SuperPI Contest!!!

[Googer]

Originally posted by: zakee00

Originally posted by: bobsmith1492
Sager P-M Centrino lappy, 1.7 stock + 25% FSB OC for 2125.5 MHz; 1 gig ram at 208.4 (up from 333)
36m, 26s

That seems kinda slow compared to all these 20-min times.... are you supposed to reboot and close everything for maximum efficiency or what am I doing wrong?

superpi seems to run better on AMD's, but try defragging/reboting your computer. and always close everything when you are running a benchmark.

Because of it's memory intensity and demand for memory bandwith. AMD's onboard controller helps a lot.

 

[Googer]

Originally posted by: zakee00
there actually is a Prescott optimized version of SuperPI. Please state if you are using it or not though.

I was unaware of that, what version is it? what is the stepping?

 

[zakee00]

Originally posted by: Googer

Originally posted by: zakee00
there actually is a Prescott optimized version of SuperPI. Please state if you are using it or not though.

I was unaware of that, what version is it? what is the stepping?

i dont know; i read that on HardOCP forums. try googling it.

 

[DrFragaTron]

Well on my first go around I got a 43m27s on a 2.4 P4 which doesn't seem too bad at all.

I'm gonna try some different stuff out and see what happens.

The Screen

 

[RussianSensation]

Originally posted by: DrFragaTron
Well on my first go around I got a 43m27s on a 2.4 P4 which doesn't seem too bad at all.

I'm gonna try some different stuff out and see what happens.

The Screen

Thats on P4 2.4 overclocked to 3.0ghz and having 250FSB. If you run 1:1 this also gives a huge boost. Regular 2.4ghz would be close to 55 min or 1 hour.

 

[Mloot]

Originally posted by: RussianSensation
Update:

Just threw in some PDP PC3200 1 gig Samsung TCCD. Reran the bench at 1:1 => 246/246 @ 2.5-3-3-6

New score at identical CPU speed of 3.2ghz => 38min 3sec

That is a decrease of 7min 9sec from 246/198@2-2-2-5. Therefore, SuperPI is definately memory bandwidth hungry.

RussianSensation, those are pretty cool results. I'm running my P4C at ~200mhz faster, but it looks like your faster FSB definitely makes up the difference. My time of 38m 01s is only 2 seconds faster.

 

[DrFragaTron]

Originally posted by: RussianSensation

Thats on P4 2.4 overclocked to 3.0ghz and having 250FSB. If you run 1:1 this also gives a huge boost. Regular 2.4ghz would be close to 55 min or 1 hour.

I can't figure out how to get to 1:1 ? I thought I knew what I was doing but then you bust out ratios and I feel like a n00b

 

[zakee00]

well, in order to run 1:1 your ram has to be good. no value stuff.
go to your BIOS and look for something like "Ram Speed", and pick "200", or "auto". thats how it is in my bios anyway.
good luck,
nick

 

[btbam]

32:49 1:1

and i was chatting on aol and had other aps open, dont know if that affects it

steady 40c as it ran is that good?

 

[zakee00]

OK, heres my updated score with my finalized (I hope) overclock, complete with pic :
29 Minutes, 16 Seconds. Not too bad, I wish i had a higher CPU multi so I could run at a lower HTT, so I could run 1:1 at 265HTT instead of 240MHz with a 166 devider.
Pic

 

[RussianSensation]

Originally posted by: DrFragaTron

Originally posted by: RussianSensation

Thats on P4 2.4 overclocked to 3.0ghz and having 250FSB. If you run 1:1 this also gives a huge boost. Regular 2.4ghz would be close to 55 min or 1 hour.

I can't figure out how to get to 1:1 ? I thought I knew what I was doing but then you bust out ratios and I feel like a n00b

In your BIOS under System or CPU configuration or something along those lines you should encounter an option called CPU:RAM ratio or RAM:CPU ratio or FSB:RAM. Ratios of 5:4, 3:2, 1:1, 5:6 and so on exist.

Here's a brief explanation:

your cpu is 2400mhz = 12 multiplier x 200 FSB
If you raise the FSB you are then increasing the cpu speed as you have done by 50FSB to 250 x 12 = 3000mhz. Originally at 200FSB the ram for P4 C platform (being i865/875 chipsets) runs at 200mhz as well. So the default setting should be 1:1 = CPU:RAM ratio = 200FSB: 200mhz. Now since ram is DDR (double data rate), to figure out the ram speed you simply multiply 200mhz x 2 = 400mhz or x 8 = PC3200 speeds.

Therefore, to run 250FSB:250mhz ram (1:1), you'd need ram that is capable of hitting 250mhz (500 effective). On an A64 platform, memory bandwidth isnt so important as are timings as Zebo has shown. But for P4, memory bandwidth benefits a lot more. So go into your BIOS and first relax all the ram timings to say 3-4-4-8 and raise the voltage to 2.8V and see if you ram can do higher speed than 400mhz by setting the CPU:RAM ratio to 1:1 and going to 205/205, then 210/210 and so on. Then you can tighten the timings once you reach the desired speed. To see what ratio you are running, just check in CPU-Z or right at motherboard bootup screen which tells you the memory speed. Since you already reached 250FSB, I am assuming your AGP/PCI ratio is fixed at 66/33.

What ram and motherboard do you have BTW?
_____________________________________________________

Mloot, I think SuperPI showed that P4 responds very well in that application to increased memory bandwidth. When I ran at 246FSB/198 memory, the score was in 45 min. So the increased FSB was not enough to offset the slower time. Clearly the 50mhz of memory increase (100 effective) is what allowed me to shave off 7 min. Still A64's mathematical calculation prowess shines in this bench. AMD always performed better in calculation benches (even XP whoops P4's arse).

