Phoronix IRC Log: 2009-01-24
^6wl: morning all, quick question, does anyone have the intel DG45FC, and has anyone had any luck with HDMI audio?
Ivanovic: michaellarabel: re the r700 2d performance test: what are the results with a slower cpu?
Ivanovic: since i could imagine that lots of features are so far not implemented in the open drivers, those are most likely done by the cpu only
Ivanovic: and you were using a cpu with a damn fast "single process" speed
Ivanovic: how do the results looks at the default speed of 3ghz or even at a speed of 2ghz
Ivanovic: would be interesting to see the 4 results in parallel
Ivanovic: so speed of r700 radeonhd at 4ghz cpu and 2ghz cpu as well as fglrx with both cpu speeds
^6wl: right, step up
^6wl: hdmi audio is working on the console
^6wl: just not in X
^6wl: I'm running 2.6.1 xf86-video-intel
^6wl: anyone know if there is anything you need to do to enable HDMI audio that was added in 2.6.0 of the drivers?
maligor: might be a different device
maligor: different audio device that is
maligor: ^6wl, using alsa 1.0.19?
^6wl: yeah, using alsa 1.0.19 and 2.6.29-rc2-git
^6wl: the audio works fine in the console
^6wl: just not in X
Kano: ^6wl: do you use dsp or alsa output
Kano: did you ask in #intel-gfx ?
^6wl: yeah, its been all silent so far
^6wl: was hoping the leets here would know =)
Kano: well i do not even own a tv with hdmi nor an intel board with hdmi connector
Kano: dont know if somebody else does here
^6wl: =) this is my htpc project
Kano: how about using a cheap gf 9300 with hdmi?
^6wl: not open source drivers
Kano: does not matter, but h264 via vdpau
^6wl: i refuse to get another nvidia after every they depreciated 2 video cards I owned
^6wl: and didn't upgrade the legacy drivers to support the latest X abi
Kano: well only 71.xx series does not work
Kano: they have got 3 legacy drivers, 2 work with 1.6
^6wl: maybe it was the kernel, cant remember off the top of my head
Kano: what card do you own?
^6wl: and a 5900 I think it is
Kano: 173.xx is supported
^6wl: had to ditch the 5200 for something or other
Kano: if you can not use it is your fault
^6wl: then the 5900 didn't work with > 2,6,26 I think it was
Kano: forget it
^6wl: forget what?
Kano: if you would use kanotix they work after calling 2 scripts
Kano: you use the wrong distro
^6wl: yeah, thats it
redeeman: you can naturally get the patches applied in any distribution
redeeman: ey kano, i got a small issue, maybe you've had it too
Kano: currently you dont need any patches
redeeman: after i upgraded to 180.22 drivers, kde remote desktop process krfb crashes after i log off the vnc session
Kano: redeeman: i only use x11vnc/vncviewer not krfb
redeeman: why not krfb?
Kano: because of the listen mode
Kano: i add a portforward to my system and those who need help do not need a portforward
redeeman: you mean so server can contact client?
redeeman: but that's for the client
Kano: they call: x11vnc -connect ip
redeeman: you could still use krfb
Kano: i call: vncviewer -bgr233 -listen
^6wl: well, just went to the nvidia site to checkout if the 5200 was support (prefer that card cos its passively cooled) - "System Message, Please try again at a later time"
^6wl: I'll stick with intel and fully open source thanks
Kano: which distro do you use?
Kano: cant you emerge it?
Kano: 173.14.15 is the driver you would need
^6wl: yeah, im sure there was a change in the > 2.6.26 kernel header with the last supported version
Kano: forget it
Kano: that works without patch even for 2.6.28
^6wl: again, forget what
Kano: i support all nvidia cards since ages. if there are patches needed in most cases nvidia provides em, at last for 180/173. for 96/71 i backported some myself
Kano: but for hdmi you would need something from series 9
Kano: which has a header to connect spdif from board
^6wl: well, fingers crossed I get this working, since i've got it working in console, assume its not far away from working in x
^6wl: got it all working
redeeman: which is the ati chipset which has R500 based graphics which is supported?
redeeman: was it 690g that's the newest supported or?
