Let's benchmark CPUs!

[Vergeltungswaffe]

P4 2.4 XP Pro

1024 PC1066 Rambus

51, 51, 53, 51, 52

[Watson & Crick]

P4 2.8 "hyperthreading" 512 RAM, nvidia 5600 128mb

1 min 15 sec

Interesting that Vergltungswaffe has a faster time, but a "slower" processor than my rig. Video card difference or ram?

[Europa]

Might be the RAM but it might also be because of what programs you run in the background. The video card has nothing to do with it thats for sure.

[Redwolf]

Originally posted by Watson & Crick:

P4 2.8 "hyperthreading" 512 RAM, nvidia 5600 128mb

1 min 15 sec

Interesting that Vergltungswaffe has a faster time, but a "slower" processor than my rig. Video card difference or ram?

Neither can be the reason. And Vergeltungswaffe's system is twice as fast as the 2.4 GHz Pentium with (probably) slow RAM posted earlier. That can't really be. Even though he has much better RAM that would be unlikely to cause a 100% speedup. Maybe the earlier one was a Celeron (probably not).

There must be something I overlooked when I was thinking of this benchmark.

The sound issue might be a clue, maybe CM is having some internal background process shuffling sounds around even when none are hearable during turn computation, trashing the CPU cache and loading the bus to the RAM.

Can't wait to get home to test without sound...

[Berlichtingen]

Originally posted by Panzerman:

Juju, it is generally know that a Mac's speed is faster than a compairable PC, and they may not messure the CPU speeds the same way, although I don't know why. At the peak, I believe CM was useing only 85-100MB of RAM.

Actually, they measure in the exact same way... unfortunately, the terms used are only partial.

###MHz only matters when comparing to like processors (example, 2 different Pentium 4s)

If you compare a P3 to a P4, MHz is largely meaningless. Even more meaningless when comparing G5 to P4 because the processor archetecture is completely different.

MHz measures numbers of processor cycles per second IIRC. Each processor has a set number of processes per cycle. So, if you have a processor that does 2 processes per cycle (low number for example purposes) at 20MHz (40 processes per second) and one that does 10 processes per cycle at 10MHz (100 processes per second) its obvious which is faster. However, computer manufactures only tell you the MHz

[Berlichtingen]

Oh, also, cache size, bus speed and bus archetechure will also effect the computer's overall speed

[Redwolf]

The usual speed measure for CPUs, at least outside gaming circles, is the "SPECmark" suite.

http://www.specbench.org/

http://www.specbench.org/cpu2000/results/cint2000.html

http://www.specbench.org/cpu2000/results/cfp2000.html

The clock speed is almost irrelevant and the Pentium-4 is the best example.

[Europa]

Slow RAM (including its bus architecture) makes a big difference in a computer since the processor will have to run empty cycles waiting for the RAM to respond. If you want to test the speed of your processor only write a little program in a language of choice that does 10 different mathematical calculations 10000 times or so.

[DesertFox]

non-OCed: EPOX EP8-RDA3+ / XP-Pro-SP2 / Barton 1.833Mhz (XP 2500+)/ FSB166 / 1024MB (2x512 dualmode) DDR333 CL2.5 / Nvidia 5950 Ultra 256MB

Average out of 5 runs: 58 sec

OCed: EPOX EP8-RDA3+ / XP-Pro-SP2 / Barton 2.200Mhz (XP 3200+) / FSB 200 / 1024MB (2x512 dualmode) DDR333 CL2.5 / Nvidia 5950 Ultra 256MB

Average out of 5 runs: 51 sec

cheers

Helge

[i.i.koshkin]

AMD/64 3800+

1 Gig PC3200 memory

49 seconds

[Pzman]

Originally posted by Berlichtingen:

Oh, also, cache size, bus speed and bus archetechure will also effect the computer's overall speed

Hmm, the real question is what is standard when it comes to a cache? Mine has L2 at 256k and a L3 at 2MB. I also know that I have 100Mhz Bus speed. It all sounds nice, but I don't really understand what that all means.