 

[Zebo]

36:35 Win 2000


28:43 WinXP

 

[GuitarDaddy]

Originally posted by: Zebo

36:35 Win 2000


28:43 WinXP

WoW! And I used to love 2k:frown:

 

[DrFragaTron]

Man I feel like you're making me do homework!

The problem with me understanding things in terms of ratio is probably due to the fact that my BIOS has no mention of a ratio other than A CPU/FSB but thats in the description of the multiplier (12x).

In the menu I see:

CPU External Freq.: (250)
CPU Ratio: 12
DRAM Freq.: 266,320,400,Auto (It's on auto as of now)
AGP/PCI 66.66/33.33, 72.73/36.36, 80.00/40.00 (On auto as well)
DDR Ref. Voltage" 2.55,2.65,2.75,2.85 (on auto as well)
I can also change DRAM timings but I have that set to determined by SPD which is an auto config. as well.

Here are a few CPU-Z shots of the memory stuff (I have 2 paired sets of memory for dual channel).

Photo

 

[Zebo]

Originally posted by: zakee00
OK, heres my updated score with my finalized (I hope) overclock, complete with pic :
29 Minutes, 16 Seconds. Not too bad, I wish i had a higher CPU multi so I could run at a lower HTT, so I could run 1:1 at 265HTT instead of 240MHz with a 166 devider.
Pic

See here is where you see low latency comming into play like my tests shown. Same processor speed but quicker due to LL.

28 flat with async 166 memory 2655Mhz.

You don't want 1:1 trust me. It's slower due to crap timings to get there. What you ideally want is 2-2-2 at highest bus possible running async.

 

[zakee00]

i wish i knew this earlier, like before i bought my system lol. i bought Geil DDR500 ram thinking that running in sync would be better. this is my first oced system, so i think i did pretty good nontheless.
wow i cant belive lowering the timings from 2.5-4-4-7 to 2.5-2-2-7 can make 1min 19sec difference in this benchmark!!!!
since latency is ovbiously more important, right now im memtesting my ram with a 133 devider, 2-4-4-7 1T. if it is stable (which i doubt ) then ill try tightening the timings some more.
thanks for all of your hard work zebo, you save so many people so much time
Nick

 

[Zebo]

PM nick1985 and see if he'll run 295 1:1 with his TCCD. I'm pretty sure it will be slower since he has to run 2.5-4-4 1T there. I sold mine so I can't do it anymore...

 

[Zebo]

Zakee try 133? Will that geil stuff even do 2-2-2 there? That would be 185Mhz... Heck value ram can do that....surly PC4200 expensive ram can.. But duvie had bad luck with geils.

 

[Stormgiant]

Super Pi is one of the benchmarks that loves Cpu Mhz + high FSB + high speed memory. That's why it is widely use for the comunity of overclockers to compare machines.

I get 31seg with my 3000+@2620 with memory @183 ratio giving 262mhz @2-2-2-5
Also get about 28seg with 3000+@2430 on Linux 64bits :Q

Gonna give a try at the 32MB bench...

 

[zakee00]

lol zebo i dont have TCCD. geil makes their own chips. my stuff maxes out at 273MHz. i ran memtest last night but someone in my family got out of it so i dont know if there were any errors or not...jeez makes me mad. i doubt my stuff can do 2-2-2-7, simply because it is ment for high speeds-loose latencies not low speeds tight latencies.
today im going to try and make this more of a "contest" and edit my OP with the lowest AMD/intel scores so far. only ones with a screenie will be considered.

 

[Googer]

I don't have the best or worst time, but here it is 1h 36m 56s.
PentiumIIIs Tualtin 1.4ghz
512k full speed cache
1 stick of 512mb cas 2-2-2 Crucial pc133
Windows 2000 Professional, Service Pack 5

Lets see who has the slowest as well as the fastest time.

 

[Googer]

Originally posted by: zakee00
lol zebo i dont have TCCD. geil makes their own chips. my stuff maxes out at 273MHz. i ran memtest last night but someone in my family got out of it so i dont know if there were any errors or not...jeez makes me mad. i doubt my stuff can do 2-2-2-7, simply because it is ment for high speeds-loose latencies not low speeds tight latencies.
today im going to try and make this more of a "contest" and edit my OP with the lowest AMD/intel scores so far. only ones with a screenie will be considered.

To the best of my knowledge GeIL is not a producer of semi-conducters, but a manufacuter of memory modules (DIMM's). They get their chips from somewhere else and re-print their name on the top of the Chip and package it on to a PCB. There are several companies that buy blank chips from a source and re-print their companies name on it.

I have see some no-name DIMM's at Best Buy where the brand was printed right on the chip it's self.


Some of my Informantion I read on Micron's website 4 years ago, so my memory is a bit faded. Micron resells chips with no silk screening on them to module manufacturers for reprinting.

 

[zakee00]

i wonder who DOES make my ram chips, and what model they are?
lol i'll add the slowest time so far to the list also

 

[Zebo]

Yup googer is right, only companies who make thier own are crucial and samsung...every other memory you buy is from someone else. I don't know what modules those geil are but frankly I've never seen any good clocks with geil...

Can you return them? I mean after all, they arn't even running stock speeds/timing which IMO is just cause for RMA.

 

[zakee00]

i dont want to go through the trouble, then my system would be down for a week etc. plus, i got them a bit before christmas so its probably too late. i mean, im talkin like 1fps difference here. its only in senthetic benchmarks will i ever tell a difference.