Phoronix IRC Log: 2008-11-06
michaellarabel: Great, Phoenix looks like they'll send out a HyperSpace-enabled PC.
maligor: heh, that looks funky
maligor: I wonder how it handles devices tho
maligor: unless it just drops it once the host os initializes native video drivers
maligor: looks very slick
maligor: I guess it must be running some sort of hypervisor
Kano: michaellarabel: don't you think the bottleneck for u vs os x was the filesystem? ext3 is really slow
Kano: compile time is depending on the compiler, no real test
Kano: at least not between different compilers
Kano: i dont use reiserfs anymore,but you it was already significant faster when you compile a kernel or create squashfs
Kano: maybe check differnet filesystems next..
Milyardo: What file system can both OS's natively use?
Kano: well usually os x uses hfs+
Milyardo: I don't know if OSX can use ext3 though
Milyardo: I know ext2
Kano: linux can read but not write it
Kano: ext4 might be interesting as 2.6.28 will rename it from ext4dev to ext4...
Milyardo: What distro will be first to use ext4 by default?
Kano: i dont think that this will be the default
Kano: but maybe optional
Kano: fedora 10 should be able to install onto ext4 when you specify a special boot option
Kano: -> called ext4 ;)
Kano: michaellarabel: if you test filesystems use a real fast hd, not a laptop...
Kano: the mini mac uses laptop drives too
Deanjo: The HD being a 2.5" vs a 3.5" is irrelevant for a A/B comparison of OS's
Kano: well not fully
Kano: because you usually install os x and linux on different paritions
Kano: and the partitions near the end of the hd are always slower
Kano: and with smaller hds the effect is more extreme
Kano: basically you would have to use the same partition
redeeman: such a test is useless if you don't use same area
Deanjo: You will see that effect on any drive though. It is all relative
Kano: well you can not get the max thruput of a filesystem when the drive limits it
Kano: therefore when you want to compare filesystems use a real fast hd
Kano: when you only get 2.5" benchmarks they are useless
Kano: best would be: use a relative small partition at the beginning of a fast drive
Kano: do you know the windows tool h2benchw?
Deanjo: I agree that the tests should have been ran by a clean wipe of the HD for each OS
Kano: that creates a ps file with all speed ranges
Kano: of a tested hd
Kano: it does not only show lowest / highest / average,but you can see the fastest part of the drive
Kano: of course you need win to run it ;)
Deanjo: All things considered, the article pretty much reflects what is seen as well on a MacPro
Deanjo: (or even a hackintosh)
Kano: dont think so, there you can use nvidia grafics and the nv binary driver for linux are more accellerated
Kano: in my own tests a few y ago about 5% difference to win
Deanjo: The Mac driver is based on the same code
Kano: i dont think so for intel os x + linux
Deanjo: Yes it is
Kano: there is no xserver running
Kano: so it cant be the same code
Deanjo: Same as the code base is shared between Linux and Windows for the nvidia blob
Kano: well parts
Kano: but do you think intel opensource drivers use the same codebase as for the binary blobs?
Kano: i highly doubt so
Kano: because usally you have got the win driver first and much later you get an linux driver.
Kano: for g33 i waited longer than a month
Kano: so the 3d benchmarks just show that the binary driver is faster, thats what i would expect in any way
Deanjo: The results are similar with the nvidia linux blobs and the OS X blobs in 10.5 although to a lesser degree then the intels
Deanjo: The reverse is seen when a ATI card is used
Deanjo: ATI + OSX = not a fun time
Kano: well ati was used in older mac books pro before they switched to nv
Deanjo: Probably on a 10.4 as well I suppose?
Kano: but ati drivers always suck
Kano: with u 8.10 the new oss driver do not flicker that much
Deanjo: ATI cards on Mac suck driver AND hardware wise. They replace those ATI cards in the MacPros like candy
Kano: but with kubuntu you have lost, kde 4 seems to run xrandr every few seconds which makes the picture flicker
Kano: and the binary driver is not really so much better.
Kano: maybe you dont see the flickering when you use a tft, no idea
Kano: but with crt you see it
Deanjo: remembers the guy that was shipped a replacement ATI card only to have DHL leave the package in the dog pen for the customer and the dog damn near choking to death as it chewed up the replacement card
Erektium: open source ati driver works like a dream in ubuntu 8.10
Kano: Erektium: tft maybe
Erektium: not maybe
Kano: and as i said: gnome does not have that much xrandr requests
Kano: but kde 4 has
Kano: you can manually execute xrandr and see if it flickers, just without any argument
Kano: with a good driver it never flickers
Kano: Erektium: and does that happen on your system or not
Erektium: i dunno
Erektium: how do I test it
Kano: open a console window and enter:
Kano: then press enter...
Erektium: I tried that
Kano: and did it flicker?
