SuperPI 32M for testing CPU and system speed and stability

[El<(')>Maxi]

Your two FX-55 have highest CPU clock and lowest SuperPI 32M run time, but in terms of the performance measure per CPU clock measured by

performance_per_CPU_clock = 1 / (runtime * CPU_frequency)

the two FX-55 are behind

Just out of curiosity, where does this run rank using this formula hitechjb1?

http://img.photobucket.com/albums/v31/canny/271x1132mbPI.jpg

 

[NuDNiK]

sorry about that, deleted it, ran a new one w/ new settings

CPU: Venice 3200+ LBBLE 0516EPGW
memory: G. Skill 4400 LE 2 x 256 MB (Samsung TCCD)
motherboard: DFI LanParty UT NForce4 Ultra-D (rev. A02, bios 03/10/05)
cooling: XP-120, 120 mm
OS: Windows XP Professional SP2

- CPU: 2736 GHz = 304 MHz x 9, 1.664 V
- memory: 304 MHz, 2.5-4-3-7 1T, 3.0 V
- SuperPI 32M completed in 27 min 10.454 sec

http://putfile.com/pic.php?pic=5/13420493388.jpg&s=x4

EDITED, messed up post, all clear now

 

[hitechjb1]

Your two FX-55 have highest CPU clock and lowest SuperPI 32M run time, but in terms of the performance measure per CPU clock measured by

performance_per_CPU_clock = 1 / (runtime * CPU_frequency)

the two FX-55 are behind

Just out of curiosity, where does this run rank using this formula hitechjb1?

http://img.photobucket.com/albums/v31/canny/271x1132mbPI.jpg

Since performance of a processor (CPU, GPU, ...) is usually measured as instructions per second or integer/floating point/pixel/graphic operations per unit time, i.e. the shorter the run time, the higher the performance. So

performance = instructions per unit time = K / run_time

where K is a constant.

Since we are interested in ranking the different types of processors (e.g. San Diego, NewCastle, Barton, Prescott) according to how much instructions per unit time regardless of how fast they are clocked, so the above performance measure is further divided (normalized) by the clock frequency.

normalized_performance = K / (run_time * frequency)

So for the same processor, whether it runs at 2 GHz or 3 GHz, the frequency-normalized performance would be about the same under the same other conditions, although in absolute terms, the 3 GHz one will complete the computation in about 50% less time than the 2 GHz one.

This frequency-normalized performance measures the relative amount of instructions (or computation) a processor performs per given amount of clock cycles, regardless of clock frequency, regardless of cooling (air/water/sub-zero). It reflects what we commonly want to find out how the different types of processors perform clock for clock.

Based on the statistics collected (may not have enough data point), the relative amount of instructions (or computation) per clock for these processors are as follows (as of 05/15/2005):

It becomes apparent that clock for clock (would like to see more data points to further refine the numbers),
- San Diego performs better than ClawHammer (~4%)
- Venice performs better than Winchester (~4%)
- Venice performs better than NewCastle (~5%)
- ClawHammer performs better than NewCastle (~2%)
- various A64 perform better than Barton/Tbred B (13-24%)
- Barton performs better than Tbred B (~4+%)
etc.


Click these two links for result tabulation and analysis:
Different core performance results in SuperPI 32M
Analysis of different core performance in SuperPI 32M

 

[deathstar13]

hitechjb1- a ? for ya.
i notice the graphs are down so i guess you recomputing them but just for instance in my case in what you said above cant those results of frequency-normalized performance be scewed by ram fsb and timings? or is that taken into account?

i say this as i lowered my speed 100mhz and fsb by 5 or so but lowered my timings and knocked off 1 min 30 sec from my time.
so in the end wouldnt this help in frequency-normalized performance scores?
if it does its only fair as i was being hurt badley at cas 3.0-3-3-7 and made my cpu look worse than mabey it should have.

anyhow i apprreciAate all the work you put into this.and i do see my times going down more soon as i have some tccd and a volt mod for my UTT coming.i just love tweaking
and one thing about super pi i really like is no tweaking is cheating and are real scores unless its outright fraud.

also is super pi mod1.4 fine to use for these runs?i like the millisec results for testing but hate throwing them out and running again a 30min test just to see what i just did again.

