Could you stress it better?

[Starfishy]

Well we have all been talking about the latest stress test, and we have discussed various issues in a multitude of different threads.

In this thread I would like us to discuss what we would do different if we were the ones performing the stress test.

1. I want to know what hardware you would use? Would you keep the original conformation (was it fair for Intel)?

2. Would you set the default priorities for all the programs the same, is this just to help us AMD fanboys out? Could they even even do this in windows? Should they have done it for both the AMD and the Intel?

3. Should they be running less programs?

4. Is there better programs that are more "equally" advantageous to both of the CPUs? Is there such a thing as a program designed equally for these rigs? Should they just test a broader range?

 

[fishmahn]

1) Original config seemed ok. Whatever I'd use, I'd try to use the same components on both sides.

2) IF it was to be done for AMD, it should also be done for Intel. However, I don't think it is necessary. (see below for more)

3) Yes, and no, and more. (see below)

4) Broader range may be better...

However, I think the purpose of this test was just that - a STRESS TEST. Not a performance test (i.e. the performance graphs should not show how much of anything it does, it should just show 100% cpu usage for X hours). I think the stress part should be revisited when they get the intel rig running properly, and they just run them for a few days.

For performance: Several different performance evaluations should be run. 2 progs, 3 progs, 4 progs, 5 progs, etc. to find out the strengths of both, and at what point does each CPU start to starve a process. Different programs - decoder that favors intel, decoder that favors amd, etc. You could try with and without priority adjustments (might be interesting), with and without HT (which makes the PD-EE just a regular PD, so If you have a 3rd system with just a PD, there's no need), with different programs as the active task, etc.

In addition, showing the # of runs through the program, FPS, etc. isn't all that useful. I would like to see a graph/chart showing % CPU time to each task, and total CPU time spent per task. Then (maybe combined with the # of times through the prog) you can determine that this prog is a hog on one system, or was starved, etc. and make some potentially meaningful comments on where each system has their strong points.

Mike.

 

[P4Man]

I just posted on this in the extremely long thread, but let me recap:

First you have to establish the objective; I assume the goal is to measure reliability of the entire system under extreme circumstances, and not a performance contest. We've seen enough performance tests, we know which cpu is faster for what. So as a stress test, I would:

-Use ECC ram to exclude random memory errors from tainting the picture as much as possible.
-Use more than one system if possible (at least 2 of each)
-Allow fairly high ambient temperature (say 35°C), which is not totally unrealistic in the summer. This is a stress test after all.
- use SLI if possible, but apparently that won't work on the intel system.
- make a huge investment in getting a second watt meter, and measure power consumption for each systems individually
- measure noise if at all possible, or even stream the audio over the web. Use either stock coolers for both chips, or identical coolers from a third party. Ideally, do both.

On the software side: make this as hard as possible for both systems, I would include one 3D game, and as many instances of CPUBurn as there are available (virtual) cores.


= The views stated herein are my personal views, and not necessarily the views of my wife. =