Sign in with
Sign up | Sign in
Your question

Super Pi Question

Last response: in CPUs
Share
November 7, 2006 11:07:02 PM

Guys,

I've never run Super PI before, and am not familiar with the program. How do I know what to bench my result against (see below)? Is this good or bad?

Thanks!

More about : super question

November 7, 2006 11:30:13 PM

Quote:
To give you some scale on the 1M calculation
E6700 1M is about 20-21 seconds
E6800 1M is about 18-19 seconds
P4 3.6 GHz is about 41-42 seconds
AMD FX-62 is about 33-35 seconds


I'll add to that list

AMD Athlonxp 2500+ 1m is about 49-51 seconds :tongue:
November 8, 2006 12:07:30 AM

Semi random: Anyone know where I can find the C++ source code for SuperPi or similar programs?
Related resources
Can't find your answer ? Ask !
November 8, 2006 12:13:10 AM

Guys,

I got 16 seconds for the 1M calc. This seems good no?
November 8, 2006 12:29:59 AM

Very good time. With my D805 @ 4.0Ghz I think I run 31 seconds.
November 8, 2006 12:58:35 AM

Cool. Thanks.
November 8, 2006 12:59:01 AM

Cool. Thanks.
November 8, 2006 1:01:37 AM

I ran 1M Super Pi in 21s.

Nothing overclocked, E6600/2GB DDR2-667.

=)
November 8, 2006 1:31:49 AM

Is Super_PI really a good benchmark though?

I've seen so many different results... makes my head spin. :?

Like for example, my P4 3.0/800 gets around 50 secs, yet I've seen others get 40-46 secs.

My dad's poor old Amd XP2400+ does around 1min 10 secs, and I've seen scores of 40-49 secs.

:cry: 
a b à CPUs
November 8, 2006 5:02:58 AM

Well for me, Super PI and other benchmarks are only intended to give you an idea of the relative performance of the CPU. Some may jump up and down a few steps depending on the task at hand. But usually the jump is realtively small, though this may be false with processors that simply have a different design. Such as memory access times between certain Pentium D dual-core vs Athlon FX. The Pentium D may be faster at video encoding but due to the FX's design, it is better at memory access and jobs such as mp3 encoding. (I am only talking about the highest end of the Pentium D)


Benchmarks usually just give you an idea. There is no absolute way to test how well a processor performs, since you also have to take into account its sorroundings (mobo, ram, hdisk, etc). Though a ton of money would surely get you the fastest pc in town.
November 8, 2006 6:14:40 AM

Heh, feeling abit tired, so I'm not sure if my question on Super_PI was clear.

So I guess what I was asking, in comparing a P4 3.0/800 system to another P4 3.0/800 or another P4 system OC to 3.0, and getting different time results, just puts a question in my mind in how that can be? Especially when its a CPU test based on computational power?

As well as comparing AthlonXP 2400+ to another AthlonXP 2400+ like I mention the different times I saw for it to complete 1M.

Just seeing lower timings based on the same CPU, I just don't seem to understand why that is.
November 8, 2006 6:28:57 AM

Quote:

P4 3.6 GHz is about 41-42 seconds


Really? Wow! I got my Prescott 3.2GHz with an OC to 3.5GHz to get 1M in less than 40 seconds. not bad for an OC n00b. I think i could do a lot better if I had better memory. You just made my day! :D 
a b à CPUs
November 8, 2006 6:38:28 AM

E6600, 2gb XMS6400C5, ASUS P5B Deluxe Wifi

Stock SuperPi = 20 seconds

At 400x8 (3200) i get 15 seconds on stock cooling

My times seem a little better?
November 8, 2006 6:45:29 AM

I think you're missing the point - the real benchmark is not SuperPi, it's a SuperPIE - to the FACE!
November 8, 2006 7:05:09 AM

I like your avatar :wink:

Thats one of my favorite skits of theirs. lol!
November 8, 2006 10:32:18 AM

Quote:
So I guess what I was asking, in comparing a P4 3.0/800 system to another P4 3.0/800 or another P4 system OC to 3.0, and getting different time results, just puts a question in my mind in how that can be? Especially when its a CPU test based on computational power?