redeeman: im thinking of getting myself a phenom 2 box
mib_euw1z5: Phenom 2
mib_euw1z5: Excellent choice sir
mib_euw1z5: You want integrated graphics?
redeeman: well.. i don't really need fast graphics on the box, so why not, i figure
RobbieAB: redeeman: which OS?
redeeman: linux ofcourse
mib_euw1z5: I do believe that 690g has R500
mib_euw1z5: Which distribution is probably what he meant...
redeeman: i don't see what distribution matters - i can get the software on any one
RobbieAB: If you don't need 3d, integrated is best bet. If you are prepared to run experimental X and drivers, any chipset should just about function
redeeman: i do want 3d, but i also know that there is one integrated that is R500 based and supported
RobbieAB: Well, 2D runs on the R6/700 and 3D should arrive in the next 6 months...
GNU\colossus: The IGP's 3D architecture is based on Radeon R420 and contains 4 pixel pipelines capable of Shader Model version 2.0b with DirectX 9 and OpenGL 2.0 compatibility but lacks hardware vertex processing.
RobbieAB: And for Phenom II you might do better waiting for the AM3 chipsets. ;)
redeeman: RobbieAB: i don't give much for the "should"
GNU\colossus: it's a pity that AMD's southbridges suck so much
revx: oh really?
RobbieAB: redeeman: Yeah, I know that feeling. But if you can bear to wait, AM3 should offer a better system than AM+
revx: GNU\colossus: I've had no issues with the SB600 series
redeeman: revx: i hear usb sucks, and ncq is non existent or buggy?
revx: GNU\colossus: in fact I like it quite a bit!
GNU\colossus: revx: well, SB600's USB2 transfer speed, for example, is abysmal
revx: I've had far worse.. *remembers the days of his VIA chipsets and shivers*
GNU\colossus: I've never had problems with VIA ;)
redeeman: usb2 gets good speeds on my via chipset
GNU\colossus: but with nforce3... oh boy.
mib_euw1z5: I've had BIG problems with VIA
revx: mib_euw1z5: same
mib_euw1z5: KT 133 and KT333 chipsets were aweful
Ivanovic: michaellarabel: around?
Ivanovic: oaky, probably it is michaellarabel_ by now
GNU\colossus: kt333 was a-ok :)
RobbieAB: VIA needs a LOT of work done on their chipsets, as they need to seriously boost the memory speed.
mib_euw1z5: Are you $%^*#$^*# me?!?!?!
RobbieAB: has a HP 2133
michaellarabel_: hi ivanovic
mib_euw1z5: KT333 was incompatible with ANY ram
Ivanovic: michaellarabel_: could you also run the 2d r700 benchmarks with with a slower cpu?
michaellarabel_: When I have the time
Ivanovic: would be nice to see how the driver does scale
Ivanovic: cf my comment from about 10h ago
RobbieAB: With decent memory speed, it would be VERY nice system if I was doing dmcrypt with AES
mib_euw1z5: Does memory speed have that much of an impact with phenom?
mib_euw1z5: I thought that the real life situations benefitted very little with anything faster than DDR400
maligor: GNU\colossus, wasn't KT333 the one with half the pci broken
mib_euw1z5: KT333 is the one they pretended never existed
GNU\colossus: maligor: if you're recalling the pain with high loads on the PCI-Bus when using emu10k1 soundcards - that was KT133, actually
RobbieAB: mib_euw1z5: memory speed is more important with the multicore CPUs.
maligor: GNU\colossus, oh, it wasn't just emu10k1 soundcards
maligor: GNU\colossus, half of the pci bus was actually broken
maligor: it's one of those WTF designs
mib_euw1z5: My KT133 was excellent apart from Non USB compliant USB
mib_euw1z5: I also had a ISA slot in it
maligor: they work fine as long as you don't use half the pci slots :P
mib_euw1z5: So I could use my US robotics 56k
mib_euw1z5: Only modem to connect at 50.666 speed at the time
maligor: speed of the devil, eh
mib_euw1z5: Slightly faster than 48.333
RobbieAB: maligor: so they don't work as a desktop, but in laptops and the like, they are ok?
maligor: RobbieAB, Never had one of those
maligor: if the HP 2133 has no problems, then it has no problems :P
RobbieAB: It has problems, there is what looks like a weird bug in the PCI system... It may be the library though.
maligor: They sold them at one shop here for 150eur at one point, which I only noticed afterwards :P
maligor: via does have some cool tech stuff, like Via Padlock
RobbieAB: Yes, which is memory bound
maligor: the laptop has padlock?
