well, its not much. but i spent the last couple days compiling a new kernel, and tweaking, and trimming and all that good stuff. this is actually my first, fully functional, custom kernel. every time i tried before, they either ended up slower, or they were missin something, or just plain didn't work, until my frustrations drove me to quit. this time i pulled through. the other thing may have motivated me, was the eee PC 900A i'm getting for my bday soon, i'm going to run arch linux on it, but with a 4g SSD, i'm going to need every bit of space i can get, so i figured i'd need to get this right soon, little practice makes perfect
for anyone interested. i'm now running a 2.6.32 rc3 kernel with the ZEN patch which adds the BFS scheduler by defaul, and some other changes, such as the option to optimize for K10, and some added smp features compared to the current stock kernel. the other big one for me right now, is the new phc-k8 driver for frequency scaling, the new driver compared to the old powernow_k8 driver, does voltage scaling as well. the new driver also seems to be much more responsive. it switches frequencies alot faster
also, for anyone interested. i think what really helped me this time as far as building it. was this
link from a gentoo user, who provides bare minimum configs (seeds) and some general info how to find out what modules you need. excellent starting point. i also cheated and used the xconfig, rather than menuconfig.......its basically the difference between dos and windows, gui vs cli.......the gui made it much easier, since it allowed searching options to make it easier to find what you need, and there's a help/info pane on the modules, which makes it easier as well to know what you need and don't need
but anyway, without going in too much detail, benchmark scores
2.6.30 Arch kernel
2.6.31 Arch kernel
2.6.32-rc3 zen kernel
big difference, some might notice the different cpu speeds reported in the tests. the first one, on the old 2.6.30 kernel, with the old powernow_k8 driver says it was 800mhz. during that test though, that was the second test run. it was scaling for the first test, for the second test i manually set the governor to performance, so it was running 3200mhz, conky was reporting the right speed, but not geekbench for some reason, its not the only app that didin't seem to recheck the speed properly while frequency scaling was on though. proper k10 support lacked in that kernel too, so i don't know. but if you look at the single threaded tests on all the scores are in line, I'm pretty sure that wouldn't be if it was running that much slower
in the new test, cpu scaling was actually still set to ondemand, so scaling was on, and i still got a higher score. but look carefully at the multithreaded tests
looks like some of the architecture settings in the new kernel paid off
edit// someone on the arch linux board pointed out the issue with reported cpu speed, which i already know. but also pointed out, the number of threads is specified on the new score, and not on the old. so that had me wondering.......i was running on the default 2.6.30 archlinux kernel when it was run, and 2.6.31 is out now. so i tried with the latest stable arch linux kernel. and the score went up a tad, not much, but a little, and it reports 4 threads now.
some linux BM's
.
.
Linear Mode