Erektium: it just gives me resolutions I can use
Kano: thats that it should do, i mean does the screen flicker
Kano: while it is running
Kano: then try with a crt connected
Erektium: I don't understand why it should flicker
Erektium: it's just text
Kano: it calls some functions in the driver which let the picture flicker here
Kano: with RV410 and ati oss driver
Erektium: when I type xrandr it takes a second and then the text pops up
Kano: it was a little wonder that the fixed the dcc detection for rv410 some weeks ago, just in time for u 8.10
Kano: but they did not solve the flickering issue
Kano: that was fixed about 1 month ago
Kano: and the card is over 4 years old!
Kano: do you call that a good driver?
Kano: for me ati is a synonym for crap
Erektium: kano do you happen to know if I must put a gartsize option to xorg.conf?
Erektium: using R430
Kano: i dont use any options when i try the card with u 8.10
Kano: that usually only restricts agp to 1x, but thats not that problematic
Kano: as i use a pci-e cards
Erektium: me too
Kano: one other thing: did somebody notice that when you change the multi for intel cpus in the bios now the current speed is shown correctly?
Kano: before even 6*400 showed up as 3600
Kano: you see 3600 in dmesg however
Kano: when the normal multi was 9
Kano: 2.6.27 seems to fix that issue
Deanjo: Yay! I finally got the new BIOS from Asus. Any bets on if they actually fixed their BIOS faux pa?
Kano: they mailed it?
Deanjo: Emailed it
Deanjo: The screwed up in their latest BIOS releases for my board. AMD-V stays disabled no matter what it is set too
Kano: i had this issue with same beta bios for a a7v board
Kano: well i sold it long ago
Deanjo: Thus why I was having issues running a 64-bit guest in any other virtualization solution other then vmware
Deanjo: It worked in earlier releases. (Which they have for some reason pulled since then from their site)
Deanjo: From what I've been getting back for feedback it pretty much plagues all AMD boards from Asus with a Award bios
Deanjo: Leading me to think that the issue is probably with the reference BIOS
Deanjo: j #opensuse-testing
Deanjo: Heh, kinda nice that SuSE Enterprise Linux 11.0 is finally going to officially support dmraid. It's forcing the opensuse devs to finally address my bug reports on it.
Deanjo: Instead of "use software raid"
redeeman: dmraid is just so ugly
Deanjo: Better then software raid
Deanjo: You read about the issues people are having with the seagate 1.5 TB drives redeeman ?
redeeman: please tell me
redeeman: i was thinking of ordering them
redeeman: dmraid is not better than mdraid :P
Deanjo: Disconnecting on sustained read and writes
redeeman: fucking hell
redeeman: why must they fuck up everything
redeeman: god damn morons
Deanjo: and dmraid is better then mdraid by the simple fact alone that the volume is not limited to use by only 1 OS
Kano: Deanjo: incl. raid 5?
redeeman: mdraid is very stable and robust
Deanjo: Yes including raid 5 Kano
Kano: because you need a kernel patch for the needed module
Deanjo: Yes it is being put into the next release of suse Kano
Kano: well i have got dmraid here
Kano: that my win raid, ich9r
Deanjo: Ya I've got it running smoothly here on Suse 11.0 using Raid 0 and Raid 5 sets at the same time
Kano: all partition created with win
Kano: look the last value
Deanjo: That's the trick, gotta partition it with windows
Kano: 72963 bigger than total size of 72962
Kano: thats stupid
Deanjo: otherwise parted does not mark the volume properly and you get that damn annoying "access beyond device" message
Kano: ich9r seems to be really buggy, only nv raid seems to work better
Kano: i tested also jmicron, had also issues
Kano: with intel same problem with raid 0 or 5
Deanjo: Silicon Image sames to be about at the same level as nvidia
Deanjo: *seems to be
Xemanth: i have weird usb problems with 780g chipset in linux
Kano: well this board has jmircon + ich9r
Xemanth: especially with 8 gb kingston usb memory sticks
Kano: 780g can be combined with sb700 or sb600
Kano: what do you use?
Xemanth: i have sb700
Xemanth: that one
Kano: in theory this should be the better chip
Kano: do you already use latest kernel?
Deanjo: USB and sata on the SB700 did not improve at all on the ATI chipsets
Kano: well you still can buy something with nvidia for amd cpus
Deanjo: Yup, still seems to be the best chipset for AMD cpus
redeeman: well, it would seem 1.5tb seagate is off the table
redeeman: how the hell can they keep selling such crap
redeeman: they should be severely fined for such shit
Kano: you bought one?
redeeman: no but i was gonna buy many
Deanjo: redeeman, t wouldn't be the same if there was one company that didn't pull off major fuckups every once and a while
redeeman: this kind of fuckup needs to cost them alot
Kano: i may buy one tb disk, maybe samsung f1, not fully sure or one of the wd caviar green
Deanjo: Wonder if it is only on the 1.5 TB's or also the smaller drives built with the same setup just with fewer platters.
