Benchmark & CoolBook

[richard67]

Hi, I was wondering, why the MBA is getting such bad benchmark results in multi cpu tests, therefore i did some experiments with CoolBook on my 1.6 GHz MBA 80G...

For stressing the CPUs i used CPUTest (download available at http://www.coolbook.se) with the following setting:
Test type: huge, Repetitions: 1 Instances: 2

To limit the fan speed to 4000 RPM I used the smc CLI tool from smcFanControl...

Here are the results:

Num, Throtteling, max Fan RPM, Minutes (less is better), Remarks
01. OS, 6200, 05:03
02. OS, 4000, 05:33
03. CB, 6200, 04:44, default voltage
04. CB, 4000, 04:47, default voltage
05. CB, 6200, 04:00, save voltage
06. CB, 4000, 04:15, save voltage
07. CB, 6200, 04:00, minimum voltage
08. CB, 6200, 03:39, minimum voltage only 1200, 1400, 1600 MHz

Try those at your own risk!
09. CB, 6200, 03:24, minimum Voltage, only 1400MHz, 1600MHz, temp >95C
10. CB, 6200, 03:24, 1600MHz fixed, temp >100C OS throtteling kicked in

1. - 2. Deactivated CoolBook, using the OS throtteling.
3. - 4. Active CoolBook using the default voltage and MHz
5. - 6. Active CoolBook using:
800 MHz, 9.000 V
1200 MHz, 9.125 V
1400 MHz, 9.250 V
1600 MHz, 9.375 V
7. Active CoolBook using the minimum stable voltage on my machine:
800 MHz, 9.000 V
1200 MHz, 9.000 V
1400 MHz, 9.000 V
1600 MHz, 9.125 V

8. Active CoolBook using:
1200 MHz, 9.000 V
1400 MHz, 9.000 V
1600 MHz, 9.125 V

Conclusion: Apple did a really bad job in implementing the CPU throtteling. Even with limited fan speed and the original voltage the CoolBook throtteling algorithm is far better. Not to mention the results using undervolting...

Cheers, Richard

P.S: I would love to see more results from other MBAs...

 

[wordy]

Pardon my ignorance, but I don't really understand where you're getting the "minutes" values. If you can explain that, I'll be glad to post my results.

My minimum stable voltages at 800, 1200, 1400 & 1600 are all 0.9V (tested for 20 mins each, using same CPU Test settings you used).

 

[stakis]

Pardon my ignorance, but I don't really understand where you're getting the "minutes" values. If you can explain that, I'll be glad to post my results.

My minimum stable voltages at 800, 1200, 1400 & 1600 are all 0.9V (tested for 20 mins each, using same CPU Test settings you used).

I think what the OP is simply timing the test and finding that under his coolbook settings the same test completes at a faster time..

correct me if I'm wrong.

I would be interesting to run Xbench or something to see what the differences are there...

Very interesting post though...

 

[richard67]

Pardon my ignorance, but I don't really understand where you're getting the "minutes" values. If you can explain that, I'll be glad to post my results.

On the Main tab of CPUTest at the lower right corner you will find "Elapsed Time". It does show the time needed to finish the choosen test. That gives a good benchmark & stability test in one...

My settings have been:
Test type: huge
Repetitions: 1
Instances: 2

Cheers, Richard

 

[richard67]

I think what the OP is simply timing the test and finding that under his coolbook settings the same test completes at a faster time..

correct me if I'm wrong.

Perfectly correct, but one doesn't have to do the timing manually, CPUTest does that for you...

I have chosen CPUTest over XBench because it gives a simple solution to saturate all CPUs for a longer period...

The resulting differences in speed are huge. Using the original Apple throtteling the test does take about 5 minutes, with maximum tweaking in CoolBook it came down to about 3:30 minutes on my machine. This is consistent with the result i've got running other "heavy calculation" apps like handbrake and such...

Cheers, Richard

 

[aveda6]

Richard,

Using the voltage settings recommended on the main coolbook thread, I ran the test you mentioned (I have the 1.6 ghz Air with 80gb hd). I had a result of 03:23 and my CPU ran from 40C at idle (before the test) to 85C during the test. This was with coolbook set to high throttle with a max temp of 85C and the following voltages:

800mhz - 0.9000
1200mhz - 0.9125
1400mhz - 0.9250
1600mhz - 0.9375

 

[richard67]

I had a result of 03:23 and my CPU ran from 40C at idle (before the test) to 85C during the test.

Thanks for the test! Your machine does for sure run a lot cooler than my MBA. Could you run the test again without CoolBook active?

Did you do additional tweaking like changing the thermal paste or something similar?

Cheers, Richard

 

[aveda6]

Okay, today proved a bit different in terms of how the coolbook worked. During routine surfing of the net and allowing time machine to back up the system, my temps were running in the low 70s/high 60s. Quite a bit higher than yesterday. I have not redone the thermal paste yet, although I may do so if the temps continue to run high.

I reran the CPUtest with coolbook off and on (with no other programs actively running).

Off: Temps 83C
Fan 6200rpm during entire test
Elapsed time 06:22

On: Temp started at 48 and ended at 86
Fan started at 2500 but maxed out around test 6 to 6200
Elapsed time 03:33

Now, a few minutes after running the test with coolbook on, the temp is 49C with the fan slowly reducing (around 5900rpm and very slowly dropping)

 

[stakis]

To limit the fan speed to 4000 RPM I used the smc CLI tool from smcFanControl...

What are the Advantages/Disadvantages of limiting the fan speed for a test like this?

 

[richard67]

What are the Advantages/Disadvantages of limiting the fan speed for a test like this?

The idea is to find out the best values for a good balance between fan noise, calculation speed and temperature under heavy load...

For example, with active CoolBook and limited fan speed, my machine runs silent, a lot cooler cooler and even faster compared to the original Apple throtteling and voltage...

So far it looks like limiting the fan speed to reduce the noise does have surprisingly little impact to the machine speed and temperature. It might be a useful and save option for daily work...

Cheers, Richard

 

[richard67]

I reran the CPUtest with coolbook off and on (with no other programs actively running).

Great, thanks a lot!

Wow, what a difference! Looks like every MBAs has it's own thermal behavior...

I am very surprised to see your 6 minutes versus my 5 minutes without CoolBook. On the other side your machine is faster and cooler with CoolBook...

I hope we will see more results from other machines to get a better picture...

Cheers, Richard

 

[n0de]

I am not questioning the results posted here, but geekbench tests show little or no difference between coolbook on or off.

Anyone here a coder who can run a big compile and see how long it takes with and without coolbook?

 

[aveda6]

I reran the CPU test with coolbook both on and off given that my initial values with coolbook off were so high. This time the values seemed to make more sense.

Off
time 03:30

On
time 03:29

So the times are much more close than previous. I'm thinking I may have had a program running that I was unaware of. Time machine seems to be always running so I downloaded a manager for it to limit how often it runs.

There is a definite difference in the overall temps of the system however. I am now consistently running around 39 - 45C instead of mid 60's up to 80's before coolbook.

jim

 

[n0de]

Coolbook is good for something and in the end it does do what it says. It's just not for performance improvement.

I do consider it money well spent.