As well as comparing AthlonXP 2400+ to another AthlonXP 2400+ like I mention the different times I saw for it to complete 1M.

Just seeing lower timings based on the same CPU, I just don't seem to understand why that is.


SuperPi isn't just a CPU computational test, though primarily it is. It exercises the cache subsystem and a little bit of main RAM - whatever is necessary to figure out Pi to several million digits. Because the XS mod version reports to such precision, even at 1M you'll see performance differences merely by adjusting FSB, RAM timings, background processes, even hard disk file system and fragmentation. But these latter factors won't have nearly the effect of a gross overclock to the CPU.
November 8, 2006 11:32:21 AM

This is my SPi1M score of my overclocked rig:
ASUS A8N5X NF4 A3Rev2 s939 PCIe
Athlon64 s939 Venice Rev E6 2.0GHz 512kB L2
2x512MB DDR-400 CL2.5 4-4-8 CR2 2.5v Patriot
Creative Labs Audigy 2 6.1
Inno3D GeForce 6600 256MB DDR2 350/700 PCIe
2x160GB Maxtor SATA 7200RPM 8MB Cache
1x120GB Maxtor IDE 7200RPM 8MB Cache
LG DVD-ROM 16x
Sony CD-RW 40x
420W Deluxe ATX 120mm low noise
Cooler Master Vortex all cooper 90mm low noise

Overclocked:
HTT 1120MHz 4x280MHz
CPU 2.8GHz 10x280MHz 1.55v
DDR 466MHz 2.5-3-3-6 CR1 2.8v
GPU 525, VideoRAM 830

http://forumz.tomshardware.com/hardware/modules.php?name=Forums&file=viewtopic&t=186085&highlight=venice
November 8, 2006 11:50:47 AM

I'd run SuperPI on my 1.7GHZ Pentium-M, but I don't want it to be calculating PI for the next couple of hours =P

Really need to upgrade...
November 8, 2006 12:13:38 PM

Does your RAM affect SuperPi in any way? (ie. frequency, timings, size etc)
November 8, 2006 12:29:23 PM

I believe yes.
November 8, 2006 12:31:18 PM

Marginally thought, some review sites use SuperPI to test RAM, but like its been said, these are all small things compared to sheer processor power.
November 8, 2006 12:41:49 PM

Maybe you guys can help. Check my specs on the sig. I need a benchmark program to measure my RAM speed/effectiveness. I changed my RAM out of the 1:1 divider (370 FSB, should be DDR740), but I used the 4:5 divider and made it 925Mhz.

I had to loosen my timings to 5-5-5-15 from 4-4-4-12.

I ran 3DMark06 and PCMark05. The results were worse (but just by very little).

Are there better programs to measure whether I should go with tighter timings or higher frequency?

PS. I used Sandra lite, and the mem bandwith was much better. Not sure how to interpret all this though. THANKS!
November 8, 2006 12:44:56 PM

What speed RAM are you running?
November 8, 2006 12:45:56 PM

OCZ Plat Rev 2 sounds like DDR2 800 to me...
November 8, 2006 12:47:58 PM

I have DDR2800Mhz, but OC'ed my chip, and maxed out the FSB at 370. Therefore my RAM was running at the 1:1 ratio of 740Mhz with timings of 4-4-4-12. I then used the 4:5 mem devider, and OC'ed my memory to 925Mhz. I had to loosen the timings to get this right.

I ran PCMark05 and 3DMark06, and the results were slightly less. I need to figure out which one I use (ie. higher frequencies or lower latencies).

