Phoronix IRC Log: 2008-10-16
Milyardo: Good Morning #phoronix
Kano: hi, did anybody collect the flash 10 beta versions?
Milyardo: Collect? For what purpose Kano?
Kano: i need some older ones which dont require glibc24
Kano: i only found version 9 in an archive zip,but there is none for betas
cxo: anyone got the 8.10 ati driver working on 2.6.27 x86_64?
Kano: no problem the 8-9 patch works
maligor: does it complain about the protos?
maligor: or whatever they were called :P
maligor: the smp thing
Kano: cxo: do you need xserver 1.5 support?
cxo: complains about 2 or 3 functions
maligor: switched to 1.5 myself just :P
cxo: smp is one of them and another cpu_ something and another one
cxo: i tried that patched version from that ukranian guy, that just reduced the errors to 2
Kano: btw. i made some instructions for installing 8.543 on other distros than ubuntu 8.10 with xserver 1.5, if somebody is bored and wants to try it
cxo: that has been patched for 2.6.27?
maligor: I just modified the dependencies a bit and installed it on debian
Kano: well 8.543 is for 2.6.27 + xserver 1.5 only
Kano: maligor: just install dkms and the ubuntu packages work
cxo: it doesnt build on vanilla 2.6.27
Kano: cxo: for xserver 1.5 or older? which distor?
maligor: no, I removed the fglrx-kernel dependency from the driver
maligor: then I don't need dkms
cxo: fc9 64bit
Kano: cxo: follow my hack instructions, pure 8-10 will never work
maligor: yeah, you'll need the modules from ubuntu
cxo: i can patch up the changes myself, but i'm trying to eat breakfast here
cxo: thought someone would have done it by now
Kano: you have use the files from the orig file
Kano: using a 2.6.27 patch alone will not fix your problem
cxo: k i'll try it out
Kano: not fully sure if you can use 8.543 as option, maybe you need 8.542, just try it a bit
cxo: you need to do line by line commands, some people cant use the keyboard you know
Kano: i will not write a script for that, you can use for example mc, thats easy enough
Kano: maybe 8.542 is better, will replace it
cxo: aesh you are making me work
Kano: basically it should not matter that much
cxo: almost worked
cxo: the driver built, but the rpm packaging script failed
Kano: if somebody needs it for gentoo the changes to an ebuild would be trivial, as you can skip the extraction and just have to create a common dir + put the 4 dirs in there which are usually there
Kano: try 8.542 not 8.543 as option if you extracted all to correct place
Kano: did you create the common dir?
cxo: yeah yeah i did your instructions
Kano: well i can not try on fc9
Kano: just try abit
cxo: there is no X11R6/lib folder at all
Kano: you will find it *g*
Kano: or you have to compare a bit more to the original 8-10 extraction
Kano: but not much i think
cxo: ah it needs to be lib64
cxo: stupid build script
EisNerd: I have a big problem
EisNerd: I need to get one hw accelerated framebuffer over 4 displays
EisNerd: each display should be rotated and has a resolution of 1920x1080
EisNerd: so that the frame buffer has in complete 4320x1920 pixel
EisNerd: currently we have a nvidia quadro fx 4500 X2
EisNerd: which seems to be limited in hw acceleration to 4096 pixel width
EisNerd: the station should be able to run x86_64 gentoo linux and optional windows vista
EisNerd: so if some one could give me a usefull xorg.conf, hints, more background or a well explenation where the problem with the current graphics board is it qould be great
EisNerd: expecially it would be interesting if amds multi view could do this with two graphic boards
copper: Deanjo michaellarabel : Tom's Hardware OpenSSL Part Deux
Deanjo: Oh what now
copper: you were fucking right
copper: they used a single binary for all CPUs, with *no* optimization, and the same environment
Deanjo: lol geee that sort of defeats the whole purpose now doesn't it?
copper: they justify it by saying that Joe six-pack will use apt-get and not a self-compiled binary, that it allows them to compare CPUs accurately, and that their results for the C2D E7200, E6850 and E6600 are close to those reported on PTS Global
Deanjo: What a bunch of horse shit
copper: I don't even know where to begin
Deanjo: They might as well use glxgears to benchmark video cards too
Deanjo: This basically means all their linux benchmarks are useless
Deanjo: Maybe somebody should clue them in that "Joe Sixpack" couldn't even tell you what openSSL is therefore why even use it as a benchmark
Deanjo: Why even run benchmarks on video drivers then as "Joe Sixpack" runs whatever MS shoves to them via software updates
Deanjo: It's not even benching the CPU per say then but more the distro
copper: I suspect they were just lazy as fuck.
Deanjo: probably was 32 bit as well
michaellarabel: Hmm, I have had an idea for a way to prevent that from occurring.
Deanjo: Burning down Tom's offices?
michaellarabel: Just if PTS detects the CPU/motherboard/GCC has been changed since the test was installed, force a reinstall
Deanjo: There is a great idea
Deanjo: (although I admit I'm still partial to the burning down of their offices as well)
copper: michaellarabel: I told them they shouldn't mention the name of Phoronix Test Suite, since they completely deviate from its intended mode of operation. I suggest you ask them yourself - it gives PTS a bad name, IMO.
copper: especially since their explanation was posted only in the french forum
Deanjo: thinks some sort of validator is needed for PTS as well
cxo: got the sucker
cxo: built catalyst 8.10 on fc9 x86_64
cxo: now to try this thing out
copper: if you were worried about cheating, who would have thought the nastier cheater of all would be Tom's Hardware…
Samuel-NotAFK: copper: Every-one knows that.
copper: knows what? That TH cheats?
