Back
[02:02:49] <skunkworks> skunkworks is now known as credak
[02:03:04] <credak> credak is now known as skunkworks
[10:47:08] <SWPadnos_> SWPadnos_ is now known as SWPadnos
[18:44:46] <SWPadnos> interesting. I booted my athlon64 (3500) machine with the EMC2 liveCD, and I get results similar to jmkasunich when I run a CPU hog during the latency tests
[18:44:57] <SWPadnos> this is a single core CPU
[18:45:53] <SWPadnos> the latency test alone gets averages in the 3355 range (+/- 10) without the other process running
[18:46:35] <SWPadnos> with the hog running, the average goes down to 2565 +/- 10
[19:12:03] <jepler> cycles or time?
[19:12:58] <jepler> could the switch from user->rt be that different (and faster!) than kernel->rt? different wouldn't surprise me, but faster would..
[19:13:43] <jepler> it could be the time to wake from 'hlt' instruction
[19:13:50] <SWPadnos> that's the number printed from the RTAI testsuite, so I think it's ns
[19:13:57] <jepler> which you'd pay frequently for an idle machine, but rarely for a CPU-hog machine
[19:14:24] <SWPadnos> right - we had considered idle or cache effects as possible culprits
[19:14:38] <SWPadnos> I didn't get this effect on mmy celeron 500
[19:16:56] <jepler> read about "C1E enhanced halt state",
http://techreport.com/reviews/2005q1/pentium4-600/index.x?pg=2
[19:17:09] <SWPadnos> jmkactually had the effect even when he niced the CPU hog to 19
[19:17:15] <jepler> Intel's new C1E halt state is also invoked by the HLT command, but it turns down the entire CPU's clock frequency (via multiplier control) and voltage in order to work its mojo. This more robust halt state requires significantly less power than the old C1 implementation.
[19:18:45] <SWPadnos> hmmm. I wonder if it's actually possible that the wall time is the same (or so), but the number of clocks is lower ...
[19:19:13] <SWPadnos> I also wonder if a similar thing is implemented in AMD CPUs
[19:20:28] <jepler> yes, there's something with the same name even:
http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_9485_13041%5E14633,00.html#114437
[19:21:31] <jepler> http://www.cpuid.com/cpuz.php
[19:21:32] <jepler> Q: CPU-Z reports my CPU speed below its stock frequency.
[19:21:32] <jepler> A: This is the effect of C1E (Enhanced Halt State) and/or EIST (Enhanced Intel SpeedStep Technology). Load your system and you will see the frequency increase to its nominal value.
[19:21:33] <SWPadnos> interesting. that means that my CPU doesn't have the feature, since it's not an X2 nor is it 65 nm (I think)
[19:23:01] <SWPadnos> it doesn't appear as though /proc/cpuinfo tells me
[19:23:17] <jepler> I dunno if it's a /proc/cpuinfo item
[19:23:55] <SWPadnos> I wasn't sure if it might be in the CPU flags, but I don't see it
[19:24:11] <SWPadnos> (or any power related stuff
[19:24:13] <SWPadnos> )