Mersenne Digest Wednesday, May 9 2001 Volume 01 : Number 847 ---------------------------------------------------------------------- Date: Sun, 6 May 2001 21:55:25 -0600 From: "Matt Goodrich" <[EMAIL PROTECTED]> Subject: RE: Mersenne: Slow CPU's in a Proliant 2500 Thanks for the response Ethan, it was quite helpful. Yes I did set the type and speed in Prime 95. I wish it was something obvious like that! The indexing service was set to manual and is now disabled. No change in times. I wasn't aware that running 2 instances on a dual machine would run 30% to 40% slower than a single instance (if I understand you correctly?) It looks like I may just have to live with it. I am loathe to change those chips when I am not sure it will help. Also, I'm not even sure I will be able to find them, even if I was inclined to change them. I will try your suggestion of running a double check on 1 CPU and LL test on the other, when these double checks are done. Thanks again, Matt - -----Original Message----- From: Ethan Hansen [mailto:[EMAIL PROTECTED]] Sent: Sunday, May 06, 2001 8:01 PM To: [EMAIL PROTECTED] Subject: RE: Mersenne: Slow CPU's in a Proliant 2500 Matt, Sorry for raising the obvious, but have you changed both the CPU type and speed values under the Prime95 Options->CPU menu? Another item to check are that you have disabled the blasted indexing service. It could explain the slight CPU time leakage you are seeing. That said, the ratio I typically see between Prime95 iteration times on a dual processor system with two instances vs. only one running is in the 1.3-1.4x range; i.e. the difference between your times of 0.448 and 0.325. I have seen a slight degradation of iteration time/MHz when upgrading from 500 to 1GHz processors, but not as much as you report. A way to boost overall throughput is to run double checking (with it's smaller memory footprint) on one CPU, and primality tests on the other. Regards, Ethan - -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of Matt Goodrich Sent: Sunday, May 06, 2001 5:26 PM To: [EMAIL PROTECTED] Subject: Mersenne: Slow CPU's in a Proliant 2500 I have just upgraded a Proliant 2500 from dual PPro 200's to dual Pentium II 333 overdrive processors. According to Intel's website this is supported. Also Compaq offers this upgrade. Now before I upgraded, I was running double check's on 2 exponents, 668xxxx. I was getting about .515 second iteration times. Now that I have upgraded, I am only getting .448 second iteration times. Yes, the affinity is set properly. [SNIP] Interesting note here. If I run just 1 copy of Prime on either CPU I get .325 second iteration times. It doesn't matter which copy of the program or which CPU I run it on. It is only when I run both copies that it slows down to .448 seconds. BTW I am running Windows 2000 Server with the latest SSD from Compaq. [SNIP] Any thoughts or suggestions would be appreciated. Matt _________________________________________________________________________ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers ------------------------------ Date: Mon, 07 May 2001 02:04:23 -0400 From: Sandy Harris <[EMAIL PROTECTED]> Subject: Re: Mersenne: Slow CPU's in a Proliant 2500 Matt Goodrich wrote: > > I have just upgraded a Proliant 2500 from dual PPro 200's to dual Pentium II > 333 overdrive processors. ... > Now before I upgraded, I was running double check's on 2 exponents, 668xxxx. > I was getting about .515 second iteration times. > Now that I have upgraded, I am only getting .448 second iteration times. Might it just be the effect of slower cache? As I recall, the numbers were: P Pro 256 or 512K one CPU clock to deliver data P II 512 K two Celeron 128 K one and on some tasks, Celeron outperforms P II at the same clock because of this. If the process is cache-bound and key data fits in both caches, this gives about the right numbers. P II cache runs at 333/2 which is to 200 roughly as .448 is to .515. Alternately. do you just need a differently optimised version of the code to get the best out of your P IIs? _________________________________________________________________________ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers ------------------------------ Date: Mon, 7 May 2001 07:55:56 +0200 From: "tom ehlert" <[EMAIL PROTECTED]> Subject: Mersenne: Slow CPU's in a Proliant 2500 Hi Matt, > Now that I have upgraded, I am only getting .448 second iteration times. > If I run just 1 copy of Prime on either CPU I get .325 second iteration As I have the same problem, some hints: I have a dual P700 (MSI 694D motherboard, PC133 RAM). running Prime twice, I get .340 (at 12xxxxxx), running only one copy, I get 270. If the second Prime is set for factoring, the first copy still tests at 270. My conclusion is, that it's not a Win2K or something else bug, its rather the limited memory througput, which prevents 2 copies two do twice