Samuel-NotAFK: copper: Yeah!
Samuel-NotAFK: copper: You're telling me you didn't know?
copper: what's more disturbing is that they even try to justify it
copper: afk for a bit
Deanjo: I wouldn't say cheating, just piss ass poor benchmarking practives
Deanjo: An article on Phononix would be nice to expose their fraud so the story could be dugg
michaellarabel: Deanjo: Phoronix has been banned from Digg for whatever reason basically. No article has got to the frontpage in months even when there's been several hundred Diggs.
Deanjo: Time to sick the slashdot crew on digg now too I guess
Deanjo: wouldn't doubt Epic had something to do with digg
copper: someone call the New York Effing Times, yo.
Deanjo: heh, if I was still at apple I would pull up Roses contact info and screa,
Deanjo: 10,000 sms messages screaming "Unban Phoronix you bastard!!!"
copper: *something* should be done
Deanjo: Probably was an effect of Jade trying to Digg his forum posts.
copper: it annoys the hell out of me that TH remains basically the main online publication as far as benchmarking goes
Deanjo: What annoys the hell out of me is this shit http://www.theinquirer.net/gb/inquirer/news/2008/10/16/steve-jobs-quit-apple
copper: at least everyone knows The Inquirer is a tabloid
Deanjo: The inquirer just can't help spreading unsubstantiated FUD
Deanjo: pities the Executive Relations crew at Apple whenever crap like that is posted.
copper: hey, even Bloomberg announced Steve's death
copper: (it was Bloomber, right?)
Deanjo: Ya and you saw what crap like that does to stock shares
michaellarabel: Deanjo: At least with The Inquirer's writer's new site, they acknowledge their information isn't always the truth
Deanjo: I really don't have any doubt in my mind that the Inq is simply trying to cause Apple stock to take a momentary hit with crap like that.
Deanjo: Which site is that michaellarabel ? Fudzilla?
michaellarabel: Deanjo: It's not publicly launched yet :)
copper: Fudzilla isn't much better than Tom's Hardware
Deanjo: IIRC Fudzilla was founded by a Inq writer as well (may have been the register though)
copper: and neither Fudzilla or Tom's Hardware seem to read any other publication than their own
copper: any publication other than their own*
Samuel-NotAFK: silentpcreview FTW
copper: I like The Tech Report and Ars Technica so far
Deanjo: Well Fudzilla seems to be a better one with tech news.
copper: TR reviews are quite complete, and Ars Technica is the one who found out about PCMark's cheat
Samuel-NotAFK: Gordon-: Brown?
Deanjo: I usually hit up the Tech Report for their Daily Bread
copper: yeah, me too
Deanjo: They keep posting articles from that Phoronix site though :P
Samuel-NotAFK: I wish Phoronix would do ETQW ATI 4670 benchmarks.
Samuel-NotAFK: I love ETQW and I'm thinking about getting that card.
Deanjo: I believe there are issues with the drivers and ETQW at present Samuel-NotAFK
Samuel-NotAFK: Deanjo: That would explain the lack of benchmarks then.
Deanjo: I think it was even mentioned in one of the last radeon reviews
Deanjo: Ya it was "Enemy Territory: Quake Wars results could not be published in this article due to regressions within the Catalyst (fglrx) Linux driver."
Samuel-NotAFK: Deanjo: I didn't see that in the 4670 review.
Deanjo: Right under the screenshot
Samuel-NotAFK: nevers reads page 5
Deanjo: "Statistics show that 78% of people will read page 4 and 6 but only 13% will read a page 5"
Deanjo: beats on the Seti@home server. Take my completed units dammit!!
Samuel-NotAFK: Deanjo: Lol really?
Deanjo: 83% of all statistics are made up
Samuel-NotAFK: Deanjo: Lol really?
Deanjo: Ya and believe it or not, did you know that if your biological parents couldn't have kids then you couldn't too?
maligor: Deanjo, that's made up, it's 86,2% really
maligor: or that the probability of surviving alive is 0
Deanjo: Ya it's a amazing stat
Deanjo: giggles thinking how many times he confused people with the parents trivia
Samuel-NotAFK: Deanjo: It's pretty obvious.
Samuel-NotAFK: Deanjo: If my biological parents couldn't have children then i wouldn't be alive thus wouldn't be able to make children.
Deanjo: Walk into a room and say that and I guarantee 3 or 4 out of 10 won't get it
Samuel-NotAFK: Deanjo: I got it straight away.
Samuel-NotAFK: Deanjo: Another made up statistic?
Deanjo: Actually that one is pretty true lol
Samuel-NotAFK: Deanjo: Lol ^^
Deanjo: Human nature is to react with a response before actually thinking about it
Deanjo: It's the whole "Say silk." and then follow it up with "What do cows drink?"