When people say slow RAM, what is that? Is SDRAM slow? (I have four 128mb SDRAM chips).

[Watson & Crick]

Originally posted by Watson & Crick:

P4 2.8 "hyperthreading" 512 RAM, nvidia 5600 128mb

1 min 15 sec

Interesting that Vergltungswaffe has a faster time, but a "slower" processor than my rig. Video card difference or ram?

I forgot that I had Outlook Express, PBM helper and Norton Antivirus "open." I reran the test and came up with the same result. I also have PC3200 RAM. :confused: :confused:

[von Lucke]

P4 2.4C (Northwood), L2 cache 512

512 PC2700 RAM

Win98 SE

5 runs averaged out at exactly 60 seconds.

[Redwolf]

OK, sound does play a role :mad: I forgot that the ambient sounds are getting played. I edited the instructions to say to turn sound off.

My times drop by 5-8 seconds without sound.

This can cause great variance for the same CPU as different soundcards are putting more or less load on the CPU. Some soundcards are really bad, I assume that explains some of the crude results we have seen.

This is also a good tip for people playing big scenarios on slow computers - turn off ambient sound before hitting "GO", turn back on when action starts.

[poppy]

Be sure to run a "spyware" or something similar to get rid of the trash before you start the test.

[86smopuim]

do you want more times?

I have a few rigs laying about, most older.

[Redwolf]

86, a few Celeron times would be nice to have.

BTW, if people want to know how much memory an application uses, you can run the system monitor in the background, set the graph so that it displays at least the last 5 minutes or so, then fire up CM and the test, iconize CM and watch the graph. Several monitors work better, of course.

The system monitor can also be used to see whether some spyware or other unwanted program sucks up CPU time.

[Daman324]

AMD Athlon XP 3200+ (Barton 400mhz bus), 1024 MB of PC3200 EL/DC RAM

I ran three tests and got an average of about 46 Seconds.

[Juju]

Here's an interesting one:

At first I had these times on my P-IV 2.4 on W2K: 2:03 (2x), 1:54 (2x) and 1:59.

But then, in order to play System Shock II on my rig I was forced to disable hyperthreading.

After disabling it I ran two more tests: 1:35 and 1:33. That's a 25 to 30 second gain! Can you explain that, Redwolf?

[Juju]

Originally posted by Redwolf:

This can cause great variance for the same CPU as different soundcards are putting more or less load on the CPU. Some soundcards are really bad, I assume that explains some of the crude results we have seen.

Aha! Last year -during a big desperate troubleshooting session to get some game to work properly- I ripped out my SB Audigy to try the onboard soundcard (C-Media something). Never put it back since then. D'you think it'd make a big difference if I put it back??

[Redwolf]

Originally posted by Juju:

Here's an interesting one:

At first I had these times on my P-IV 2.4 on W2K: 2:03 (2x), 1:54 (2x) and 1:59.

But then, in order to play System Shock II on my rig I was forced to disable hyperthreading.

After disabling it I ran two more tests: 1:35 and 1:33. That's a 25 to 30 second gain! Can you explain that, Redwolf?

Well, you know what hyperthreading is, do you? It is like a dual-processor system, except the second CPU exists only partly, other parts are shared between the two virtual "CPUs". A performance gain will result when two programs run at the same time and both programs are written in such a way that they don't make optimal use of the processor (if you want more into I can give the saw example I once posted on GF I think). Since very few programs make optimal use of the P4 this is a smart idea, in theory.

That means that you can never be faster with hyperthreading than without unless you run two programs or one program doing its work in two (or more) threads (concurrent program flows within one program). Obviously CM is single-process, single-threaded, so you can never see a speedup.

But you shouldn't see a slowdown either, because if only CM runs then one of the "CPUs" should be idle and not disturb the other. The other virtual CPU should run just like one physical CPU.