Kano: dont know which of em would last longer...
Deanjo: No matter what brand, I've had early failures pretty much with every brand.
Deanjo: They all seem to have their "bad series" of drives
redeeman: Deanjo you said seagates china drivers would die within 6 months
redeeman: i've had mine for more than that now, no issues
Kano: well i had bad wd raid hds
Kano: died after 1 year power use
Kano: samsung 500 gb died even after 1 week, maybe overheat, now i added extra fans for hds
Kano: i really dont know if there is a brand which is better than the others...
Kano: i had failures with every brand
Deanjo: That's the bad thing about HD's a lot of the times issues are not known about until a extended period of time has gone by
Deanjo: Deskstars and Fuji's being really bad a few years back
Deanjo: for example
Deanjo: Samsungs back in the 20/30/40 Gig days had bad chips on them (The old freezer trick would usually work enough to grab data off them for recovery)
mcgreg: anyone in here using a radeon 4670? any experience how good it runs with "fglrx" , or "radeon" (stable and git?)
cxo: is really hungry
redeeman: so eat something?
cxo: Good call
cxo: is going to wash the car today before winter comes back
cxo: We had a random snowfall last week, and now its 20degC
cxo: 20degC= 20*1.8 + 32 degF
Erektium: it's -4Celsius here ;S
cxo: haha global warming is on my team this year!
cxo: Yay Obama!
cxo: Or as I like to call him, Mandela 2.0
Xemanth: < Kano> do you already use latest kernel? <- Kubuntu Intrepid 8.10 with kernel 2.6.27
michaellarabel: Creative Gives In, They Open-Source Their X-Fi Driver - http://www.phoronix.com/vr.php?view=13083
GNU\colossus: ha, cool
redeeman: a fat load of good that does
redeeman: they should instead pull some documentation out their asses
kkuno: i have a question on phoronix tests
kkuno: why in the mac os test ubuntu 8.10 has 60 seconds
kkuno: while in the other test it has 116 seconds?
kkuno: they said that ubuntu 8.10 is much slower than 7.x
mha: and I say that all your fail are belong to you.
kkuno: mha, ?
mha: please make your time.
redeeman: find . -iname *.png -print0 | xargs -0 -n 5 -P 4 optipng -zc1-9 -zm1-9 -zs0-3 -f0-5
redeeman: damn sweet
michaellarabel: Intel GEM benchmarks should be out in the near future
GNU\colossus: michaellarabel: 2d or 3d? or both?
maligor: slow as heck with blender ;)
GNU\colossus: michaellarabel: one issue that boggles my mind whenever I view PTS results - it would be MUCH better imho if the charts indicated whether "higher is better" or "lower is better" in a better readable way
GNU\colossus: michaellarabel: is there anything planned in that regard?
michaellarabel: colossus: It says "Low Is Better" on all such tests
michaellarabel: 2D and 3D
GNU\colossus: michaellarabel: yes, I know it does. but it's fscking tiny and non-obvious imho
GNU\colossus: it would be cool if "lower is better" charts would differ from "higher is better"-charts in some way
GNU\colossus: e. g. by distinction in colour
GNU\colossus: (if sth. like that isn't already patented, of course ;))
maligor: I wonder what's with creative and linux drivers, they stall and then they release the sources
cxo: michaellarabel, We really need a Overall chart for performance benchmarks
michaellarabel: For what?
cxo: percentile each chart with the max, and then add up all of them, divide by the number of charts
cxo: for everything, example the osx vs ubuntu you just did, its difficult to see which is faster or which is slower without an overall graph
cxo: i'll pastebin an example
cxo: michaellarabel, http://pastebin.com/m4c2bcabc
cxo: michaellarabel, this way, immediately one can tell A is 30% faster than B, without out needing to eyeball every single metric
cxo: do you capiche?
cxo: Intermediary data is only for people who want to dissect the performance gaps, us n00bs just want to know which is faster overall
Azerthoth: cxo, if you weight the chart higher is better, you would subtract the lower is better scores
cxo: I do not want weighted charts
Azerthoth: then you would have to have higher and lower as seperate metrics instead of a single score
cxo: Nope, you dont
Azerthoth: if lower is better (4) than (9) and you add that to a scale of higher is better, which one gains the most?
Azerthoth: you have to subtract the lower is better score
cxo: Did you even look at my algorithm?
cxo: I think you are confused
cxo: Have a look at what i proposed
Azerthoth: ah I see, your assigning a 'win' value, not the scores
Milyardo: ls -l
cxo: ls -l = ll on redhat/fedora
cxo: and i put `pwd` in $PS1 so i dont need to ever do pwd
redeeman: rpath pwns that
Milyardo: pwd -p ftw
cxo: i also put the time into $PS1