Deanjo: Hell, "How many eggs a day can a rooster lay?" still get people
Samuel-NotAFK: roosters lay gold, silly
Deanjo: I wish they laid full cases of beer
Deanjo: *sigh* openSUSE 11.1 beta 3 will be delayed by several days. We originally scheduled the release for today (October 16) but the power outage last Friday left us unable to check in packages temporarily, and pushed the schedule back by several days.
Samuel-NotAFK: OpenSUSE is too unstable anyway
Samuel-NotAFK: I always had issues with it.
Samuel-NotAFK: I liked it a lot but it was too annoying.
Deanjo: I find the exact opposite
Samuel-NotAFK: Well I haven't used it for about a year now.
Samuel-NotAFK: 10.1 10.2
Samuel-NotAFK: Both had annoying bugs
maligor: Bug #1) Not Debian :P
Deanjo: Got servers across the world running rock solid on it with less then 1 hours downtime in 2 years (downtime for kernel updates and such)
Deanjo: And most of those are still running 10.2
Deanjo: Pre 10.3 the most annoying thing was their package manager but that got fixed when zypper came out
Ivanovic: michaellarabel: uhm, 8.10 is officially out: http://ati.amd.com/support/drivers/linux64/linux64-radeon.html
Ivanovic: posted already yesterday
Kano: Ivanovic: but the changelog is empty ;)
Ivanovic: Kano: do you really trust the ati changelogs?
Ivanovic: those that only list 5% of the actual changes?
Kano: well i only discovered one etch fix
Kano: with correct permissions now
Kano: 8-9 had a dri permission problem
Kano: but basically 8-10 is unusable on etch, broken beryl, broken wine
michaellarabel: Deanjo: Commited to git: pts-core: Track motherboard, processor, OS, and compiler version when a test is installed, and if any one ends up getting changed, force re-install of that test
Samuel-NotAFK: Ivanovic: Does ETQW work with the ATI 4670 card now?
Ivanovic: Samuel-NotAFK: no idea
maligor: the 8-10:s have myserious kcl-*.c/h in the drm module wrapper too
Deanjo: Nice commit michaellarabel
Deanjo: By doing this though, PTS should default to keeping the installation packages in the cache, that way moving a hd to another system will result in it not trying to redownload all the packages again.
Deanjo: Maybe a simple question upon first run of the PTS should ask if a download cache should be made with the default set to yes
copper: you should add a EULA denying Tom's Hardware the use of PTS :P
copper: or add a --toms-hardware-bias-compensation switch ;)
maligor: it's good advertisement for pts, even if they have weird results
Deanjo: Then once that option is selected each subsequent added test gets automatically either put into the cache or not depending on how the question was answered upon first install
copper: maligor: "There is no such thing as bad publicity except your own obituary."?
Deanjo: I wouldn't say posting results that give inaccurate data good publicity
maligor: copper, quite so
Deanjo: In the case of a benchmark it put's it's value into question
maligor: most benchmarks have questionable value
Deanjo: Not to the extremes that we see on Toms
michaellarabel: Deanjo: It should already keep it cached
maligor: I wonder if it's malicious or unintentional
copper: there's one big difference with PTS, it's open source, and uses open source software
copper: maligor: it's probably just incompetence
copper: "omg too many linux distros, too many GCC optimization switches"
copper: "hey, let's not use any optimization whatsoever, unlike with Windows software"
Deanjo: michaellarabel, it doesn't, it makes the download-cache folder but does not store the packages unless you run the make download cache command
michaellarabel: Deanjo: Correct, but there are two levels of caching :)
Deanjo: I gotta admit I'm not a fan of it being cached in the tmp folder
michaellarabel: It shouldn't use /tmp/ anymore at all
Deanjo: Oh I see it keeps them in the installed tests
Deanjo: didn't notice that
michaellarabel: running make-download-cache moves those to that download-cache/ folder. But PTS is aware of all locations where it could be and checks appropriately
Deanjo: OK then just pretend you didn't see my previous irc posts :P
Deanjo: copper must have taken over my nick or something :P
copper: lies, damn lies
Deanjo: That's my story and I'm sticking to it dammit
copper: quit lying, you Inquirer double-agent
Deanjo: Must be because I'm not using a directional NIC cable :P
copper: what bugs me most about the TH story is that people who comment don't comment on the OpenSSL thing (except for one dude)
Deanjo: It only receives packets and doesn't transmit them :P
copper: they don't seem to care, or see the implications with other benchmarks
copper: a bit like I don't care about what's in the BigMac I eat every once in a while
Deanjo: Ya, Tom's is one of the leading sites where people go for reviews to base their purchasing decisions. It's really to bad that they misrepresent the product by their shoddy practices
copper: and I'm not one to be content with knowing, erm, that I know better
Deanjo: I admit back in the day (before there was 10,000 tech sites) that tom's was a regular stop for me
copper: I wish I would not give a shit about anyone or anything and become a trader and make millions at the expense of penny-less americans
copper: Man, I wish I was a Lehman Brothers CEO.
maligor: how so?
copper: Make millions even when you utterly fail. Best job in the world, man.
maligor: I wouldn't
maligor: Being a CEO is a ass job
Deanjo: Their testing method is totally bubkus. They might as well be using a DSL modem to test NIC speed
maligor: You make millions but it's the equivalent to watching The Bold and the Beautiful with your eyes force kept open
copper: no no no
copper: you make millions and get to have sex with a horde of expensive hookers on your big-ass yacht
GNU\colossus: poor CEOs :(
Deanjo: thinks Macgyver should set up a Tech review site.
maligor: I expect there are plenty of CEO's who actually love the bold and the beautiful
maligor: Deanjo, like measuring operation temperature using gum, a 10cm stick, a swiss army knife and a bottle of coke?
copper: diet or regular?