RobbieAB: VERY VERY VERY memory bound, if their theoretical figures are to be believed
RobbieAB: Yes, all C7-M have padlock
maligor: benched it yourself then?
RobbieAB: Seen enough non-Via figures
RobbieAB: Via say 25Gb/s AES speeds, most non-Via benchmarks are saying 800Mb/s as peak.
maligor: so 100MB/s?
maligor: that's pretty decent
RobbieAB: Yes, but with faster memory...
redeeman: a single E8400 core does 100MB/s
maligor: redeeman, and the point is?-)
redeeman: so if that's what they reach with dedicated silicon, im not impressed
redeeman: the via stuff?
GNU\colossus: are we talking bits or bytes?
RobbieAB: redeeman: that's on a 1.5GHz clock as well...
maligor: C7-M is a mobile chip
maligor: you can compare it to atom rather than a friggen E8400
RobbieAB: The basic problem padlock has is getting the data to the chip fast enough.
redeeman: maligor: but if it has a dedicated piece of hardware for the encryption stuff...
RobbieAB: It looks like the AES benches are choking on the memory bandwidth
RobbieAB: redeeman: what clock is that E8400? What power draw?
maligor: E8400 is a 3.0GHz top of the line dual core
maligor: I mean jeez
redeeman: i don't think you understand what it is im saying
maligor: you're comparing it to a lowpower mobile part
maligor: no, i don't
RobbieAB: So a single core chip, clocked at 50%, drawing a fraction of the power, is competitive, and it's not pretty good?
redeeman: what im saying is, that if a general purpose cpu, with a software aes implementation, granted, on a very fast cpu, can match dedicated encryption logic, im not impressed, even if that dedicated logic is contained in a much slower cpu
RobbieAB: redeeman: that general purpose CPU is CPU limited on AES, NOT memory limited
maligor: 1.5GHz C7-M is 12Watts
maligor: that's on 90nm tech
ZeXx86: Is possible to start hunting for UT3 Linux Client informations ? Its not normal to wait years for this game port
RobbieAB: Improve the memory performance for the Via chip, and watch that benchmark grow.
mib_euw1z5: Normal, probably not, common... I suppose it is
redeeman: the atom does 20MB/s AES
redeeman: in software
mib_euw1z5: redeeman: Then perhaps VIA should have been making CPU's the whole time
redeeman: i wonder what the amd geode companion chip does in speed, for it's 128bit aes
maligor: lots of embedded devices have crypto engines
maligor: 800Mb/s means you can saturate a gigabit network with encrypted data
mib_euw1z5: Which would do what exactly?
RobbieAB: pretty impressive for a cheap, 12W CPU, don't you think?
mib_euw1z5: 650$ for HP2133 with SLED
RobbieAB: Now lets me see... I got my 2133 forr 350Eur, and they are currently hipping in IE for 300...
RobbieAB: Ok, that's with 1GB ram, and a 1.2GHz chip, but...
RobbieAB: That is the 1280x768 display btw, which is probably a better bet. :)
mib_euw1z5: I find it's a bit expensive
mib_euw1z5: And I'm woondering about the graphics
mib_euw1z5: VIA Chrome 9, 128MB shared system memory
mib_euw1z5: Is Chrome really good at something? Frome what I've gathered it's not very powerful, but then again for a system like that, energy consumption is probably what's best
mib_euw1z5: The one I'm looking at doesn't have that resolution
mib_euw1z5: 1024x600 I believe
RobbieAB: Also, you really don't want SLED on it.
mib_euw1z5: Why not?
RobbieAB: You want something lighter.
mib_euw1z5: What would you recommend?
RobbieAB: I'm running gentoo on it, once I got it set-up, I build over night.