the work, that on CPU would do alone. tom _________________________________________________________________________ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers ------------------------------ Date: Mon, 7 May 2001 00:15:52 -0700 From: "Aaron Blosser" <[EMAIL PROTECTED]> Subject: Re: Mersenne: Slow CPU's in a Proliant 2500 Besides other nefarious things I'm known for, I am also a Compaq ASE, so hopefully that'll carry some weight in my response. :) > > I have just upgraded a Proliant 2500 from dual PPro 200's to dual Pentium II > > 333 overdrive processors. ... > > Now before I upgraded, I was running double check's on 2 exponents, 668xxxx. > > I was getting about .515 second iteration times. > > Now that I have upgraded, I am only getting .448 second iteration times. > > Might it just be the effect of slower cache? As I recall, the numbers were: > > P Pro 256 or 512K one CPU clock to deliver data > P II 512 K two > Celeron 128 K one > > and on some tasks, Celeron outperforms P II at the same clock because of > this. One problem is most likely due to the Pentium II Overdrive's cache architecture which, if I recall correctly (I hate abbreviations), runs at half-speed like the other PII's. The PPro cache ran at full speed, and the Proliant 2500 (and I've got a couple of the 2500's around) came with 512MB L2 cache chips, but could be ordered with the 1MB or 2MB L2 PPro's also... believe me, even on a 200 MHz machine, that extra cache, and running at full speed, does make a difference. Sure, a PII running at 333 would be faster, but you have to deduct "speed points" for the smaller cache (256K I think) running at half the full speed (166 MHz, even less than the cache on a PPro 200)... Also, please DO NOT DO NOT DO NOT DO NOT replace the BIOS chips on your 2500 with the E50 BIOS. There are a LOT more differences in the systemboards for a Proliant 2500 E24 (the PPro system board) and the E50 board (the PII board). Not only would those BIOS chips not work, but they would *really* not work (I'm not sure what that means...) The odd performance stemming from running NTPrime on both CPU's is a direct result of the memory architecture not being optimized for PII accesses and speeds. The E24 board was made to work great at handling mem requests for a pair of PPro's. Compaq does a GREAT job at SMP architecture, even on their older boards, using their custom chipsets and all that (much better than off-the-shelf dual CPU boards). However, when you go beyond what the motherboard was designed for, the peak performance of the mem architecture goes away. I would say that the BEST option you could come up with would be to forget the PII Overdrive chips (they're almost worthless, in my opinion, performing worse than a regular PII at the same speed... go figure. It shouldn't be that way, but that's my experience... The thing to do is to get the PPro-Celeron socket adapter from Powerleap (www.powerleap.com). They're compatible with the Proliant 2500 (and a few other dual PPro machines)... One caveat... you can get 2 upgrades for your 2500, but due to some physical constraints, some people have been forced to shave off a few millimeters of PCB to get both CPU's working in the 2500... but it can be done. Another caveat... only the older Celeron 550's are dual CPU capable... the newer ones are now missing that capability altogether. Powerleap sells kits of 2 Celeron 550's that are SMP capable, so that's one way to go. Another way to go is to forego the dual CPU and just get the fastest single Celeron they support, which is a Celeron 766MHz (the fastest 66MHz Celeron there is). Given the trade-offs of running dual 550's or a single 766, you might get better performance with the single, faster CPU than with 2 slower ones... only some benchmarks of either would really tell you though. Hmmm... I'm currently debating which way to go on that for the 2 Proliant 2500's I currently run. I'm leaning towards the single faster option for my own machine, which doesn't really need dual CPU's for much, and getting the dual 550's for the server at work since it does more things. Hey, that way I can tell you all which works better at NTPrime. :) To sum up: ... E24 boards don't work with E50 BIOS (there's actual physical differences in the board, which should be obvious since one is socket 8 and one is Slot 1. :) ... PL-Pro/II upgrades using Celeron (cache runs at full speed!) are a MUCH MUCH better option than the cheesy PII Overdrives (if you can even find one of those dinosaurs) ... Besides NTPrime, judge whether dual Celeron's at 550 would be better or worse than a single Celeron at 766. Aaron _________________________________________________________________________ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers ------------------------------ Date: Mon, 7 May 2001 11:49:52 -0000 From: "Brian J. Beesley" <[EMAIL PROTECTED]> Subject: Re: Mersenne: Slow CPU's in a Proliant 2500 On 7 May 2001, at 2:04, Sandy Harris wrote: > Might it just be the effect of slower cache? As I recall, the numbers were: > > P Pro 256 or 512K one CPU clock to deliver data > P II 512 K two > Celeron 128 K one > > and on some tasks, Celeron outperforms P II at the same clock because of > this. Though when I investigated why a C266 system I had outperformed a PII- 266 running Prime95, I found that the reason was that the LX chipset on the PII was assuming that the memory was very slow, whereas the EX chipset on the C266 system was taking advantage of the memory timing information set in the PROM on the SDRAM boardlet. > > If the process is cache-bound and key data fits in both caches, this gives > about the right numbers. P II cache runs at 333/2 which is to 200 roughly > as .448 is to .515. Key data does NOT fit in the caches. In fact if you look at the timings page you will see that where the same processor is available in "slow" 512K L2 cache & "fast" 256K L2 cache formats (e.g. PIII vs PIIIE, or Slot A Athlon vs Socket A Athlon) the "slow" 512K cache version usually runs Prime95 a few percent faster. The key is the memory bus loading and the efficiency of access to the memory. > > Alternately. do you just need a differently optimised version of the code > to get the best out of your P IIs? No, the PPro and the PII are the same hardware architecture. Regards Brian Beesley 1775*2^332181+1 is prime! (100000 digits) Discovered 22-Apr-2001 _________________________________________________________________________ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers ------------------------------ Date: Mon, 07 May 2001 10:09:49 -0400 From: "Brian Last-Name" <[EMAIL PROTECTED]> Subject: Re: Mersenne: Slow CPU's in a Proliant 2500 >I have set the switches on the I/O board for 66/333 speed. > That is probably your problem. Your bus is only 66 mhz. You are not able to access your memory as fast as you want to process the data. I once upgraded a 486 100Mhz to a 486 150Mhz (I think, it was a while ago) and there was almost no change in the speed. I overclocked the memory bus from 66 to 75, and lowered the multiplier (so the CPU was at the right speed) and gained more of an improvement. You may have to buy new, faster memory to make this possible, but the stuff is CHEAP right now. (just bought 128MB PC-133 Cas 2 for $30!) Good luck! Brian Peltzer [EMAIL PROTECTED] _________________________________________________________________ Get your FREE download of MSN Explorer at http://explorer.msn.com _________________________________________________________________________ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers ------------------------------ Date: Mon, 7 May 2001 09:09:29 -0700 From: "John R Pierce" <[EMAIL PROTECTED]> Subject: Re: Mersenne: Slow CPU's in a Proliant 2500 > That is probably your problem. Your bus is only 66 mhz. You are not able > to access your memory as fast as you want to process the data. I once > upgraded a 486 100Mhz to a 486 150Mhz (I think, it was a while ago) and > there was almost no change in the speed. I overclocked the memory bus from > 66 to 75, and lowered the multiplier (so the CPU was at the right speed) and > gained more of an improvement. You may have to buy new, faster memory to > make this possible, but the stuff is CHEAP right now. (just bought 128MB > PC-133 Cas 2 for $30!) Good luck! the PPro chips only supported 66Mhz FSB operations, so thats going to be all the motherboards support. _________________________________________________________________________ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers ------------------------------ Date: Mon, 7 May 2001 10:36:53 -0700 From: "Pardoe, Richard (PRDR)" <[EMAIL PROTECTED]> Subject: Mersenne: OT: Distributed Computing: YETI@home Appears that SETI@home has some new competition - YETI@home. Details (and perhaps a few chuckles) to be found at: http://www.phobe.com/yeti/index.html On a bit more serious note - humour does often try to capture what people feel uncomfortable about. I think the comments about "aggressive algorithm" that takes control of the computer probably echo the reservations many have towards participating in distributed computing. Rich Pardoe _________________________________________________________________________ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers ------------------------------ Date: Mon, 7 May 2001 22:02:05 +0000 (GMT) From: Russel Brooks <[EMAIL PROTECTED]> Subject: Mersenne: W2K screen saver vs. Prime95 I just up(?)graded from NT4 to Win2000. In the old NT4 screen saver selections was a "Blank Screen" option that I used. It would blank the screen when it locked my keyboard but it didn't run any active screen program that would take cycles from Prime95. Under W2K I don't have any Blank Screen option. If I want the screen saver