you'll cause the kernel to actually allocate a physical page. By doing that one after the other, we hope to eliminate interleaving issues.) -- SUSE Labs, Novell Inc. Send instant messages to your online friends http://au.messenger.yahoo.com - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: strange CPU speedups with SMP on Athlon 64 X2
Nathan Becker wrote: I would be happy to post my exact C source that I use to do the benchmark, but I wanted to get some feedback first in case I'm just doing something stupid. Also, since I'm not subscribed to this list, please cc me directly regarding this topic. Hi Nathan, Cache issues may explain this. When 2 processes are allocating memory in parallel they'll be given different interleavings of pages which could explain the speedup. Start one process, get it to memset all its memory, then pause it and do the same thing. Then set them both running at the same time (ie. after they've each touched their memory in turn), what do you see? (By memset()ing the memory, you'll cause the kernel to actually allocate a physical page. By doing that one after the other, we hope to eliminate interleaving issues.) -- SUSE Labs, Novell Inc. Send instant messages to your online friends http://au.messenger.yahoo.com - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: strange CPU speedups with SMP on Athlon 64 X2
Hi, stupid question : isn't it possible that your motherboard does some sort of overclocking when it detects high cpu usage (bus activity, etc...) ? It should not be easy to check (rdtsc every second ?), but you might want to explore such a possibility. Regards, willy On Tue, Aug 30, 2005 at 12:16:04PM -0700, Nathan Becker wrote: > Hi, > > I'm having a strange problem when I benchmark some of my physics > simulation code on my new Athlon 64 X2 4800 machine. It occurs on all > current kernels that I have tested including 2.6.12.5 and 2.6.13. > > If I run my benchmark single threaded, so that one of the two CPU cores > is just idling then the calculation goes pretty fast. But if I load both > CPU cores simultaneously but with INDEPENDENT calculations, then each > calculation runs about 12-15% faster than when running alone. I have > found this to be always reproducible. There is no disk access involved > in the calculation and RAM usage is fairly minimal so this is not caused > by caching. Also, if I compile the kernel to disable SMP then the machine > runs a single calculation at the same speed as when running alone when > SMP is enabled. > > I am aware of the timing issues on these machines (especially since I > reported the bug http://bugzilla.kernel.org/show_bug.cgi?id=5105 ). > However, I double-checked my benchmark with a stop-watch, so this is > independent of something strange happening in the timer. > > I also checked the cpufreq governor and according to the logs, my CPU is > holding steady at the maximum setting of 2.4GHz. I set the governor to > "performance" mode which should prevent unintended downclocking. > > I would be happy to post my exact C source that I use to do the > benchmark, but I wanted to get some feedback first in case I'm just doing > something stupid. Also, since I'm not subscribed to this list, please cc > me directly regarding this topic. > > Thanks very much, > > Nathan > - > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in > the body of a message to [EMAIL PROTECTED] > More majordomo info at http://vger.kernel.org/majordomo-info.html > Please read the FAQ at http://www.tux.org/lkml/ - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
strange CPU speedups with SMP on Athlon 64 X2
Hi, I'm having a strange problem when I benchmark some of my physics simulation code on my new Athlon 64 X2 4800 machine. It occurs on all current kernels that I have tested including 2.6.12.5 and 2.6.13. If I run my benchmark single threaded, so that one of the two CPU cores is just idling then the calculation goes pretty fast. But if I load both CPU cores simultaneously but with INDEPENDENT calculations, then each calculation runs about 12-15% faster than when running alone. I have found this to be always reproducible. There is no disk access involved in the calculation and RAM usage is fairly minimal so this is not caused by caching. Also, if I compile the kernel to disable SMP then the machine runs a single calculation at the same speed as when running alone when SMP is enabled. I am aware of the timing issues on these machines (especially since I reported the bug http://bugzilla.kernel.org/show_bug.cgi?id=5105 ). However, I double-checked my benchmark with a stop-watch, so this is independent of something strange happening in the timer. I also checked the cpufreq governor and according to the logs, my CPU is holding steady at the maximum setting of 2.4GHz. I set the governor to "performance" mode which should prevent unintended downclocking. I would be happy to post my exact C source that I use to do the benchmark, but I wanted to get some feedback first in case I'm just doing something stupid. Also, since I'm not subscribed to this list, please cc me directly regarding this topic. Thanks very much, Nathan - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
strange CPU speedups with SMP on Athlon 64 X2
Hi, I'm having a strange problem when I benchmark some of my physics simulation code on my new Athlon 64 X2 4800 machine. It occurs on all current kernels that I have tested including 2.6.12.5 and 2.6.13. If I run my benchmark single threaded, so that one of the two CPU cores is just idling then the calculation goes pretty fast. But if I load both CPU cores simultaneously but with INDEPENDENT calculations, then each calculation runs about 12-15% faster than when running alone. I have found this to be always reproducible. There is no disk access involved in the calculation and RAM usage is fairly minimal so this is not caused by caching. Also, if I compile the kernel to disable SMP then the machine runs a single calculation at the same speed as when running alone when SMP is enabled. I am aware of the timing issues on these machines (especially since I reported the bug http://bugzilla.kernel.org/show_bug.cgi?id=5105 ). However, I double-checked my benchmark with a stop-watch, so this is independent of something strange happening in the timer. I also checked the cpufreq governor and according to the logs, my CPU is holding steady at the maximum setting of 2.4GHz. I set the governor to performance mode which should prevent unintended downclocking. I would be happy to post my exact C source that I use to do the benchmark, but I wanted to get some feedback first in case I'm just doing something stupid. Also, since I'm not subscribed to this list, please cc me directly regarding this topic. Thanks very much, Nathan - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: strange CPU speedups with SMP on Athlon 64 X2
Hi, stupid question : isn't it possible that your motherboard does some sort of overclocking when it detects high cpu usage (bus activity, etc...) ? It should not be easy to check (rdtsc every second ?), but you might want to explore such a possibility. Regards, willy On Tue, Aug 30, 2005 at 12:16:04PM -0700, Nathan Becker wrote: Hi, I'm having a strange problem when I benchmark some of my physics simulation code on my new Athlon 64 X2 4800 machine. It occurs on all current kernels that I have tested including 2.6.12.5 and 2.6.13. If I run my benchmark single threaded, so that one of the two CPU cores is just idling then the calculation goes pretty fast. But if I load both CPU cores simultaneously but with INDEPENDENT calculations, then each calculation runs about 12-15% faster than when running alone. I have found this to be always reproducible. There is no disk access involved in the calculation and RAM usage is fairly minimal so this is not caused by caching. Also, if I compile the kernel to disable SMP then the machine runs a single calculation at the same speed as when running alone when SMP is enabled. I am aware of the timing issues on these machines (especially since I reported the bug http://bugzilla.kernel.org/show_bug.cgi?id=5105 ). However, I double-checked my benchmark with a stop-watch, so this is independent of something strange happening in the timer. I also checked the cpufreq governor and according to the logs, my CPU is holding steady at the maximum setting of 2.4GHz. I set the governor to performance mode which should prevent unintended downclocking. I would be happy to post my exact C source that I use to do the benchmark, but I wanted to get some feedback first in case I'm just doing something stupid. Also, since I'm not subscribed to this list, please cc me directly regarding this topic. Thanks very much, Nathan - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/