CLC Efficiency Enigma

[RhoXS]

SETISpy defines efficiency as "CPU cycles / FLOP". I am running SETI CLC on two machines, both PIIIs.

One is a 600 MHz CPU running at 840 MHz (140 MHz FSB - ABIT P3V4X mobo - 128 Meg SDRAM).

The other is a new 1.2 Gig Taulatin running at 1.5 Gig (125 MHz FSB - Asus ST6 mobo - 512 Meg SDRAM).

The 840 Mhz machine runs with an average efficiency of 6.45 CPU cycles per FLOP but the 1.5 Gig machine averages a dismal 8.47. Both machines boot under XP Professional. The only software I believe that competes for CPU time is hmonitor. SETI Client Priority is set Low in both instances. Zonealarm is active on both machines but only the 1.5 Gig (the faster but less efficient machine) has Norton Antivirus running.

Why are the efficiencies so radically different? The 1.5 Gig machine is turning in WU at approximately 6 hrs/WU, even using the CLC. I am assuming this slow spped is a result of the lousy efficiency.

[13oots2]

First question, have you configured Seti Spy for your processor. If not you can find details how to Here (http://pages.tca.net/roelof/setispy/faq.htm#3.3) . Also you need to set your processor to RAM speed ratio in the Performance/Compare it's at the bottom. This should clear up any problems. If Norton Antivurus scan is running, turn off the Auto Protect Realtime scanning. You can still set it so E-mail and downloaded files are scanned. Virus scanners are a real hog of CPU time and disk access these will really affect Seti's performance.

[wolfsid]

well i don't know about U but i have 4 systems one a xp 1600 running at 2000+ 13X 133fsb and i have 3 windows open both running seti clc and i am doing them all in about 5 1/2 hrs all 3 setis are done with at least one so i think people who ever say that they can only run one seti on a system has to be wrong.. running one might improve the rate that one moves at but if U run 2 or more i think U get more work done in just about the same amount of tim..

[TC]

Is that 600 the older core with 512K cache? If so it will show better efficiency than a tualatin with 256K cache, even though it's full speed.

[wolfsid]

Originally posted by TC
Is that 600 the older core with 512K cache? If so it will show better efficiency than a tualatin with 256K cache, even though it's full speed.


what were U talking to me cuz i already said that is a xp 1600 running at 2000+ or 13x133 which is 1755 i think..

[Sir-Epix]

Originally posted by TC
Is that 600 the older core with 512K cache? If so it will show better efficiency than a tualatin with 256K cache, even though it's full speed.

No it is the Coppermine core, because the Katmai core would never reach those MHz. But you have the right idea. Just because it is not as effiecent doesn't mean that it is slower. What really matters are completion times :P

[TC]

Originally posted by Sir-Epix


No it is the Coppermine core, because the Katmai core would never reach those MHz. But you have the right idea. Just because it is not as effiecent doesn't mean that it is slower. What really matters are completion times :P Actually the Katmai core topped out at 600 with the 512K cache - check the Intel chart. I actually have one lying around here. Stepping kc0 s-code SL3JM. If his is indeed the last Katmai with larger cache that might explain the better efficiency.

[Sir-Epix]

Originally posted by TC
Actually the Katmai core topped out at 600 with the 512K cache - check the Intel chart. I actually have one lying around here. Stepping kc0 s-code SL3JM. If his is indeed the last Katmai with larger cache that might explain the better efficiency.

He also said it was running @ 140FSB. No way a Katmai core could do that. This is why I know it is a CuMine. I run on the Katmai core myself, and I get the better effieciency.

[TC]

Originally posted by Sir-Epix


He also said it was running @ 140FSB. No way a Katmai core could do that. This is why I know it is a CuMine. I run on the Katmai core myself, and I get the better effieciency. Well I ran the one I have up to 900 with a peltier, but I just noticed his motherboard is socket 370, so you're right it has to be a cu.

[RhoXS]

It is a CuMine SL44Y/1.65V core. Sorry I have been unable to clarify that sooner.

I will be replacing it sometime this comming week with a 700 MHz PIII that runs completely stable with a 142 MHz FSB (on the P3V4X). The 600 MHz chip, with Intel boxed cooler, is a slot one chip that fits OK in my current (Inwin) case. The 700 MHz chip is an FCPGA chip. Because of the required Slotket, the CPU cooler interfers with the power supply so it has to wait until I change cases.

I still use the screen saver to simply blank the screen. I am assuming that can have no adverse effect when running the CLC.

[TC]

The OS's power saving monitor shutoff won't bother the CLC. I'm puzzled now that you clarified the cpu in question. There shouldn't be a huge difference in efficiency between the coppermine and tualatin core when comparing same size L2 cache. The tualatin core has a few enhancements that should put it slightly ahead, but not by a wide margin. It may just be the memory bandwidth between the two systems you're comparing.

[wolfsid]

Originally posted by TC
The OS's power saving monitor shutoff won't bother the CLC. I'm puzzled now that you clarified the cpu in question. There shouldn't be a huge difference in efficiency between the coppermine and tualatin core when comparing same size L2 cache. The tualatin core has a few enhancements that should put it slightly ahead, but not by a wide margin. It may just be the memory bandwidth between the two systems you're comparing.

memory bandwidth is very important to overclocking mainly for seti or any crunching program U want very high memory bandwidth and if possible alot of ram it helps...

[RhoXS]

Maybe memory bandwidth is indded the key to the problem. The 600 MHz CPU is on an Asus P3V4X running with a 133 MHz FSB and 128 Meg SDRAM. The 1.2 Gig Tualatin is on a Abit ST6 running with a 125 MHz FSB and 512 Meg SDRAM. It is beggining to appear that the difference between a 125 MHz FSB and a 133 MHz FSB does indeed result in a noticeable effect.

I did notice that CPU cycles / FLOP did improve considerably on both machine when they were left unattended overnight. The 600 MHz CPU indicated about 5.6 and the Tualatin came in at just under 7. Still slowler but both much better than the numbers I saw yesterday with the only use primarily being just looking at SETISpy.

[lonewolf1983]

dont all the tulatins run on a 133mhz bus by default??

[wolfsid]

Originally posted by lonewolf1983
dont all the tulatins run on a 133mhz bus by default??


I Don't know about that i think some run on 100 and some run on 133 just like amd... i have 2 processors that are defult 100 which sucks but the multplier go higher then some of the 133 chips i have...

[RhoXS]

Intel is currently marketing a 1.2 Gig (12X/100 Mhz FSB) Tualatin. It is actually a full blown PIII (256K etc.) labeled as a Celeron. It is cheap, circa $100.00, and is easily overclocked. It will easily do 1.5 Gig (125 MHz FSB) if the core voltage is raised from 1.475 to 1.600 volts. Many have reported success at 1.6 Gig (133 MHz FSB) but that usually requires wire wrapping three pins to trick the mobo so the core voltage can be raised to 1.800 volts. Some have also removed the IHS (Integral Heat Sink) to achieve better heat transfer rates at this speed.

I was previously running a PIII at 998 MHz (142 MHz FSB). I was concerned that the increase in core speed might be offset by decreasing the FSB to 125 MHz. My concerns were misplaced. This cheap little wonder at 125 MHz blows the doors off my previous setup. Probably within the next month or so I intend to remove the IHS for better cooling and try to get to 133 MHz.

[lonewolf1983]

i thought u were running a Pentium 3 tulatin,not the celeron version
hence why i thought it was 100mhz default
:)