Deanjo: ARRRGH!!!! what a waste of a domain http://macgyver.com/
maligor: wouldn't touch diet, destroys teeth
copper: Deanjo: bwahahahha. http://www.head-fi.org/forums/f105/d3-python-p2-heron-arrive-365831/index20.html#post4867453
maligor: heh, fortune
mha: I think I drew the jackpot on my new used laptop purchase. A Portege 2000 for about 90 EUR. with extended battery inclusive. :)
mha: too bad the 'trident' xorg driver is slow ass *HELL*. even the VESA gfx driver is faster. which says quite a lot
Deanjo: SOB!! GRRR!! http://www.macgyvertech.com/
Deanjo: goes and registers http://www.macgyver-techreview.com/
Ivanovic: no xorg 1.5.x support in "normal" 8-10
Ivanovic: that is: the folder michaellarabel wrote about is *not* included in the binary release at the amd website
maligor: Ivanovic, yeah, you can download it from ubuntu
Deanjo: hehehe n00b question of the day " if i delete fstab and mtab then what happens?"
maligor: it's different version too
maligor: Deanjo, have you tried it?
Deanjo: what the driver or the fstab question?
maligor: doubt much would happen except it wouldn't boot
Deanjo: Ya you wouldn't get too far
maligor: or it'd boot with ro root
Deanjo: heh Maximum PC just did a article on "All the new features of OOO 3.0". I still don't think is was worthy of a major version release title. Should have been OOO 2.6 or something like that. http://www.maximumpc.com/article/%5Bprimary-term%5D/open_office_30_revealed_all_features_you_dont_want_miss
maligor: well, it is snappier than 2.4
maligor: and they did add the mac version which is pretty big
Deanjo: really wishes websites would get with the times and adjust their width dynamically to the browser window
maligor: yes, static width pages annoy me too :P
Deanjo: Open office has been available on OS X for quite some time. That is not new to 3.0
maligor: neo was a separate project
maligor: and is
maligor: or do you mean the x11 ports?
Deanjo: The x11 ports
maligor: 3.0 has that shiny aqua blob
Deanjo: Well I suppose that would help system's pre 10.5 not having to install X11 support
Deanjo: michaellarabel, you could update the p7zip package. The latest version is 4.58
michaellarabel: Deanjo, it is 4.58
Deanjo: Hmm using yesterdays git it was still grabbing 4.57
michaellarabel: http://internap.dl.sourceforge.net/sourceforge/p7zip/p7zip_4.58_src_all.tar.bz2, http://voxel.dl.sourceforge.net/sourceforge/p7zip/p7zip_4.58_src_all.tar.bz2, http://ovh.dl.sourceforge.net/sourceforge/p7zip/p7zip_4.58_src_all.tar.bz2
Deanjo: http://global.phoronix-test-suite.com/?k=profile&u=dean-946-26450-26411 Nice little 5% bump in performance just using proper optimization flags
copper: Deanjo: OOo 3.0 != OOo3 ;)
Deanjo: OOo 3.0 = Same shit, different pile as OOo 2.4
copper: they wanted to get it out there so that people would actually test it :)
copper: I'm just waiting for Gnome 3.0 != Gnome3
Deanjo: Gnome < cave drawings
Deanjo: KDE > directional patch cables
copper: Openbox > Gnome > KDE > your mum
copper: (I'm such a skilled debater)
Deanjo: Gnome < Command Prompt < Openbox < KDE < my mum
Deanjo: err correction
Deanjo: Gnome < Windows 1.0 UI < Command Prompt < Openbox < KDE < my mum
copper: Moose > your mum
Deanjo: AMD > *
Deanjo: intel < diseased whore
copper: depends what kind of diseased whore
copper: asthmatic whore > Intel > seropositive whore
copper: "Rihanna's Fourth Studio Album !" thread on Head-Fi.org… something doesn't register…
dante_2core: ^^ you guys are entertaining.
copper: Thanks. We're here all week. Don't forget to tip the waitress.
Deanjo: lol Have you received our cut of the cover charge yet copper ?
Deanjo: Because I think michaellarabel is taking a little more off them then our agreed 15% agent commission
copper: I… I did the waitress, if that's what you mean…
Deanjo: w00t! shaved almost a whole second off my pbzip2 results. http://global.phoronix-test-suite.com/?k=profile&u=dean-14146-12098-1666
Deanjo: It feels soooo much faster though :P
copper: I bet
Deanjo: is now using audiophile mentality when it comes to system optimization
copper: hey, would you wire money to a Hong Kong shop via Western Union?
