Benching SuperPi tweaks @32M

[Jochenp]

I'm going to bench a lot of those so called 'tweaks' and see if they actually make a difference in SuperPi 32M
I'm gonna use XPLite, X-Setup 7.2 and some tweaks in the sticky.
Before I start, I wanna know if anyone has any tweaks they wanna see benched @ 32M?
BTW, I'm using a fresh install of Windows XP Pro SP2 for this.
I already disabled unnecessairy services, so we won't see the effect of that, I'm sorry :(


Ok, first run, this is without tweaks (except for disabled services)

http://img117.imageshack.us/img117/7829/notweaks7fe.th.jpg (http://img117.imageshack.us/my.php?image=notweaks7fe.jpg)

XPLite did a decent job, it shaved 11 seconds (don't know if this within the margin of error) and it freed >200MB harddisk space, and that is the main purpose of this program.

http://img117.imageshack.us/img117/930/naxplite4rf.th.jpg (http://img117.imageshack.us/my.php?image=naxplite4rf.jpg)

Another update, testing the DisablePagingExecutive=1 tweak, of which all tweak experts say that it doesn't make any difference. Because of that, I reran this test 3x and I reran it 3x without the tweak, to make sure my results weren't an error or something screwing up. It shaved 2 seconds off. Quite funny, if I disable it, I always get 29min 51sec, if I enable it I always get 29min 49sec. It seems SuperPi 32M doens't have a margin of error and always shows the same results for the same settings :) (I like that, it makes my benches more accurate!)

http://img490.imageshack.us/img490/6486/disablepagingexecutive19om.th.jpg (http://img490.imageshack.us/my.php?image=disablepagingexecutive19om.jpg)

Going to test with a registry cleaner, it won't do too much right now, since this is a fresh install, but we'll see...
Using this one http://www.majorgeeks.com/download460.html
EDIT: RegCleaner is a POS, it added 5 seconds :(
http://img493.imageshack.us/img493/21/afterregcleaner1oa.th.jpg (http://img493.imageshack.us/my.php?image=afterregcleaner1oa.jpg)
Pretty weard huh, deleting obsolete values in a registry makes it slower, that doenst make sence :confused:
Maybe some other stuff slipped in by installing it...
Well, all I can say is, don't use RegCleaner, because once you deleted those values and you (like it did, stupidstupid me :slap: ) deleted it's backup files, there is no way of adding them again.


The LargeSystemCache=1 did a wonderfull job :clap:
It shaved 1min and 27sec off, but yesterday, before I undid this tweak, and redid it, I shaved even 10 seconds more:( Too bad, but still, this is, untill now, the best tweak you can do!
http://img210.imageshack.us/img210/7125/nalargesystemcache10qx.th.jpg (http://img210.imageshack.us/my.php?image=nalargesystemcache10qx.jpg)


JochenP

[Jochenp]

*Reserved for future posting*
You never know how fast I'll exceed the image limit
(and no, this isn't spam posting for a higher post count)

[Jochenp]

*Reserved for future posting*
You never know how fast I'll exceed the image limit
(and no, this isn't spam posting for a higher post count)

[Jochenp]

Sorry lads, I'm gonna redo the benches , my system wasn't fully overclocked, and I want to make these benches as accurate as possible.

[[XC] moddolicous]

Your also gonna want to include a cpuz shot and an A64 tweaker shot to show us that you didnt cheat (I beleive you, but others might just say you raised clocks). Also, why not make the pics like thumbnails instead so that modem people can see this also. I cant wait to see the results.

[Jochenp]

K, thnx for the feed-back.
For more than 2 years now, I wanted to see an actual benchmark-comparison between all of these tweaks.
Since I never saw one, I decided to make one myself.

[Jochenp]

DisablePagingExecutive=1 update

[Jochenp]

Very strange indeed, comparing my screenshots, I see that after every tweak, my 'available real memory' decreases, and I rebooted everytime, so I don't know how this happpened :confused:
It went from 802893824 to 725753856 (comparing first to last screenshot).
Does anyone have an explanation for this strange behaviour?

[SquireSCA]

I have to ask the question... Overclocking and tweaking the hardware to get better scores makes sense. I have been overclocking since 1997, so I get that.

But tweaking the OS and the test itself, does that make sense?