RobbieAB: It's nice enough in use.
mib_euw1z5: I see
RobbieAB: Basically, you don't want a current SLED for the same reason you don't want Vista.
mib_euw1z5: Although I did get my mac G3 running with OpenSuse 10
RobbieAB: It's too heavy for a lightweight machine. XP or something that runs trimmed down is going to be a nicer experience.
mib_euw1z5: AND KDE4
RobbieAB: Well, if you are happy with that level of responsiveness, disregard my comments. ;)
mib_euw1z5: I wasn't thrilled
mib_euw1z5: But I was excited to see that it did run
mib_euw1z5: I need to re-install because I fubared it
RobbieAB: Yeah, that's kind of the point, you want to use the system much, you'll want something responsive.
RobbieAB: That said, if you really know your way around Suse, you may be able to strip a SLED down easier than getting anything else working
mib_euw1z5: How lightweight is XFCE anyway?
mib_euw1z5: Compared to KDE4
RobbieAB: Depends how much of KDE4 you use, I suspect.
RobbieAB: If you use KDE4, with FF as your browser, TB as your email client, OOo for your office software, XFCE4 will be lighter.
mib_euw1z5: I see
mib_euw1z5: Looking at the screenshots, it looks pretty stripped down
RobbieAB: If you use KDE4, and non-KDE4 apps, it's going to be a lot more competitive.
RobbieAB: But the minute you install The Gimp, or FF, ...
RobbieAB: I run with only one Qt app installed, and that's statically linked, so I have no Qt or KDE libs installed.
RobbieAB: (The Qt app is Skype!)
RobbieAB: There is also LXDE (For the ultra lightweight approach)
mib_euw1z5: Sounds good
mib_euw1z5: It's an older system with probably 1/3 the power of your 2133 :P
mib_euw1z5: ATI Rage 128
RobbieAB: lol, so about comparable with my 3 PIII 500s
RobbieAB: Same graphics as well...
RobbieAB: I got a stripped GNOME running on the PIII
maligor: 1.3GB the whole system is using here with kde4 and a few apps running :P
mib_euw1z5: Although I think PPC 400mhz may be a little more performant than a P3 500
RobbieAB: Not the most responsive system I've ever used, but certainly usable
maligor: weird having 1.5GB in caches
mib_euw1z5: On you 2133 or P3?
RobbieAB: the P3.
maligor: KDE 4.2 is pretty sweet
RobbieAB: That's one benefit of gentoo... ;)
RobbieAB: Once the build is done, with properly targetted CFLAGs, it does give better performance.
RobbieAB: -march=pentium3 is great ;)
maligor: not a whole lot
maligor: not on recent systems anyway
RobbieAB: Less noticable on recent systems, partly because the 64bit binary distros all build with much the same flags gentoo users would use anyway.
RobbieAB: But -march=pentium3 should beat -march=686 on a P3 ;)
maligor: have you compared?
maligor: as in: empirical study
maligor: before you do, that statement is pretty meaningless
RobbieAB: I haven't benched, no.
RobbieAB: But to say that renders the comment worthless is to ignore the whole point of the march flag
RobbieAB: Which every gentoo dev I have spoken to has said is the single most important optimising flag.
mib_euw1z5: What does the march flag do??
maligor: pentium3 just enables MMX and SSE
maligor: mib_euw1z5, instruction sets
maligor: it's entirely a different question on how well gcc can use them
mib_euw1z5: I'll got with RobbieAB and his hyppothsis
mib_euw1z5: I'll put 5 space bucks on it
maligor: recent gcc versions do have auto-vectorization features too
maligor: but it's not enabled by default
maligor: and I really wouldn't use them for compiling main systems :P
RobbieAB: maligor: It's more than just instruction sets, hence the distinction between mtune and march
RobbieAB: If it was JUST instruction sets, the distinction wouldn't exist.
maligor: you could try compiling something with both flags and disassembling them
maligor: hmm.. I might just have the thing to test it
mib_euw1z5: I would assume that -march=686 would be much slower if the software decides to rely on CPU instruction sets that are really not there
mib_euw1z5: Possibly freezing
RobbieAB: mib_euw1z5: if it's trying to use cpu instructions that aren't there, it should crash
RobbieAB: However, it's also a matter of scheduling. If there are any scheduling differences between a pentium3 and a pentium4 (for example) march with tell the compiler to take them into account.