Deanjo: Ya they could do that
Deanjo: I wouldn't get it sent via boat though. Hong Kong seems to have an issue with high-seas piracy nowdays
copper: I'm just wondering what kind of protection that gives me
Deanjo: No idea
copper: What if the shop doesn't ship my unit after they've received the payment?
Deanjo: You call in your Triad buddies then :P
copper: I wish
copper: then again if I had Triad Buddies, I wouldn't be ordering a mere $165 product
copper: I would *steal* the $2,999 product
Deanjo: looks for other tests to optimize in PTS
copper: then kill the shop owner in front of his wife and kids, then rape the wife in front of her daughter, then rape the daughter
copper: then go home and enjoy the ridiculously overpriced DAC
copper: "AMD posts an operational profit for the third quarter"
copper: did Hell just freeze over?
Deanjo: Hmmm, you know I just thought of a potential issue with some of the compilation tests in PTS
copper: hit me
Deanjo: using just plain configure to run it would alter the compilation time results system to system depending on what preinstalled libraries you already have on the system
copper: that's not the only problem
copper: some folks will have CFLAGS set in the environment, others wont
Deanjo: for example, in mplayer the amount it has to compile could be HUGE
copper: that's why people shouldn't be using Ubuntu :p
michaellarabel: Deanjo, yeah that's a thing I've been wanting to work on is adding in switches to disable a lot of the build stuff there. Patches are welcome :)
Deanjo: autodetect flags should not be enabled
Deanjo: Ya, I'll piss around with it michaellarabel
copper: is a touchy feely 15 year old girl
Deanjo: starts printing out the mplay compile flags
copper: michaellarabel: and pleaaase, switch to VBR encoding with LAME
copper: Deanjo: think of the rain forest trees
Deanjo: Don't give anything copper wants
Deanjo: He's getting to needy :P
michaellarabel: copper: Patches welcome :)
Deanjo: "I want..., I want..., I want...." that's all he says. Well what about my needs dammit :P
copper: you're not really asking me to get off my lazy ass, are you?
copper: michaellarabel: would be a patch against PTS 1.4.0a3 be acceptable?
Deanjo: searches for that damn custom ppd file again Grrr
copper: btw since we've been bitching about Tom's Hardware not using optimizations with OpenSSL, we should enable optimizations everywhere
copper: for isntance require nasm for LAME, and use the --enable-nasm --enable-expopt=norm switches
Deanjo: Well the forced recompiling will take care of alot of that stuff
copper: not with all apps
copper: especially with the audio encoders, they often require nasm/yasm and specific switches
copper: like --enable-mmx and whatnot
Deanjo: the config file *should* detect the capabilities
copper: and I should be rich and strong and beautiful
copper: (which I am)
Deanjo: if not then the lame crew should be shot
michaellarabel: copper: Against git would be ideal, but 1.4.0a3 should work
copper: I remember that's why I didn't submit more patches
copper: (the lack of optimization I mean)
copper: can't blame Tom's Hardware for running an unoptimized binary when PTS compiles audio encoders without nasm/yasm present
copper: also, C(XX)FLAGS="-march=native" or something to that effect should be added globally
copper: hey, that would be a good way to prevent Tom's Hardware of using the same binary on different CPUs
copper: hey, that would be a good way of preventing Tom's Hardware from using the same binary on different CPUs
copper: calls the grammar nazi police
Deanjo: You could just set your environment prior to installing PTS copper
Deanjo: also there are many tests that use their own static flags (pbzip for example and p7zip(
copper: then TH could say I'm cheating
copper: look at the Windows benchmarking suites, they spend time and money on optimizing their software for every CPU they support (and occasionaly cheat)
Deanjo: michaellarabel, does the kernel compile test use a set config already?
copper: in other words they make sure that every CPU runs their software as fast as possible (or slower than Intel CPUs)
Deanjo: Damn mplayer has to have one of the most extensive set of config switches known to man
copper: I think we should come up with a set of generic enough CFLAGS, and concentrate on improving profiles one by one (e.g. add --enable-nasm for LAME, --enable-mmx for WavPack, etc…)
Deanjo: Ya there are a few tests that could be tweaked
Deanjo: But distros already have a set of generic CFlags as it is
copper: but that requires a go-ahead from michaellarabel
Deanjo: Fork it into the "Macgyver Test Suite"
michaellarabel: copper: Huh? You propose a patch and I either accept it or reject it, as easy as that :)
copper: no no no
copper: you accept the general idea first, then I work on patches
michaellarabel: What's your idea?