What I mean, is that tweaking my bike by altering the gearing, changing the exhaust and remapping the FI system to take my 10.74s 1/4 mile down a tad, is like overclocking. But I am not altering the track. I still have to run the same 1/4 mile track as the other guys, at the same altitude, same temperature and relative humidity.

The track does not change, and it is therefore the control factor, the constant that tells us who's bike is fastest.

If I change the track, or test/OS in order to improve my score, doesn't that make the score kinda irrelevant? If I modified the track and beat the other bike, unless he is on that very same track with the same track "tweaks" that I had, any comparison between our results is kinda meaningless.

I was just curious what other people thought if this. I have always been for benchmarking, but not when it involves tweaking the app or the OS. That means that it is not an apples to apples comparison, and so my score of 27 seconds means nothing compared to your score of 32 seconds in SuperPI.

I see a lot of people tweaking systems and their OS to the max to get a better score, but doesn't that process make many of those score irrelevant?

It is almost like saying that your bike ran a 1/8 mile faster than mine did a 1/4. Does that make your bike faster?

Thoughts?

[massman]

you don't change the track, you change your tyres, brakes and so on. Changing the track would be changing your benchprogram.

tweaking is just a part of overclocking :)

[SquireSCA]

you don't change the track, you change your tyres, brakes and so on. Changing the track would be changing your benchprogram.

tweaking is just a part of overclocking :)

No, I think that when you alter the OS, you are not altering the car, but rather the environment that it runs under.

So let's say that the track(benchmark) is the same, but you lower gravity(OS/Environment)... When you compare to someone else running normal gravity... is the comparison valid?

I am talking more philosophically here. For any comparison to have merit, you need a control. You need to run the identical test on two different hardware configs. When you change the OS, you alter the test itself in a way. Now it it no longer about hardware, but who can alter the OS more than the other guy.

I see threads where guys have 55mb WinXP installs with all sorts of stuff ripped out in order to get a higher number, and all I am saying, is does that number actually mean anything when you have to go to such lengths to get it?

I always ran benchmarks on an real everyday system. Tweak the hardware out to the max, but if the system is not stable and it is not the settings that I use on a daily basis, then I never considered my scores valid.

Like guys posting benchmark scores when the 3Dmark graphics are corrupted with artifacts. You would never game that way, so why would such a score even be considered valid?

This is not intended to start a huge debate, I was just curious about the mindset that many people have towards getting higher benchmark scores. I wondered if people realized that in the pursuit of getting a higher number, that they might render that number meaningless.

[Jochenp]

No, I think that when you alter the OS, you are not altering the car, but rather the environment that it runs under.

So let's say that the track(benchmark) is the same, but you lower gravity(OS/Environment)... When you compare to someone else running normal gravity... is the comparison valid?

I am talking more philosophically here. For any comparison to have merit, you need a control. You need to run the identical test on two different hardware configs. When you change the OS, you alter the test itself in a way. Now it it no longer about hardware, but who can alter the OS more than the other guy.

I see threads where guys have 55mb WinXP installs with all sorts of stuff ripped out in order to get a higher number, and all I am saying, is does that number actually mean anything when you have to go to such lengths to get it?

I always ran benchmarks on an real everyday system. Tweak the hardware out to the max, but if the system is not stable and it is not the settings that I use on a daily basis, then I never considered my scores valid.

Like guys posting benchmark scores when the 3Dmark graphics are corrupted with artifacts. You would never game that way, so why would such a score even be considered valid?

This is not intended to start a huge debate, I was just curious about the mindset that many people have towards getting higher benchmark scores. I wondered if people realized that in the pursuit of getting a higher number, that they might render that number meaningless.
Just keep in mind, that if tweaks increase your benchmark number, they'll probably increase you reallife performance too

[SquireSCA]

Just keep in mind, that if tweaks increase your benchmark number, they'll probably increase you reallife performance too

Well. that's debatable. Many times one person or reviewer claims that one PC is faster than the other, when the real truth is that there is no real-world difference between them... It takes some synthetic benchmark to spit our a number to let us know that there is a 2% difference.

I even stopped overclocking as much as I used to, because the real world benefits hit the wall of diminishing returns. Fast parts are so cheap these days, that the idea of adding heat and possible instability just to get a higher arbitrary score in a given benchmark no longer seemed worth my time.