 

[Clevor]

Just picked up a 3700+ San Diego. Has a real ugly stepping, but can't complain so far. Here is a 32M at just 2700 mhz. Not bad at all for the clock.

The clock ain't big, but figure this is at 1.376 VCORE for starters. The rig passes all my 3D benches of course. If you want, I can even show you a 1M time of 30 sec. Believe it or not, I am running TCCD at 300x9/166 divider, so the ram is only running 245, 2.5-3-3-6! I've never clocked this low with ram that slow!

Board is a 9NDA3+. I'd be happy if this chip matches the 2763 mhz of my Winnies, as the 1MB L2 is killer in 3DMark2000/2001 (I'll be posting shortly on this).

 

[hitechjb1]

Just picked up a 3700+ San Diego. Has a real ugly stepping, but can't complain so far. Here is a 32M at just 2700 mhz. Not bad at all for the clock.

The clock ain't big, but figure this is at 1.376 VCORE for starters. The rig passes all my 3D benches of course. If you want, I can even show you a 1M time of 30 sec. Believe it or not, I am running TCCD at 300x9/166 divider, so the ram is only running 245, 2.5-3-3-6! I've never clocked this low with ram that slow!

Board is a 9NDA3+. I'd be happy if this chip matches the 2763 mhz of my Winnies, as the 1MB L2 is killer in 3DMark2000/2001 (I'll be posting shortly on this).

Your SuperPI 32M window is too small and the total run time does not show up. You need to enlarge the SuperPI 32M window before the completion of the run in order for the run time to show up. Also please run SuperPI 32M using the SuperPI_mod version (download link in first post of this thread).


Please post your result per the format of post 2 as outlined in the first post of this thread. It would help me to get the information needed for the analysis and tabulation of the results:

Different core performance results in SuperPI 32M
Analysis of different core performance in SuperPI 32M

 

[hitechjb1]

hitechjb1- a ? for ya.
i notice the graphs are down so i guess you recomputing them but just for instance in my case in what you said above cant those results of frequency-normalized performance be scewed by ram fsb and timings? or is that taken into account?

i say this as i lowered my speed 100mhz and fsb by 5 or so but lowered my timings and knocked off 1 min 30 sec from my time.
so in the end wouldnt this help in frequency-normalized performance scores?
if it does its only fair as i was being hurt badley at cas 3.0-3-3-7 and made my cpu look worse than mabey it should have.

anyhow i apprreciAate all the work you put into this.and i do see my times going down more soon as i have some tccd and a volt mod for my UTT coming.i just love tweaking
and one thing about super pi i really like is no tweaking is cheating and are real scores unless its outright fraud.

also is super pi mod1.4 fine to use for these runs?i like the millisec results for testing but hate throwing them out and running again a 30min test just to see what i just did again.

Currently, the frequency-normalized performance score is only normalized using the CPU frequency.

frequency_normalized_score = K / (run_time * CPU_frequency)

I attempted to include memory frequency such as using

frequency_normalized_score = K / (run_time * (A CPU_frequency + B memory_frequency))

but since I don't have the weights A and B known and worked out, I defer that.

In general, for SuperPI 32M and other benchmarks, I would say

run_time = A0 + A1 / CPU_freq + A2 / memory_freq + A3 (memory_timing) + ...

Or even further

run_time = A0 + A1 / CPU_freq + A2 / memory_freq + A3 tCL + A4 tRCD + A5 tRP + A6 tRAS + ...

where those constants A0, A1, A2, A3, ... depends on individual benchmark and have to be determined empirically (by changing the variables of frequency, timing and curve fitting the run_time with them, more details left out for now). The constant A0 accounts for program overhead. For most programs, I think A0, A1, A2, A3, A4 are the main dominating components. When I get a chance, I may attempt to do that (not sure when for now).

For memory frequency and memory timing, there is an equivalence between the two and there are ways to trade-off between them. E.g. 2-2-2-x 1T at 250 MHz ~ 2.5-3-3-x 1T at 300 MHz (roughly speaking).


Regarding to other questions,

SuperPI mod is actively supported and being revised once a while. I am not aware of any hidden problems (if any). Version 1.4 is current.

In your results, the memory voltages are missing, please fill them in.

If possible, use the format as listed in post #2.

