Running The Linux 3.16 Kernel Might Be A Bit Slower On An Ultrabook
Phoronix: Running The Linux 3.16 Kernel Might Be A Bit Slower On An Ultrabook
For those running an Intel ultrabook, here's some benchmarks using the Linux 3.16 kernel on this portable x86 hardware compared to Linux 3.15. Unfortunately, the results aren't too promising...
Does the performance drop have anything to do with the new power management features in the scheduler? If so, check out how the battery life is changed under normal workloads. A minor change in performance is certainly worth it for a laptop on battery power, though not for one plugged into the wall.
That performance drop is so minor and statistically insignificant, I can't see any end user losing sleep over it and skipping a kernel upgrade unless they are just being pedantic.
cpu frequency in 3.16 is way up if idle, backlight keys on my T440p don't work anymore and system won't enter lower package states even if display is completely off. idle drain is now 19 watts from my battery with powertop --auto-tune. nice job!
Are other people affected similarly?
Originally Posted by saski
Well, I said this in other thread, but I guess it could be useful here, too. With kernel 3.16-rc2, my system (AMD A6-3400M, iGPU Radeon HD 6520g + dGPU 6750M) feels slow. I don't think it's the disk, though...
Not just a bit slower on a Lenovo IdeaPad Y500
After seeing extremely poor performance with bcache in its infancy, I swapped the HDD to an SSD in my Lenovo IdeaPad Y500, but made a habit of running phoronix-test-suite benchmark pts/disk when upgrading the kernel, just to see how things went.
Some tests have begun to fail, but the ones that run show pretty grim numbers - especially for 3.16-rc*.
The results are at http://openbenchmarking.org/result/1...SO-Y500DISK045 from 3.12 up to 3.16-rc3.
I am nowhere sure what is happening, but I seem to experience extreme interaction lag with 3.16-rc*, reminiscent of the bad old days where any sort of heavy disk I/O would semi-freeze any interaction with the desktop.
One possibility could be bugs in the new block-mq, but I have neither the time nor the kernel insight to verify this.
Look especially at IOZone, Apache Benchmark and DBench...