In the old days, it could mean the difference between a game being playable or choppy. Now it is the difference between 98fps and 110fps. No difference at all. And as for tweaking out a CPU to calculate PI, well, I am a gamer, so the video card makes the biggest difference. Going from a 2Ghz CPU to a 4Ghz CPU really would not make a difference in gaming performance. Not really. I do not play at a resolution that would have me CPU limited. Might make a couple fps difference, but without some benchmark telling me there is a difference, I would never know.

Possibly it is just that it got boring. Overclocking used to be a challenge and it was good to do it because not everyone could afford the $1100 top Intel CPU. Now you can get a dual-core 3Ghz chip for a couple hundred bucks. Sorta takes the fun, and need out of it.

[Jochenp]

This freaking regcleaner is a POS, it raised my score by 5 seconds, it should burn in hell for that, now let's test LargeSystemCache=1, which seems promising.

[Jochenp]

Well. that's debatable. Many times one person or reviewer claims that one PC is faster than the other, when the real truth is that there is no real-world difference between them... It takes some synthetic benchmark to spit our a number to let us know that there is a 2% difference.

I even stopped overclocking as much as I used to, because the real world benefits hit the wall of diminishing returns. Fast parts are so cheap these days, that the idea of adding heat and possible instability just to get a higher arbitrary score in a given benchmark no longer seemed worth my time.

In the old days, it could mean the difference between a game being playable or choppy. Now it is the difference between 98fps and 110fps. No difference at all. And as for tweaking out a CPU to calculate PI, well, I am a gamer, so the video card makes the biggest difference. Going from a 2Ghz CPU to a 4Ghz CPU really would not make a difference in gaming performance. Not really. I do not play at a resolution that would have me CPU limited. Might make a couple fps difference, but without some benchmark telling me there is a difference, I would never know.

Possibly it is just that it got boring. Overclocking used to be a challenge and it was good to do it because not everyone could afford the $1100 top Intel CPU. Now you can get a dual-core 3Ghz chip for a couple hundred bucks. Sorta takes the fun, and need out of it.
In my case, overclocking is almost necessairy to play the newest games at decent settings. It makes the difference between running CS:S @1024*768 2AA 2AF No HDR @ 80FPS stresstest or @1024*768 6AA 6AF Full HDR @ 95FPS stresstest, so it makes it worthwile for me. Normally SuperPi doens't interest me a lot, to be honest, this thread was the first time I ran SuperPi 32M.
But I just bugs mee that people who are 'tweaking' their OS just apply tweaks explained on site's without even caring to see the actual numbers behind it. The only way to proove that one settings is faster than another is running a benchmark. Since 3DMark01 is the only benchmark where CPU and Memory performance make a big difference, I first thought running this as comparison. But 3DMark01 has such a big margin of error, that my results would be very inaccurate. So then SuperPi popped in my mind. If these tweaks do anything, they should at least show some improvement in calculating Pi to fill 32MB, now shouldn't they? And SuperPi almost hasn't got a margin of error, so it seemed perfect to me for this job.
After all these 'tweaks' I will see if they make any difference in CS:S stresstest, and if they don't I'll just undo em all with system restore and be happy with the thought that my OS still has all funtions a non-overclocked system would have...

JochenP

[K.I.T.T.]

How do you set this largesystemcache=1 and disablepagingexecutive = 1 ?

:confused:

[Jochenp]

Google est ton ami!
It's a basic tweak so I didn't think I needed to put this in my thread, but I will if a lot of people don't know how to do em.

[metro.cl]

i wanna know more about largesystem cache, is it recomended for 24/7 use??

where is it on the registry??

you could add a task manager shot to see the disbled process


p.d also update your amd64 tweaker its on v0.6 (http://home.exetel.com.au/codered/A64Tweaker/A64Tweaker_V0.6beta.zip) now

also check this link we have a small guide to optimize super pi chw super pi guide (http://www.chilehardware.cl/index.php?name=PNphpBB2&file=viewtopic&t=6137)


hope this helps a bit


LargeSystemCache is the one of the two registry settings manipulated by the Control Panel dialog not available in NT Workstation. If you turn on LargeSystemCache=1, assuming you have 96MB+ RAM, then you should see a significant performance boost for CPU and AV intensive applications but little or no improvement for I/O bound applications. I would use 128MB and 256MB as minimums for Windows 2000 Pro and Windows XP Pro, respectively.

