Pentium M And Wu 1481

[Sqopt]

The project: p1481_tet_2nm

 

Pentium M 1200, 1MB Cache: 190 ppd/GHz

Pentium M 1895, 2MB Cache: 226 ppd/GHz

A64@1800 Mhz, 512k Cache: 183 ppd/GHz

A64@2400 MHz, 512k Cache: 166 ppd/GHz (overclocked)

 

 

1.) P4M owns on this WU.

 

2.) Is the behavior I see on the A64 (declining ppd/GHz with increasing OC) atypical? If it is typical, why?

 

Sqopt

[bigred]

1: it's not a Pentium 4 M. it's a Pentium M. there is a HUGE difference between the two chips

 

2: sometimes you don't always get the best choice of WU's. Many of my folders are FX-55's or better. some of which are OC'd to near double the speed of your athlon 64. some days they tear up the WU's and others thy don't do too well

[Sqopt]

1:  it's not a Pentium 4 M.  it's a Pentium M.  there is a HUGE difference between the two chips

 

2:  sometimes you don't always get the best choice of WU's.  Many of my folders are FX-55's or better,  some of which are OC'd to near double the speed of your athlon 64.  some days they tear up the WU's and others they don't do too well

584912[/snapback]

 

Ah, I stand terminologically corrected on the Pentium M stuff. Dang chip nomenclature... Oh, and one more data point: the 1.895 GHz Pentium M also soundly trounces my Prescott 630 @ 3GHz/2MB Cache (same wu!) I just don't have the number at hand.

 

As to #2, does that still hold true where, as in my cited examples, one considers the <i>same</i> WU?

 

I would have thought that, within reason, one would expect more or less a linear decrease in processing time with increase in CPU frequency; and thus the ppd/GHz should also remain roughly constant. Yet, in the case of the 1481 WU and my (dedicated-to-folding) Athlon system, that assumption seems not to hold particularly well.

 

I'm not really sure, when all's said and done, that this is "trouble." So I'm not looking for "troubleshooting." Still, one could make a colorable argument that there's an inefficiency apparent in my benchmarks. I'd love to understand the basis of that inefficiency. Any ideas?

 

 

Sqopt

[hardnrg]

you cannot compare the GHz between A64 and P-M... the P-M has more effective power per clock cycle...

 

it would be like saying the A64 owns for folding compared to a P4, basing your argument on clock speed... meaning, you are stating the obvious... sorry, but that's what you are doing

[bigred]

technially speaking no two work units are identical. if they were what would the point of folding all those seperate ones be? if WU A is the same as WU B and we already know what A is, then why calulate B? see waht I'm saying here.

[BabyBalrog]

A) The P-M architecture is very efficent and it's large, low latency cache is perfect for certain WU's that fit within it, but on large ones they suffer from the mempry interface.

 

 

B) That is an atypical result. When i OC my 3000+ to 2.7 my PPD doubles. So I"m guessing that the OC isn't stable adn you are having to either A. Repeate steps. or B) turn in incomplete WU's so try lowering it and seeing if that improves thigns

[hardnrg]

some WU's (usually some of the 600 pointers or 354 or so pointers) give me ~400ppd... some of the nasty little bastard WUs give me like 80ppd... :|

[swifty11212]

some WU work best under certain procs.

[ccokeman]

some WU's (usually some of the 600 pointers or 354 or so pointers) give me ~400ppd... some of the nasty little bastard WUs give me like 80ppd... :|

585048[/snapback]

Man am i feeling that one right now. Revenge of the 46 point wu's. It just depend on the wu and the type of processor. On some of the 1140 and 1150 series my A64 runs about 5 minutes a step faster than the p-4 3.0. Just depends.

[Sqopt]

A) The P-M architecture is very efficent and it's large, low latency cache is perfect for certain WU's that fit within it, but on large ones they suffer from the mempry interface.

B) That is an atypical result.  When i OC my 3000+ to 2.7 my PPD doubles. So I"m guessing  that the OC isn't stable adn you are having to either A. Repeate steps. or B) turn in incomplete WU's  so try lowering it  and seeing if that improves thigns

585009[/snapback]

 

Ok, now I'm really feeling overclock envy. I have (on air cooling) not been able to get past 2.4 GHz on my A64 3000+. A nifty 2.7, I'd say. I'm both mprime and fah stable at 2.4, though, so I don't think I'm repeating WUs or having them fail.

 

Question, though: assuming your A64 3000+ has a stock clock of 1.8 GHz like my Winchester does, how could going to 2.7 GHz, which is an overclock of 50%, double your PPD? In my experience, FAH WUs, kernel compiles, and mprime benchmarks all scale in about a 1:1 proportion with CPU speed. Are you getting some disproportionate speedup by way of a tricky guru maneuver?

 

 

Sqopt

[BabyBalrog]

Well some of it has to do with being faster CPU, but also memory, and just WU luck. Things that fit in my cache etc. I haven't tweeked this CPU out quite yet. or i mean windows. but I'm trying.. need water cooling

[apostolics]

every chip is designed to work at its mhz speed they do not compare. my pentium -m 2.13 stomps but my amd 3000+ @ 2.2 isnt as fast. remember that ram is also a factor my 2.13 has 2gb of ddr2 ram compared to my 3000+ that only has 512 also remember the cache size