Sporadic Behavior With Linux 3.16 vs. 3.17 vs. 3.18 Test Results

[phoronix]

Phoronix: Sporadic Behavior With Linux 3.16 vs. 3.17 vs. 3.18 Test Results

Here's some more Linux 3.18 kernel benchmarks I did with the final release due out likely today. However, these results seem to be a bit odd...

http://www.phoronix.com/vr.php?view=MTg1NjM

[gens]

2-3x improvement on some tests is a lot odd
must be some bug fix or something (pstate?)

[dungeon]

To me all looking fine there on 3.18 kernel, just nouveau regressed a lot And also it looks to me that CPU does not like earlier kernels at all for some reason.

'Timed Linux Kernel Compilation' need 6 minutes with 3.16/3.17 kernels ... that sound right only for low power devices.

[chrisb]

2-3x improvement on some tests is a lot odd
must be some bug fix or something (pstate?)

That would be my guess too. Maybe the ASRock BIOS is setting a lower p-state default for some cores, and for some reason the kernel isn't overriding it until 3.18? Or perhaps some of the cores are disabled, the results from <3.18 look like what you might get with only one or two cores enabled. Check the BIOS is the most recent version and in settings has all cores active and at same CPU core ratio and Turbo enabled. You could try plotting the frequency of each core when running the benchmarks, I'm guessing it would show that some of the cores are either disabled or underclocked.

[Gloomwing]

Hi!

I'm feeling that there is some regression on performance for me on 3.18.
But how can i test it (quickly)?

I'm on Samsung NP535U3C - AMD 4355M APU.

Regards!

[chrisb]

Hi!

I'm feeling that there is some regression on performance for me on 3.18.
But how can i test it (quickly)?

I'm on Samsung NP535U3C - AMD 4355M APU.

Regards!

If only there were some kind of Test Suite you could use...

[Gloomwing]

As always on such forums. There is someone who is smarter than anyone.

Check this one: https://www.google.pl/webhp?sourceid...0on%20phoronix

And if there is anyone who can ANSWER question which i ASKED...

[toyotabedzrock]

It is the pstate driver. Iirc you tested CPU throttle drivers and one of the oddities I remember seeing with pstate was much faster compile times with GCC.

[chrisb]

As always on such forums. There is someone who is smarter than anyone.

Check this one: https://www.google.pl/webhp?sourceid...0on%20phoronix

And if there is anyone who can ANSWER question which i ASKED...

The test would depend on what you thought had regressed? ie. if you think CPU is slower, then choose one of the computational benchmarks, if you think 3D is slower, then choose one of the game benchmarks etc. Then run the same benchmark under both kernel versions and compare the results.