Hive: HKEY_LOCAL_MACHINE\System\CurrentControlSet\Contro l\Session Manager
Key: Memory Management
Name: LargeSystemCache
Type: REG_DWORD
Value: 0 or 1

[Jochenp]

Thnx for the info, but could you translate some of that in english for me please?
I know there is a version 0.60, but i like this one better (more stable if ou ask me)

[Shinuza]

Dis you use Ramdisk? Writing on the ram instead of the HDD will help you for sure :)

[massman]

Well. that's debatable. Many times one person or reviewer claims that one PC is faster than the other, when the real truth is that there is no real-world difference between them... It takes some synthetic benchmark to spit our a number to let us know that there is a 2% difference.

I even stopped overclocking as much as I used to, because the real world benefits hit the wall of diminishing returns. Fast parts are so cheap these days, that the idea of adding heat and possible instability just to get a higher arbitrary score in a given benchmark no longer seemed worth my time.

In the old days, it could mean the difference between a game being playable or choppy. Now it is the difference between 98fps and 110fps. No difference at all. And as for tweaking out a CPU to calculate PI, well, I am a gamer, so the video card makes the biggest difference. Going from a 2Ghz CPU to a 4Ghz CPU really would not make a difference in gaming performance. Not really. I do not play at a resolution that would have me CPU limited. Might make a couple fps difference, but without some benchmark telling me there is a difference, I would never know.

Possibly it is just that it got boring. Overclocking used to be a challenge and it was good to do it because not everyone could afford the $1100 top Intel CPU. Now you can get a dual-core 3Ghz chip for a couple hundred bucks. Sorta takes the fun, and need out of it.

to be honest, I just overclock to benchmark a few times and then set everything stock, because I don't game very much, so I don't need a fully overclocked pc. But if I bench, it's pure a hobby, so I want to be as good as possible :)

[SquireSCA]

to be honest, I just overclock to benchmark a few times and then set everything stock, because I don't game very much, so I don't need a fully overclocked pc. But if I bench, it's pure a hobby, so I want to be as good as possible :)

Well, that's what I was trying to ascertain. I see some pretty exotic tweaks and lenghts that people go to just to get a couple artificial points in some obscure benchmark, and wondered why. I will tweak and overclock if needed in order to get better performance for some real world task that I want to use my PC for, but to get liquid nitrogen cooling, running the system outside of a case and running a hacked and modded OS, just to get a higher number on what amounts to a long division equation... well, it just seems strange to me. hehehe

[metro.cl]

Dis you use Ramdisk? Writing on the ram instead of the HDD will help you for sure :)


thats the idea.

for the translation maybe you could use google language tools :) then i can help you with the ones that you still dont understand

[coop]

@squireSCA, the Pi program is the track. "The track does not change, and it is therefore the control factor, the constant that tells us who's bike is fastest." Tweaking the hardware and software it runs on is like the areodynamics of your helmet, modding the bike and being able to cut a good light. *****Well, that's what I was trying to ascertain. I see some pretty exotic tweaks and lenghts that people go to just to get a couple artificial points in some obscure benchmark, and wondered why. I will tweak and overclock if needed in order to get better performance for some real world task that I want to use my PC for, but to get liquid nitrogen cooling, running the system outside of a case and running a hacked and modded OS, just to get a higher number on what amounts to a long division equation... well, it just seems strange to me. hehehe *** It is just something to do. Going 130+ mph in a quarter mi. might seem strange to some as well. :D

[SquireSCA]

Going 130+ mph in a quarter mi. might seem strange to some as well. :D

No way man, that rocks. Besides, doing a wicked fast 1/4 mile on a bike is more likely to get you laid than breaking 25s in SuperPI. ;-)

[M.Beier]

SquireSCA, sure you've not gone to the wrong homepage?

Its the second thread tonight I've seen you asking the question "why do you bother" in..
Xtremesystems.org is about hardware, and how to make it better & faster.... - Not about how the odds of getting laid are, or how a benchmark program affects your game, or how great a wast of time it is to bench...

EDIT: My point is, your preaching to the wrong quier (bad gramma, I know..)

[SquireSCA]

SquireSCA, sure you've not gone to the wrong homepage?