Deanjo: "..... this test requires duct_tape.7.90 to be installed......"
copper: hmmmpf, I just said
michaellarabel: You and Deanjo have a lot of commentary mixed in :)
copper: we're tight like that, yeah
Deanjo: Well I'm tight, him not so much
copper: 22:09:52 UTC I remember that's why I didn't submit more patches
copper: 22:10:13 UTC (the lack of optimization I mean)
copper: 22:18:24 UTC look at the Windows benchmarking suites, they spend time and money on optimizing their software for every CPU they support (and occasionaly cheat)
copper: 22:21:36 UTC I think we should come up with a set of generic enough CFLAGS, and concentrate on improving profiles one by one (e.g. add --enable-nasm for LAME, --enable-mmx for WavPack, etc…)
michaellarabel: I am fine with accepting optimizations as long as they don't present any new burdens on the user or have other problems
copper: well it might not always be simple
copper: but you can't favor convenience over performance and then claim to be on the same footing as popular Windows benchmarking suites
Deanjo: You would have to setup Processor profile support to do it effectively copper
michaellarabel: I can easily expose processor information to test profiles...
copper: maybe, I guess that's one way
copper: I'm just saying, it's not just a matter of a couple of patches
copper: ROFL http://www.rue89.com/files/20081016mccainlangueinside.jpg
Deanjo: goes over 4 fricken pages of mplayer options
copper: switching to some other software would be simpler
Deanjo: Well it's needed for the encoding tests anyways
Deanjo: what codecs are being used on the mplayer encoder test BTW
Deanjo: Hmmm, might have to have 2 separate builds of mplayer, one for compilation time and one for encoding
copper: I can't believe how much crap they post at Head-Fi.org - you could read it all day long
copper: right now I'm on the look-out for something to watch before I go to bed
copper: (no, porn doesn't count)
Deanjo: lol "Did I just see the words 'profit' and 'AMD' in the same sentence? Wait, I see some pigs in the air..."
michaellarabel: Deanjo: Have that already
Deanjo: Oh they are already 2 separate builds, OK, just popped into my head going over these build flags
Deanjo: Then I guess for the compile test the march could be permanently set
Deanjo: set it for something like i586
Deanjo: so you can have consistent build across architectures
Deanjo: for the timed compile test
Deanjo: because your not benching the compiled library but the time of compilation you want to have as static of build options as possible
Darklock: lost your t?
Marox: found some goo
Marox: anyone playing 'world of goo' ?
Darklock: gibbet dat scho?
RobbieAB: copper: the argument AGAINST custom tuning of your benchmarks to the CPU is most users won't be running custom tuned software.
copper: RobbieAB: most users may not run X software or Y game either, some may run i686 Ubuntu, others may run Gentoo with custom CFLAGS
copper: you're benchmarking hardware, you don't level down to what you think most users would use and how they would use it
copper: That's for reviews to talk about. You don't ignore SSE4 for instance, just because most users won't run apps that take advantage of it.
copper: You can't accurately compare one CPU to another (or one GPU or whatever) if you don't benchmark their true potential (or as close to their potential as possible)
copper: I made a forum post (in french) about a case where my 2.3GHz Phenom 9600 would encode audio *slower* than my 1.6GHz Intel Atom: the Phenom would be running a stock encoder with no optimizations, just like what an Ubuntu user would get, and the Atom would run an alternate, 100% compatible encoder, but with many optimizations
copper: There's too much of a gap between optimized apps and unoptimized ones to ignore them.
RobbieAB: copper: it depends on the target market for the benchmark
copper: Where do you draw the line?
RobbieAB: If you are intending benchmarks for the "average" user (if such a creature exists) you want to benchmark "average" set-ups
copper: What's the point?
RobbieAB: Obviously, I would NOT use the phoronix benchmarks to choose hardware for a HPC cluster.
copper: That's like benchmarking a Ferrari in a city at rush hour
RobbieAB: But if I'm picking hardware for my new Ubuntu box, I want to know what hardware gives me the best performance in a ubuntu box.
RobbieAB: Not what gives the best performance in a HPC cluster.
copper: then you would need separate benchmarks for Ubuntu and Gentoo - where does it stop?
copper: there's only one constant within all those variables: what the processor is capable of
RobbieAB: Which is entirely dependant on the compiler
copper: well not entirely - the apps too, of course
RobbieAB: icc compiled code often beats gcc code.
copper: yes, it often does
copper: and maybe you noticed that many benchmarks in the windows world were compiled with Intel's compiler
RobbieAB: And at one point (When AMD where still spanking Intels chips) ICC compiled benchmarks on Intel hardware was matching GCC benchmarks on AMDs hardware.
copper: Deanjo: getouttahere
RobbieAB: copper: which means Windows benchmarks are inherently biased against AMD.
copper: I wouldn't believe it when you hypothesized on TH's way Deanjo
copper: RobbieAB: and as Ars Technica revealed, PCMark is too
Deanjo: RobbieAB, even with icc the intel's chips fall way short of the AMD's
Deanjo: on SSL
RobbieAB: SSL is not the only benchmark...
copper: anyway our conversation isn't about Intel vs. AMD
Deanjo: AMD whoops intel in GMC as well RobbieAB
RobbieAB: And all I was saying was that within the usecase for the the phoronix test suite, using stock Ubuntu images is actually more valid than using a custom compiled set-up.
copper: again, you would be benchmarking a Ferrari at rush hour, it makes no sense
Deanjo: Then you are benchmarking the OS and not the processor RobbieAB and it should never be in bloody CPU charts
RobbieAB: Deanjo: What is the target of the benchmark?
copper: if you want to talk about joe six-pack usecases, don't benchmark high-end CPUs, benchmark "good enough" CPUs
Deanjo: CPU benches is to show off the CPU of course
copper: (and factor their prices in)
RobbieAB: Deanjo: only if you are in the CPU S&M team.