 

[Clevor]

I'm slowly creeping up the clock of my 3700+ San Diego. This is at just 2810 mhz and nope, it's not BH-5 either, but TCCD at 2.5-3-3-6. Nothing fancy like phase or FX either, just a Corsair HC200. Apparently the 9NDA3+ is a fast board:

 

[Clevor]

Ram is 2x512 GSkill LEs at 2.5-3-3-6. CPUZ says 2 GB, but obviously, Epox needs to fix the BIOS, just as maybe CPUZ needs an update to recognize the San Diego properly.

 

[Gautam]

Lol, or maybe you need to update to the latest version of CPU-Z...

Nice score.

 

[Clevor]

Here's the first CPUZ screenie. Excuse the (big) size of the screenies, I'm still learning the ropes. 2810 with TCCD also breaks me into the 29 sec SuperPI 1M bracket.

 

[Clevor]

Lol, or maybe you need to update to the latest version of CPU-Z...

Nice score.

I only have time to bench, not to take screenies and update my software!!! I only recently upgraded to the SuperPI program Hitechi recommended. Hell, the SuperPI file I was using is dated sometime in 2003. No cheats anyway .

 

[hitechjb1]

Update of results:

Watch how the yellow bar trending downwards in the last graph as performance clock for clock decreases.



For background and explanation of the analysis:

Analysis of different core performance in SuperPI 32M

 

[Clevor]

Here is the 32M of that 3700+ San Diego at 2904 mhz at 1.568 VCORE. The time is not that impressive since I am running TCCD at 264, 2.5-3-3-6 (don't want to be flogging that CPU with UTT and 3.4 VDIMM). Plus the board don't have the spectrum of BIOS tweaks like DFI boards so A64 Tweaker says the TCCD is running 2-3-2-4 Trrd-Trwt timings.

Spec:

9NDA3+, Rev. 2.1
3700+ CABGE 0517UPAW (another xPAW winner!)
2x512 GKill LE at 264x11, 2.5-3-3-6
6800 Ultra at default clock
HC200
VCORE=1.568
VAGP=1.5
VCHIP=1.65
Room/water/CPU temps(load)=22C/27C/27C

 

[Clevor]

CPUZ shots (I upgraded!):

 

[Clevor]

Other CPUZ:

 

[stang8118]

I ran Super_Pi 32m to check for stability. I was still using the computer when it was running the loops. I got 28m 10s, is that good even though i was still using the computer to surf the net and stuff? System is in sig.

 

[Clevor]

BTW, on this particular CPU anyway, I am finding that SuperPI 32M correlates VERY closely with 3DMark2005 clearing (particularly the two CPU tests). If I can get by 3DMark2005 in its entirety, 32M is OK. This is good enough for me since I didn't buy my rig to watch it run Prime overnight.

At one clock, 3DMark2005 kicked out of CPU2, a common occurance when taking a CPU to the max. 32M failed at the 20th loop. Upped VCORE a notch so the Mark passes, 32M also passed. I'm not saying this is the case with all CPUs, but this one at least.

There seems to be a lot of bashing recently of using 3D tests for stability, e.g. if you didn't run Prime it's not stable. Sorta like there is only one God on Earth and it is _____.

 

[Clevor]

I ran Super_Pi 32m to check for stability. I was still using the computer when it was running the loops. I got 28m 10s, is that good even though i was still using the computer to surf the net and stuff? System is in sig.

That's very good if it did that, as far as stability is concerned. But it may have taken something off your time. I personally don't bother jumping thru hoops when I run SuperPI; stuff like disabling all TSRs, minimizing the window, configuring special memory settings for XP, etc. I always get that stupid Microsoft automatic updates box popping up at the bottom, asking if I want to check for updates. I can never figure out how to disable it anyway .

 

[HousERaT]

Here is my new time.... not the fastest but should increase my performance per cpu clock

CPU: A64 3000+ Venice LBBLE 0516 CPBW
Memory: Mushkin Redline PC3200
Motherboard: DFI LanParty UT NForce4 Ultra-D BIOS 5/10-2
Cooling: XP-90/Delta 92mm Triple Fin
OS: Windows XP Pro SP2

CPU: 2.7 GHz = 300x9 (with 166 Divider)
Memory: 245 MHz, 1.5-2-2-4 1T, 3.2v
SuperPI 32M completed in 26 min 46 Sec