active to lock the unattended keyboard I have to pick one of their 'active' display selections. Any ideas on how I can get automatic keyboad lockup with a blank (or cleared) screen AND leave the most cycles for Prime95? Cheers... Russ _________________________________________________________________________ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers ------------------------------ Date: Mon, 7 May 2001 16:46:06 -0700 From: "Aaron Blosser" <[EMAIL PROTECTED]> Subject: Re: Mersenne: W2K screen saver vs. Prime95 > I just up(?)graded from NT4 to Win2000. In the old NT4 screen > saver selections was a "Blank Screen" option that I used. It > would blank the screen when it locked my keyboard but it didn't > run any active screen program that would take cycles from > Prime95. Under W2K I don't have any Blank Screen option. If I > want the screen saver active to lock the unattended keyboard I > have to pick one of their 'active' display selections. > > Any ideas on how I can get automatic keyboad lockup with a blank > (or cleared) screen AND leave the most cycles for Prime95? You should see a screen saver called 'Default Screen Saver' That's what you want... just a blank screen. Note: If you have power saving on your monitor, once your monitor turns off, any running screen saver is "supposed to" stop running. Whether that's the case or not for many screen savers, I have no idea. Besides the blank screen, another good, low impact saver is the "logon" one. Aaron _________________________________________________________________________ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers ------------------------------ Date: Mon, 7 May 2001 18:06:24 -0600 From: "Matt Goodrich" <[EMAIL PROTECTED]> Subject: RE: Mersenne: Slow CPU's in a Proliant 2500 I just wanted to thank everyone for all the great responses I got. I have decided to just leave it alone. Perhaps at a later time I will look into putting a Celeron in here as Aaron Blosser suggested. Again, thanks to everyone! You were all very helpful. Matt _________________________________________________________________________ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers ------------------------------ Date: Tue, 8 May 2001 05:21:28 -0500 From: "Steve" <[EMAIL PROTECTED]> Subject: Re: Mersenne: W2K screen saver vs. Prime95 >From my experience with dozens of cases on all flavors of Windows, the power save mode has no effect on screensavers. They will continue running with the monitor off. The ones with sound effects (such as "underwater") continue producing their noises. If anyone has any ideas on how to make it otherwise, I would love to hear them. Steve Harris > >Note: If you have power saving on your monitor, once your monitor turns off, >any running screen saver is "supposed to" stop running. Whether that's the >case or not for many screen savers, I have no idea. > >Besides the blank screen, another good, low impact saver is the "logon" one. > >Aaron > >_________________________________________________________________________ >Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm >Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers _________________________________________________________________________ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers ------------------------------ Date: Tue, 8 May 2001 21:28:00 +0000 (GMT) From: Russel Brooks <[EMAIL PROTECTED]> Subject: RE: Mersenne: W2K screen saver vs. Prime95 > On my Win2K boxes here the "default" screen saver is a blank screen. It > doesn't look that way when you select it, but it is. I just tested it and it > didn't steal a single cpu cycle from Prime95. Yeah, it doesn't look like a blank screen so I didn't try it. Thanks for the pointers to the "Default" screen saver, I'm now using it so Prime95 gets the max cycles when I'm away. Cheers... Russ _________________________________________________________________________ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers ------------------------------ Date: Wed, 09 May 2001 16:47:13 -0400 From: "Joshua Zelinsky" <[EMAIL PROTECTED]> Subject: Mersenne: Really minor suggestion. Not sure if this is even worth it to bother writing. It wouldn't be dificult to have Prime95 accept non-integer quantities for the estimated amount of time on. Why not do this? Granted, the benefit, in terms of accuracy of completion dates will be small, but for computers that are only on rarely (under about 3 or 4 hours) it would really matter. Any thoughts. Regards, Joshua Zelinsky [EMAIL PROTECTED] _________________________________________________________________ Get your FREE download of MSN Explorer at http://explorer.msn.com _________________________________________________________________________ Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers ------------------------------ End of Mersenne Digest V1 #847 ******************************