Deanjo: You don't take a ferrari and put low octane gas and then run 1/4 mile times
copper: RobbieAB: it goes like this: you benchmark the CPU to the fullest of its potential, and _then_ you talk about average-joe performance, in a _review_
RobbieAB: copper: ok, so you write the benchmark suite in ASM as often can't get at the full range of the CPUs capabilities using a compiler.
copper: if Joe uses a Core 2 Quad Extreme Edition with no optimizations, that's his problem
RobbieAB: Worse still, if you go down that route, you need to consider microcode tweaks.
copper: Joe's faults shouldn't reflect on the CPU
Deanjo: copper, that first "toms way" run was with the core2 gcc flags set, second run was with it letting it autoconfigure and recompile
RobbieAB: It is possible, and there are speed ups there.
copper: RobbieAB: that's why there are synthetic benchmarks
copper: but yeah, we live in the real world, and would like to see how CPUs fare in real apps
RobbieAB: And as ANYONE who takes benchmarks seriously knows, synthetic benchmarks are worthless as a fine guide to performance.
Deanjo: RobbieAB, part of the strength of linux and why it gets better performance is because of it's ability to tune itself to your hardware. If you don't want that capability then a person might as well run windows
copper: but you have to eliminate variables as much as you can
RobbieAB: Deanjo: that's nonsense.
Deanjo: No that is fact
RobbieAB: It gets better performance because it is, for the most part, a lot better written and tuned
RobbieAB: It's not down to compiler magic
copper: RobbieAB: if you don't eliminate variables, you get cases such as my aforementionned experiment where the much more powerful Phenom runs slower than the Atom
RobbieAB: copper: Under some loads, it might well do.
Deanjo: RobbieAB, alot comes from compiler magic and the supporting optimized libraries
copper: you know what the Intel Atom is, right?
RobbieAB: lowpower chip
copper: *much* lower
RobbieAB: Under some, properly optimised loads, a P4 blows away many C2Ds
copper: in every respect
Deanjo: You build a binary to the lowest common denominator and you lose any advantage that newer instruction sets give
RobbieAB: Deanjo: you optimise to the chip, and you get P4s beating C2D if it's done properly
copper: that's my point actually
copper: it makes no sense
RobbieAB: Because IF you can keep the p4 pipeline loaded, it's an animal.
Deanjo: On applications that can take advantage of a long ass pipeline absolutely
Deanjo: Unfortunately a long ass pipeline has more disadvantages then advantages
RobbieAB: Problem was the compilers had trouble do it. Hand optimised asm for a whole raft of jobs is too expensive.
copper: If you don't use the processors' technological advances, there's no point in benchmarking them at all, much less buy them
RobbieAB: But if you want to give a fair benchmark, showing WHAT the P4 can do under optimal conditions, you need to do that.
Deanjo: Why by a top of the line CD player for the car if all you listen too is AM Radio
copper: RobbieAB: dude that's what I've been saying all along
RobbieAB: copper: I don't agree, I think optimal condition benchmarks are basically artificial.
copper: they're not artificial if you can reproduce them in a useful manner
RobbieAB: You want to see how the PC performs under sub-optimal conditions, as frequently that's when you notice the difference between highend and lowend hardware
copper: RobbieAB: example: there's stock Ogg Vorbis, and there's a SSE-enabled, multithreaded, 100% compatible modification of the codec, which is 4 times faster. Which do you think people should use?
RobbieAB: copper: which is included in ubuntu?
copper: what is with you and Ubuntu
copper: fuck Ubuntu
RobbieAB: Because THAT'S what most people will use
copper: most people use Windows
Deanjo: For example RobbieAB if I build Handbrake without using proper assembly flags the performance of Handbrake drops nearly 50% when encoding 264
RobbieAB: In the context of Phoronix Teste Suite, I think we can ignore windows users
copper: then you're benchmarking Ubuntu, not the CPUs, like Deanjo said
RobbieAB: Deanjo: how many linux users do you know who build all their own software?
Deanjo: Pretty much all
RobbieAB: copper: correction, you are benchmarking that CPU under ubuntu
copper: how many people do you know who own a C2Q QX9xxx?
RobbieAB: Not many, most of the people I know with that kind of hardware buy AMD
Deanjo: When it comes to libraries that see a drastic boost the linux dude all compile those important libraries
RobbieAB: Deanjo: that you hang with a VERY geeky crowd.
copper: that's beside the point
Deanjo: Ya I hang with the linux crowd
RobbieAB: copper: no it's not.
Deanjo: People who are not afraid to get their hands dirty
copper: if you want to benchmark Ubuntu, start an Ubuntu review site
Deanjo: If they were they would still be running windows
RobbieAB: The point is that Phoronix Test Suite does not aim to be the definitive guide to the CPU performance.
Deanjo: The point of PTS IS to test the hardware
RobbieAB: It aims to be a guide to the CPU under the average Linux distro, as defined to be Ubunut.
copper: I'm sorry, I though PTS aimed at benchmarking hardware
Deanjo: That's why it was made distro agnostic
RobbieAB: It's benchmarking hardware under certain conditions.
copper: nowhere did I read anything about Ubuntu
Deanjo: Otherwise it would have been much easier to have the suite install packages instead of compiling them
copper: if it were, PTS would be distributed exclusively as an Ubuntu LiveCD
RobbieAB: I picked Ubuntu at random, it could just as easily be Fedora, or Slack, or Deb...
copper: you're not making any sense
copper: how many people use Slackware, vs. Ubuntu?