Are there any accurate benchmarks out there I can use to figure out what's my best bet here.
November 8, 2006 12:48:05 PM

Then he should be running his FSB at at least 400Mhz.
November 8, 2006 12:51:54 PM

I used Sandra, but it was a litle difficult to interpret. With higher frequencies, my bandwidth was MUCH better. I'll post the results tonight. With PCMark, my score dropped 400 points. Seems odd.

What do you guys do?

NOTE: Cant break 370 on FSB, so forget about that.
November 8, 2006 12:53:30 PM

DDR2 on Intel platform relies on higher bandwith for performance, whereas AMD platform benefits from better timings... I'd run my memory at 5-5-5-15 timings, see how far I can take the FSB, and once I max out then start tightening timings and see how it goes.
November 8, 2006 12:55:09 PM

Well for sure, then the FSB to RAM will be a limiting factor with your tests. But the thing is, do you see a difference in you computers performance. Bungholio Mark will just give you incremental differences.
November 8, 2006 1:00:01 PM

It's hard to tell. I need to try some vid encoding or something. Where does RAM frequency have the most impact?
November 8, 2006 1:08:01 PM

In everything basically. But in your case, with Super Pi
November 8, 2006 1:14:06 PM

Are timings and frequency really that important?

What it really comes down to is the frequency of your CPU. This should be priority number 1 because this is what impacts your overall system performance the most. If you consider lowering your CPU by 75 MHz just so you can run your RAM at a higher frequency, I highly suggest not doing so; it’s not worth it. The best procedure to yield the highest overall system performance is to first, OC your CPU with your RAM at the loosest settings and at a divider, and then once you've reached your CPU max, play with divider and frequency of you RAM. Your RAM timings and frequency don't affect overall performance all that much. The biggest aspect of RAM performance is the capacity; you often need at least 1GB, preferably 2GB.

Got this from THG:

November 8, 2006 2:09:47 PM

Quote:

P4 3.6 GHz is about 41-42 seconds


Really? Wow! I got my Prescott 3.2GHz with an OC to 3.5GHz to get 1M in less than 40 seconds. not bad for an OC n00b. I think i could do a lot better if I had better memory. You just made my day! :D 

I just got 49 Secs with my P4 3.0. (no OC) .5Ghz = 9sec faster. I think there may be other factors then just CPU Power that affect these results even by 1-2 secs (other running tasks, ram..etc.)
November 8, 2006 7:25:42 PM

Quote:
Are timings and frequency really that important?

What it really comes down to is the frequency of your CPU.


This would depend on your application. As you increase the number of digits SuperPi has to calculate, it will bottleneck more and more on your RAM subsystem to a point where it looks like WinRAR. WinRAR is notorious for taking advantage of fast low latency RAM (as much as it likes fast CPUs), even on the Core 2 platform. You'll notice SuperPi is likewise limited by FSB/RAM when in the Task Manager it's failing to use a full core to calculate Pi.

On the other hand, most games and encoding applications are affected by a balance of CPU, RAM, and GPU speeds/latencies. And there are other types of CPU intensive applications where RAM timings play a trivial role.

Additionally, there is a common misconception about Intel's MCH (memory controller hub) and FSB:D RAM ratios. While a 370MHz FSB will not take full advantage of 925MHz RAM (peak RAM bandwidth obviously exceeds peak FSB bandwidth) and additionally may suffer a miniscule performance penalty associated with running asynchronous FSB:RAM, it is also untrue that the additional RAM speed would have no effect because of the bottlenecked FSB. RAM operates on many multi-clock latencies and thus is not always transferring at its peak burst rate. Modern memory hubs buffer RAM data, so faster or lower-latency RAM will always enable the MCH to fill the FSB more fully than otherwise - it's just a matter of how much.

The best way to know the impact of RAM latencies/timings is still to test your specific applications. If the review sites happen to have benchmarked already, then you have your homework done for you.