Its the second thread tonight I've seen you asking the question "why do you bother" in..
Xtremesystems.org is about hardware, and how to make it better & faster.... - Not about how the odds of getting laid are, or how a benchmark program affects your game, or how great a wast of time it is to bench...

EDIT: My point is, your preaching to the wrong quier (bad gramma, I know..)

No, no, you misunderstand.

I am all for more performance. I am questioning some of the particular methodology. I wa hoping that perhaps someone would give a good example of why a higher value in SuperPI would equate to a measurable and tangible benefit in something else, with specifics.

Or, is it just a simple and easy benchmark to run for competition's sake?

That's why I am asking. If someone can show me where it could actually help me, I would dive right in.

edit: And the bike thing, actually will get you laid. :-) x2

[M.Beier]

Well, some times a small research'll do... But, the short version, SuperPI reflects the speed of your CPU & RAM, and how they coroperate, a better time reflects a better performance, in VERY heavy games, cause mostly in games, the bottleneck is the graphic card, since nobody plays in 640*480, but... A boost is a boost

[coop]

If you do pi faster, then everything will probably be faster. I'm not going to argue your other point. :)

[Dumo]

Lets not crap on this Jochen thread :)
Any other thought (other than tweaks for 32m)..... open your own thread..

[Jochenp]

Thnx Dumo:) Anyone want a specific tweak to be benched?

[Onepagebook]

Well the most efficient way I use is set as tight as you can for TRFC and TRTW, these two value are really the keys for reducing the last couple iterations running time.
ex::)

http://img382.imageshack.us/img382/5733/3560pi32m0522tpmw8mg.jpg

[Jochenp]

Yes, offcourse ram-timings are much more important then any windows-tweak, but this thread is about OS-tweaks only :)

[[XC] moddolicous]

Nice thread jochenp. Can u try like tref=200 15.6us & idle cycle limit=256clks, and then tref=100 1.95us & idle cycle limit= 000? Always wanted to see how much of a difference it makes. Nevermind, just saw that it was windows tweaks, not ram timings. If you have extra time though, might be worth a shot. Does the maxmen tweak work with SPI 32m?

[SquireSCA]

Yes, offcourse ram-timings are much more important then any windows-tweak, but this thread is about OS-tweaks only :)

Well, that sorta addresses my point. A RAM timing tweak would yield real benefits in most areas of computing performance. But an OS tweak done to slightly improve the calculation time of one iteration of SuperPI, may yield no benefit to the end user, other than the miniscule improvement in that one benchmark.

[targ]

Well, that's what I was trying to ascertain. I see some pretty exotic tweaks and lenghts that people go to just to get a couple artificial points in some obscure benchmark, and wondered why. I will tweak and overclock if needed in order to get better performance for some real world task that I want to use my PC for, but to get liquid nitrogen cooling, running the system outside of a case and running a hacked and modded OS, just to get a higher number on what amounts to a long division equation... well, it just seems strange to me. hehehe

its like a 100m sprinter wearing lyrca out fit, or good set of spikes.

when your going for a WR then it has to be done.
targ

[Weee]

i gotta go try these tonight =P

[Jochenp]

I'm sorry, the next 2 weeks I'm not coming online anymore, and I don't have time to test any tweaks.
I've got exams starting tomorow and I really need all my attention focused on them.

[da-key]

Well the most efficient way I use is set as tight as you can for TRFC and TRTW, these two value are really the keys for reducing the last couple iterations running time.
ex::)

http://img382.imageshack.us/img382/5733/3560pi32m0522tpmw8mg.jpg

Good find. Not the easiest to lower trfc at the limits ;)

[goreblast]

SquireSCA stop junking this guys thread. And the only thing that riding a bike at 130MPH will most definately get you is dead.

[SquireSCA]

SquireSCA stop junking this guys thread. And the only thing that riding a bike at 130MPH will most definately get you is dead.

Why are you telling me to drop something that I have not taken part in, in over a month?

I did drop it, so you are actually bringing it to the forefront.

And who said I was driving at 130mph?

Geez, it goes faster than that, give me some credit. I hit 154mph a few months back. ;-)

[Jochenp]

Well, thats just great, now back to 32MB Pi!
After a lot of experimenting, I noticed that 1 and a half minute of prime blend stable = 32MB SuperPi stable! :D
And @ moddolicous, I'll try settings those tonight.