21) Questions and Answers : Unix/Linux : Possibly Optimized Linux model to download for Beta Testers (Message 1457) Posted 22 Aug 2004 by old_user147 Post: Incredible, Carl! Good work! On my P4 2Ghz the sec/ts went down from 3.4 to 2.67. This is 27 % faster! I hope it calculates correct. What have you done with the compiler?

22) Message boards : Number crunching : Client performance (Message 1145) Posted 17 Aug 2004 by old_user147 Post: > About the benchmarks... > > I ran them 5 times over a few hours period on my laptop. The first time I ran > them they were really high (2190 float point, 4400 integer) for a p4 3Ghz > laptop, .5GB ram, XP Pro, the next time about half (1400 fp, 2800 i), next > (1100 fp, 2600 i), next about normal (1050 fp, 2500 i), and this is where they > stayed. The first was run on a cold (literally) computer. The only > difference between one and the next was time and temperature. Seems a little > odd to vary by such a huge amount. Makes you wonder about the accuracy of the > benchmark. > > Andrew > > If the temperature of the p4 gets too high (~70°C) "thermal throttling" is activated, which means that the processor gets slower. in this article it was tested: http://www.digit-life.com/articles2/p4-throttling/ the only thing that would help is better cooling.

23) Message boards : Number crunching : Visualisation application (Message 756) Posted 11 Aug 2004 by old_user147 Post: > I'm also getting ~50% CPU use with Linux visualisations. Thanks for providing > it in the first place, by the way. > > Specs: > Debian GNU/Linux 'sid' > XFree86 4.3.0 > Radeon 8500, proprietary ATI driver version 3.11.1 > Here the same problem (Debian Sarge, XF4.3, Intel 845G onboard graphic)

24) Questions and Answers : Unix/Linux : kernel panic - preempted kernel (Message 727) Posted 11 Aug 2004 by old_user147 Post: > But in the end, the kernel panic occured, while some other process (I don't know which process) wanted to preempt the model. Linux has preemtive multitasking, which means that the kernel shedules the cputime for every process. If time is over a process is preempted (if it is in user mode) and has to wait until the sheduler gives him the next time slice. The new feature "preemptive kernel" in the kernel 2.6 is that it is "allowed to preempt a process even if it is in kernel mode executing a system call". I would try to switch "preemptive kernel" off. If it still crashes, it must be a hardware error.

25) Questions and Answers : Windows : Bad Performance (Message 689) Posted 11 Aug 2004 by old_user147 Post: > > @carl: > > have you tried -tpp7, which imho uses sse2 on p4, athlon64...? > > or does the model not like this? > > I can't remember if I tried that, I may give it a go and see if it works OK. > > I once more read the ifort manual, now i think -tpp7 doesn't use see2, it only optimizes for p4s. to use sse2 you would need something like -axN

26) Questions and Answers : Windows : Bad Performance (Message 676) Posted 10 Aug 2004 by old_user147 Post: @carl: have you tried -tpp7, which imho uses sse2 on p4, athlon64...? or does the model not like this?

27) Message boards : Number crunching : Client performance (Message 583) Posted 9 Aug 2004 by old_user147 Post: Just ran the client on my slower computers for a few minutes, here the results: AMD K6-2+ 500 MHz: 22.5 sec/ts --> 6.8 month/model VIA C3 533 Mhz: 34.5 sec/ts --> 10.4 month/model Pentium MMX 200 MHz (100MHz FSB,RAM): 49 sec/ts --> 14.7 month/model The K6 isn't so bad, i think.

28) Questions and Answers : Wish list : Feature which checks, if the pc is calculating correctly? (Message 458) Posted 8 Aug 2004 by old_user147 Post: &gt; Unfortunately, that only partially solves one of the problems. Many people &gt; are running significantly overclocked and stable. Poorly maintained PCs, PCs &gt; with poor RAM, PCs with inadequate HeatSink/Fans or case cooling, PCs running &gt; CPUs in spec but trying to eek the most out of their RAM by decreasing &gt; timings, all can also go unstable. No way to test for those. &gt; &gt; This "control timestep" should be like a normal timestep. The only difference is that the variables (temp, pressure, ...) at the beginning of this step are always the same and the correct variables after this step are known. If the pc calculates something different, the pc is faulty. This may be due to cpu, ram or something else. <br> But, it's probably no easy to implement it into the client.

29) Questions and Answers : Wish list : Feature which checks, if the pc is calculating correctly? (Message 441) Posted 8 Aug 2004 by old_user147 Post: Would it be possible to add something to the client, which checks, if there is a hardware failure, like a to far overclocked cpu? <br>I think of running one \"control timestep\" every model-month or so. Then the calculated result of this timestemp is compared with the correct result. If the calculated differs from the correct one, the client gives a error message and stops working. That would probably help to avoid some short runs.

30) Message boards : Number crunching : Client performance (Message 297) Posted 7 Aug 2004 by old_user147 Post: My P4 (Northwood) 1.6 GHz @ 2 Ghz has 3.39 sec/ts. The cp-bonic-client is about 20% faster than the classic cp-client was.

Previous 20

©2024 climateprediction.net