If it doesn't then that leads me to believe that you either have

1) a program running in the background drawing CPU power

2) Win2K is so dumb to do graphics or sound updates or other homework in a different thread which will end up munching on the second virtual CPU. That would be very dumb if the graphics and sound updates are for the same program

3) the dual-processor core of Win2K might be much less efficient than the single-processor core.

4) other Win2K braindamage

You can check 1) with the system monitor, the CPU graph. It should show nobody taking up CPU while you don't do anything on the computer.

I have a hyperthreading P4 but only used it on FreeBSD, which was showing no such problems. My tests showed hyperthreading to be a disappointment, while I wasn't seeing slowdowns I had a very hard time to gain advantages as well.

[Redwolf]

Originally posted by Juju:

</font><blockquote>quote:</font><hr />Originally posted by Redwolf:

This can cause great variance for the same CPU as different soundcards are putting more or less load on the CPU. Some soundcards are really bad, I assume that explains some of the crude results we have seen.

Aha! Last year -during a big desperate troubleshooting session to get some game to work properly- I ripped out my SB Audigy to try the onboard soundcard (C-Media something). Never put it back since then. D'you think it'd make a big difference if I put it back?? </font>

[Juju]

Originally posted by Redwolf:

Originally posted by Redwolf:

I assume you have a different card now? Or do you use onboard sound or USB sound?

It would be great to see if and how the results with sound on change depending on different models of soundcards.

I have been using the AC-97 (C-Media) on-board sound card that's on my mobo. I just now re-installed my SB Audigy and did 3 further test runs.

With sound: 1:26 and 1:33

Without sound: 1:28

Thanks for all the hyperthreading info. I believe it's best if I just leave the whole thing off. It isn't even intended to work properly on W2K, so that probably has at least something to do with it.

[Redwolf]

Originally posted by Juju:

I have been using the AC-97 (C-Media) on-board sound card that's on my mobo. I just now re-installed my SB Audigy and did 3 further test runs.

With sound: 1:26 and 1:33

Without sound: 1:28

That means your Audigy is a little faster than the onboard? Did you actually use the Audigy or was it just present?

[Watson & Crick]

Originally posted by Redwolf:

</font><blockquote>quote:</font><hr />Originally posted by Juju:

Here's an interesting one:

At first I had these times on my P-IV 2.4 on W2K: 2:03 (2x), 1:54 (2x) and 1:59.

But then, in order to play System Shock II on my rig I was forced to disable hyperthreading.

After disabling it I ran two more tests: 1:35 and 1:33. That's a 25 to 30 second gain! Can you explain that, Redwolf?

Well, you know what hyperthreading is, do you? It is like a dual-processor system, except the second CPU exists only partly, other parts are shared between the two virtual "CPUs". A performance gain will result when two programs run at the same time and both programs are written in such a way that they don't make optimal use of the processor (if you want more into I can give the saw example I once posted on GF I think). Since very few programs make optimal use of the P4 this is a smart idea, in theory.

That means that you can never be faster with hyperthreading than without unless you run two programs or one program doing its work in two (or more) threads (concurrent program flows within one program). Obviously CM is single-process, single-threaded, so you can never see a speedup.

But you shouldn't see a slowdown either, because if only CM runs then one of the "CPUs" should be idle and not disturb the other. The other virtual CPU should run just like one physical CPU.

If it doesn't then that leads me to believe that you either have

1) a program running in the background drawing CPU power

2) Win2K is so dumb to do graphics or sound updates or other homework in a different thread which will end up munching on the second virtual CPU. That would be very dumb if the graphics and sound updates are for the same program

3) the dual-processor core of Win2K might be much less efficient than the single-processor core.

4) other Win2K braindamage

You can check 1) with the system monitor, the CPU graph. It should show nobody taking up CPU while you don't do anything on the computer.

I have a hyperthreading P4 but only used it on FreeBSD, which was showing no such problems. My tests showed hyperthreading to be a disappointment, while I wasn't seeing slowdowns I had a very hard time to gain advantages as well. </font>