SuperPi Measurements and OS

[BenchZowner]

Lo everybody,

I don't know if there's another similiar comparison thread already around, but since I performed these tests for my new website ( not yet ready/online ) I thought I should give them a go here aswell :)

Operating Systems:
Windows XP SP2
Windows 2003 Enterprise Server [ Defaults * ]
Windows 2003 Enterprise Server [ "Modified" * ]
Windows XP Professional x64
Windows 2003 Enterprise Server x64

Benchmark:
SuperPi Mod v1.5 XS (http://www.xtremesystems.com/pi)

System Specs:
Motherboard: Asus P5W DH Deluxe [ BIOS v1503 ]
CPU: Intel Core 2 Duo E6600 [ 3.3GHz / 367MHz FSB * 9 ]
RAM: OCZ PC2-6400 S.O.E. Urban Edition [ 2GB Kit, @ DDR734 4-4-3-11-4 ]
[ timings = tCL-tRCD-tRP-tRAS-tWR ]
[ BIOS Settings => Download (http://rapidshare.com/files/774755/SuperPi-And-OS-BIOS-Config.rar)

What's the difference between the "defaults" & "modified" Windows 2k3 E.S. Measurements?

The Windows 2003 Server Series Operating Systems, are set up by defaults to give priority ( Processor Scheduling & Memory Usage ) to background services.
So I decided to test these settings aswell, so:
Defaults = Priority to background services [ CPU & MEM ]
Modified = Priority to programs [ CPU & MEM ]

During the testing session I noticed that in Windows 2003 Ent. Server the scores were better when the system was set to give priority to background services, that's why I decided to test the x64 version only with the default settings ( priority to background services ).

For a fair comparison I decided to have the same services running on every OS test ( matched the Disabled & Manual Startup services, and the automatic ones ) [ whereas applicable ]
After finishing up the tests I've noticed that I forgot to enable the "Themes" service in the x64 Editions, but since the results were already worse than the other OSs, I decided to skip re-testing them with the "Themes" service enabled.

Here's what I've changed in every OS config (http://rapidshare.com/files/774407/SuperPi-And-OS-OSconfig.rar)

Final results table:

http://img79.imageshack.us/img79/7029/superpiandosresultstablsi0.jpg

Download the Results screenies here (http://rapidshare.com/files/764633/SuperPi_Measurements_And_OS.rar)

[kemo]

Well nice work with a lot of efforts and I haven?t seen any similar thread
BTW from where you got all that time and patience I haven?t even run 32M on my rig
________
Ford Tempo specifications (http://www.ford-wiki.com/wiki/Ford_Tempo)

[BenchZowner]

Well nice work with a lot of efforts
Thanks :)


BTW from where you got all that time and patience I haven’t even run 32M on my rig

Well, when it comes down to business ( profit or non-profit ) & my hobbies, I give my best right at it :D
When I say somethings, or think of testing/checking something, I'll go for it no matter what it takes ( time usually :D ).

[uOpt]

If you want fast SuperPi you have to use Linux.

Linux constantly gives better times at the same clocks for large SuperPis.

The reason is almost certainly that the page walk on TLB misses is simpler.

[before]

Nice work ;)

[I)ickie]

What was the change between Windows 2003 ent Server and Windows 2003 ent Server (modified)?

Were you referring to the rar file. If you were you might want to change first thread from Here's what I've changed in every OS config to something else.

Nice results man must have taken you a long time. Very interesting read.

Thanks

[BenchZowner]

If you want fast SuperPi you have to use Linux.

Linux constantly gives better times at the same clocks for large SuperPis.

The reason is almost certainly that the page walk on TLB misses is simpler.

I know, I still remember those crazy 8+ seconds difference with the same config back in the Athlon XP days :D

But this article is all about our beloved SuperPi Mod v1.5 :)

[BenchZowner]

What was the change between Windows 2003 ent Server and Windows 2003 ent Server (modified)?

Well, to make things clear:

The only difference between Windows 2003 Ent. Server "Defaults" & Windows 2003 Ent. Server "Modified" is:

http://img122.imageshack.us/img122/686/makingthingsclearif2.jpg


Nice results man must have taken you a long time.

Around 45-50 minutes per OS ( Operating System )

Thanks for your comments ;)

[lawrywild]

My windows 2003 server standard, came with those "modified" settings already set.

[BenchZowner]

My windows 2003 server standard, came with those "modified" settings already set.

Two thoughts:

1) Windows 2003 Server Standard Edition comes with priority preset to programs [ I don't think so ] { but I haven't tested/used the Standard Edition, so I can't be sure }

2) Perhaps your W2k3SE disc was modified ?

[BenchZowner]

Just posting because I've just finished installing Windows 2003 Enterprise Server on another machine, and the priorities defaulted to "Processor Scheduling = Background Services [ logical stuff :) ] and Memory Usage to...programs :D"
Seems like each install is unique ? :p:

[goreblast]

I did a similar comparison which was posted here; Im guessing its archived now. There is a link to a copy of my findings on my own website.

[DAK1640]

Good work BenchZowner...I guess there's not that much difference in mathmatical computations between the various windows OS...I was always curious, now I know for sure...:toast:

[uOpt]

Good work BenchZowner...I guess there's not that much difference in mathmatical computations between the various windows OS...I was always curious, now I know for sure...:toast:

The math doesn't matter here.