There are still two general rules. Firstly, you're right that a faster CPU core alone is better than a slower one, and if you have to cut CPU frequency by 5% to extract 5% faster RAM, it's probably not worth it. Secondly, you can simply divide your RAM frequency by your CAS cycles to estimate your latency. Quite a few applications are affected by latency and almost not at all by bandwidth. If you find your latency is equal to or less than with a slower RAM setting, then it is unlikely any applications will slow down as a result of your higher-clock RAM with "looser" timings.

P.S. Your SuperPi times are exemplary. And your RAM latency is the same in your two scenarios, so I'd keep the higher 925MHz rating for those few applications (like SuperPi) which actually need RAM bandwidth.
November 8, 2006 8:13:00 PM

Quote:

To give you some scale on the 1M calculation
E6700 1M is about 20-21 seconds
E6800 1M is about 18-19 seconds
P4 3.6 GHz is about 41-42 seconds
AMD FX-62 is about 33-35 seconds



Hmmm....my E6600 does about 20-21 seconds...(stock) 8)

Intel XBX975X Board
2GB DDR2 - 533mhz @ 4-4-4-12
November 8, 2006 8:23:43 PM

In my experience Pentium m does well running superPi, not that far behind C2D.

I managed to do 1m in 33secs when I first built my system. That was @2.6Ghz, 218Mhz RAM.

Now I get 35secs running 2.5Ghz 208Mhz RAM.

32M completes in 31 mins.

I ran 3 simultaneous SuperPI 32M threads for fun.

The results were interesting in that the last instance finished in around 48 minutes and the first in around 40 minutes, the middle finished in 47 minutes.

Totalling about 135 minutes. That's 41 minutes longer than it would take to run them separately.

I don't currently own a C2D so can't compare. :( 
November 8, 2006 8:56:46 PM

Why PI? Why bother attempting to calculate a number with no end? I would rather see a program that counted to 1 googol. That would be definitive.
November 8, 2006 9:18:34 PM

Quote:
Why PI? Why bother attempting to calculate a number with no end? I would rather see a program that counted to 1 googol. That would be definitive.


What are you talking about?
November 8, 2006 9:19:42 PM

Pi doesnt end.
A googol does. It would tax a system more to count to a googol then to endlessly count out the digits of pi.
November 8, 2006 11:54:50 PM

Quote:
Pi doesnt end.
A googol does. It would tax a system more to count to a googol then to endlessly count out the digits of pi.


It does not count continiously. I counts to 1m,2m....32m...decimal places, right? seem to be a fixed number to me.
November 9, 2006 1:06:56 AM

Quote:
In my experience Pentium m does well running superPi, not that far behind C2D.

I managed to do 1m in 33secs when I first built my system. That was @2.6Ghz, 218Mhz RAM.

Now I get 35secs running 2.5Ghz 208Mhz RAM.

32M completes in 31 mins.

I ran 3 simultaneous SuperPI 32M threads for fun.

The results were interesting in that the last instance finished in around 48 minutes and the first in around 40 minutes, the middle finished in 47 minutes.

Totalling about 135 minutes. That's 41 minutes longer than it would take to run them separately.

I don't currently own a C2D so can't compare. :( 


It seems to me your analysis is a bit off.

31 mins for a single SuperPI process.
running three processes back to back 31x3 = 93 mins

If i understand your post correctly when running the processes concurrently all three processes fininsh in less than 48 minutes.

Speedup = oldtime/newtime = 93/48 = 1.9375
This is the speedup you get when running the three SuperPI processes at the same time. (%increase = 1 - speedup = 94%)

Which means that when running three instances of SuperPI each process runs about twice as fast. This seems a bit strange since you said that you have a pentium M, are you sure you don't have dual cores?
November 9, 2006 12:21:49 PM

Silly me! of course it takes less time.

I was tired at the time of writing.

My system is an Athlon X2.

I have purchased several centrino notebooks from dell and have run SuperPI on them.
November 9, 2006 1:04:02 PM

If anyone does find a link with tests done on the E6600 $ RAM (timings vs frequency), then plz post. Thanks.
!