Deanjo: Why do you think PTS compiles instead of apt-get,yum,zyppering every package RobbieAB ?
RobbieAB: How many people re-compile system libs?
copper: people do what they want with their hardware
RobbieAB: Outside of the Gentoo users I know none.
copper: rich people drive their Ferraris in New York City
Deanjo: What do you NOT get about PTS being a hardware benchmark and not a OS benchmark
RobbieAB: It's not a pure "Hardware Benchmark"
RobbieAB: It's a Hardware under an OS benchmark.
copper: RobbieAB: obviously Michael disagrees with you, since he disagrees with Tom's Hardware methodology as far as PTS is concerned
RobbieAB: It uses gcc, therefore it underrates intels hardware.
copper: why, is gcc known to favor AMD CPUs?
Deanjo: you can compile it on icc if you want RobbieAB
copper: GCC is certainly more neutral than ICC
RobbieAB: Well, icc is known to produce a different performance.
copper: yeah there's even a "compiler" field in PTS reports
Deanjo: So you point is? There are AMD optimized compilers out there as well like Pathscale?
RobbieAB: Good luck getting linux code to compile cleanly on them.
Deanjo: Do it all day RobbieAB
RobbieAB: Than you don't use much of the stock linux code.
Deanjo: I use a hell of a lot
RobbieAB: I've seen what happens when you try to build a distro with ICC.
RobbieAB: It ain't pretty.
copper: anyway, PTS is not what you think it is, or want it to be
RobbieAB: And ICC is probably the most gcc compatable compiler around
copper: either take it up with Michael or fork PTS into an Ubuntu LiveCD or something
copper: discussing it further would be pointless
Deanjo: I would put pathscale way ahead of icc in compatibility. Especially when it requires GCC to even function
RobbieAB: copper: I thought you were complaining about the lack of custom compilation, not me...
copper: yeah and I took it up with Michael. According to you on the other hand, PTS already uses way too much optimization.
copper: In its current state, PTS should be unacceptable to you.
RobbieAB: I pretty much ignore it.
RobbieAB: Benchmarks, when I need them, will be run on test data using the code that will be running on the hardware.
RobbieAB: And than I will decide what to buy ;)
copper: so basically you were just trolling
copper: not that you care, but I do agree that benchmarking distros would be interesting - I'm sure there are people who'd like to know which distro would give them the most performance right out of the box, with no fine-tuning required
Deanjo: Well we will see if I get a reply now that I posted the bench on Tom's http://www.tomshardware.com/forum/254006-28-openssl-benchmark-results#t1868649
copper: nice post
copper: I should post a translation of my "Intel Atom beats AMD Phenom" post but I'm too lazy
Deanjo: Well it's just plain stupid. They don't bench video card playback without using Purevideo or UVD
Deanjo: They might as well judge picture quality of a screen using youtube video streaming
copper: Yeah but see, that's alright, since Joe sixpack install NVIDIA's drivers and runs optimized Windows apps ;)
Deanjo: Uhhuh right, that's why when you ask Joe Sixpack what video drivers he's running they usually reply, "The ones that came with the system"
copper: well it's not Ubuntu anyway!
copper: So the benchmark's rigged. Using Windows, and all.
Deanjo: pfft I bench using nothing but OS/2
copper: We should all just buy Pentiums (non-MMX) and be done with it, eh.
Deanjo: Pretty much, they should just get rid of all the extension sets since we will never use them
copper: Screw progress. It kills baby seals in Antartica.
copper: Make that the Arctic Ocean.
Deanjo: Screw FPU's as well
copper: It's all a big conspiracy to make us buy CPUs.
Deanjo: Give me a Xeon sx dammit
Deanjo: Hell just give me a Texas Instruments Ti-30 calc. That's more then anyone would ever need
copper: no no no, the Hammish are the ones who got it right.
Deanjo: And I could go years on one charge with the Ti
Deanjo: Screw this 5-6 hours and recharge bullshit
copper: lol, $250 for a "DAC cable" which probably costs $10 to make http://www.devilsound.com/DAC/
copper: that one is definitely directional ;)
Deanjo: JUNK!!! It's not officially endorsed by the Pope.
RobbieAB: copper: yep, progress does kill baby seals.
RobbieAB: What do you think is causing all the global warming :P
copper: get off IRC then
copper: every minute you stay on IRC, a baby seal dies in part because of you
copper: me, I eat baby seals for BREAKFAST.
RobbieAB: Hey, who said I cared about the baby seals?
RobbieAB: Though I suppose I am typing this on an old p3 500,,,
RobbieAB: Silly laptop had the LVM got down the tubes on me, :(
copper: that's what it's about… you're just jealous of my Phenom.
copper: Oh, and I'm sorry if my Intel Atom whoops your P3's ass.
copper: Deanjo! Show him your overclocked Phenom!
RobbieAB: Hey, my P3 does everything I ask solidly and well.
RobbieAB: In fact it's proved remarkably good as a desktop while my laptop was out of action.