What matters is that huge areas of memory are touched during the high SuperPis. That trashes both the caches, which doesn't matter, and the TLB. Trashing the TLB does matter because a TLB miss causes a page table walk and the speed of that can be dependent on the OS, and it is why Linux is so much faster.

[mat128]

wow good job! thanks for the info, keep the hard work!

[dinos22]

stock is one thing but tweaked OSes is another.......................with a tweaked XP no other is match.....or they are close....

Linux SPI benching i'd like to see that heheheh

please show us some examples how linux consistently beats XP (but make sure you tweak both OSes for SPI and not just run it standard)

[uOpt]

Linux SPI benching i'd like to see that heheheh

please show us some examples how linux consistently beats XP (but make sure you tweak both OSes for SPI and not just run it standard)

Please pick one existing result of SuperPi 32M on a socket 939 AMD64 with all clockspeeds, memory speed and memory timing given.

I will clock one of mine precisely the same, run SuperPi under Linux and post the result. So far it always came out top.

[dinos22]

Please pick one existing result of SuperPi 32M on a socket 939 AMD64 with all clockspeeds, memory speed and memory timing given.

I will clock one of mine precisely the same, run SuperPi under Linux and post the result. So far it always came out top.
see my link in sig or go to 32M Low Clock Challenge thread here and see my results..............you'll see all the settings there too so you should be able to do something.................sub-24m times you should compare if you want...........i did a lightning fast run at 2960MHz 23M 5Xs :)

if you beat that time I AM MOVING to linux for benchmarking SPI :D

BUT SPI Version has to be 1.5 XS version or 1.4 which is same as 1.5 'cept for the name

[uOpt]

see my link in sig or go to 32M Low Clock Challenge thread here and see my results..............you'll see all the settings there too so you should be able to do something.................sub-24m times you should compare if you want...........i did a lightning fast run at 2960MHz 23M 5Xs :)

if you beat that time I AM MOVING to linux for benchmarking SPI :D

BUT SPI Version has to be 1.5 XS version or 1.4 which is same as 1.5 'cept for the name

We need to do that socket 939. My Core2 and other non-939 systems are in actual use.

Can you post a concrete link to one result?

[dinos22]

here is a couple of tough ones


http://img113.imageshack.us/img113/8493/296mhz253363603vtightertimings.jpg (http://img113.imageshack.us/my.php?image=296mhz253363603vtightertimings.jpg)



here is one with CH-UTT RAM in case you don't have TCCD
http://img480.imageshack.us/img480/5149/2990mhz249onramtref312023m4789.jpg (http://img480.imageshack.us/my.php?image=2990mhz249onramtref312023m4789.jpg)

[uOpt]

dinos22, you don't get the point.

We don't want to see who's memory goes higher.

We want to see what Windows and Linux have for SuperPi 32M times for something that I can reproduce without too much fiddling.

Having said that, I can beat the TCCD clock but this is less fun than I thought.

[dinos22]

dinos22, you don't get the point.

We don't want to see who's memory goes higher.

We want to see what Windows and Linux have for SuperPi 32M times for something that I can reproduce without too much fiddling.

Having said that, I can beat the TCCD clock but this is less fun than I thought.
stop with the excuses already.........i'm giving you a fast time

any half decent kit of BH5 or UTT will do 258MHz 2-2-2-5 so just try beat that

or if you can beat TCCD show some screenshots

[uOpt]

It is time for you to take a pick in your life.

Do you want a pissing contest about higher clocks?

Or do you want to know what the times of Linux and FreeBSD are at the same speeds?

[Drager2]

http://www.goreblast.com/superPI.htm - there are more tests... but nice work BenchZowner!

[dinos22]

It is time for you to take a pick in your life.

Do you want a pissing contest about higher clocks?

Or do you want to know what the times of Linux and FreeBSD are at the same speeds?
dude i stopped benching AMD ages ago..........i dug up a couple of screens for you which i remember were decent numbers for AMD........i also told you about 32M SuperPI Challenge thread...............so what else do you want >>>>

now put you money where your mouth is and show some real world numbers for everyone to enjoy................why is that so hard seriously?

i think i'm trying too hard to provide material i think......i don't think you've got anything to show otherwise info would be up by now

come to think of it...........i don't need to provide any info man,.....we all contribute here so do a little bit of homework and test on both windows and linux and compile results into a thread and enlighten everyone how linux beats windows OSes

but you have to use SuperPI mod1.5 which is XS version.......any other version apart from mod1.4 will yield different results and be deemed unusable

[farksy]

It is time for you to take a pick in your life.

Do you want a pissing contest about higher clocks?

Or do you want to know what the times of Linux and FreeBSD are at the same speeds?
Ok put it this way.

The other time, i tried tweaked and untweaked Xp at 2.98Ghz on an Opteron 146 with UTTs. Was the exact same divider dinos was using. I just about hit under 24min with the help of the tweaks. On an untweaked OS that would have ballooned up to around 24-25 minutes.

He was trying to state a small comparison but have no untweaked results to fall back on. Say, if your tweaks to reduce AMD times by half a minute, 15 seconds in Intel shouldn't be a problem either. Hence referring back to the original results you can see that the Xp with tweaks would have come out tops in 32mil cat as well. =D

Cheers guys.

[vanquish]

^ Nice work there BenchZowner !

So after all, SP2 aint that bad at all for 1M benching ;)