RobbieAB: As will mtune
maligor: RobbieAB, I be surprised
maligor: it does generate slightly different results
maligor: one 'mov' was replaced by a 'or'
RobbieAB: This is something small, I take it?
maligor: and there's 'inc' instead of 'add 0x1'
maligor: dunno really what the differences between execution cycles they take are
RobbieAB: probably marginal.
maligor: but since it's just a few odd instructions, you'd have to run the whole thing millions of times
maligor: it was just a simple euler solver tho
maligor: my analysis could be flawed, I'm too lazy to read the longer solvers
maligor: hmm... diff to the rescue
RobbieAB: However, I suspect that it tries to make it so that more complex programs can context switch easily, which is very CPU specific... ;)
RobbieAB: I will be the first to admit I have no clue how to check for something like that though.
RobbieAB: On such matters I bow to the superiour knowledge of the GCC coders.
mib_euw1z5: superiour eh?
RobbieAB: Hey, I never said I could speel
mib_euw1z5: lol :)
maligor: I'll just bow down to the superior knowledge of the amd engineers who designed amd64
RobbieAB: That works. The other critical flag used on x86 is mfpmath=sse, which is enabled by default on 64bit systems
maligor: I wonder how useful that is with SSE1 only
RobbieAB: Standard, sane, flags: -march=<> -O2, -mfpmath=sse
maligor: I'd leave O levels to the apps
RobbieAB: Moderately useful, as a lot of graphics work and multimedia work will use float, and not double.
maligor: tried running kde4 on that p3?
RobbieAB: Not yet, it could be fun
maligor: this thing has more bling than vista
maligor: 4.2 anyway, and I'm surprised how good it is
revx: I wait for the day when gcc can figure out stuff like this:
RobbieAB: But I'm not a KDE user, and I never bothered even trying to get the graphics chip working properly
revx: float a, b; a or b could be stored in a sse reg
revx: added, scaled, etc
revx: and subject to other optimizations
revx: you can do this with the SIMD types
maligor: SIMD stuff tend to want aligned values
revx: but GCC can't see this without explicitly using non-standard types
maligor: I think it had unaligned ops too tho, but they take a hit
RobbieAB: funroll-loops should rarely be used as a system wide cflag
maligor: I had fun writing SSE assembler at one point
revx: maligor: indeed.. actualy a change for AMD Phenom II is that the prefetch SSE2 instructions can now take unaligned addresses!
maligor: to get good results you have to pretty much have the whole app using aligned
RobbieAB: march is safe (basically, any code it breaks is already broken), -O2 is also generally safe, and a generally sane "default" optimisation flag
maligor: so it's a pita if it's a old program :P
revx: unaligned random access is bad
revx: but unaligned streaming access only has a small penalty
maligor: RobbieAB, I've seen programs that explictly didn't use O2 because it breaks something
maligor: it was a while back tho
RobbieAB: -O3 is silly, and can sometimes actually degrade performance. sometimes -Os can beat -O2.
revx: maligor: well O2 breaks stuff on different arches
revx: maligor: in gcc on k10, it doesn't do anything "unsafe" but O3 does
revx: unless you have some extreme hackery :P
RobbieAB: maligor: Oh, yes, but the few programs that break with -O2 will tend to have a flag filter on distro build systems
RobbieAB: So everything else builds with it sanely.
revx: RobbieAB: how many source based distros with build systems do you know of?
RobbieAB: revx: Gentoo, not sure of any other large source distros.
maligor: OE/Angstrom is another big name
RobbieAB: But if it's to be a distro, it needs a build system, I would have thought.
maligor: it's a embedded setup tho
RobbieAB: Also, don't most binary distros build their apps with -O2?
RobbieAB: Except for the few apps that don't, of course.
maligor: in debian it depends on the mood of the package builder
RobbieAB: won't comment on that.
RobbieAB: And She Who Feeds Me is calling, so later.
mib_euw1z5: I feed she
mib_euw1z5: She does dishes
mib_euw1z5: I and she are going to drink tonight :P
Roberth1: hmm im trying to play return to castle wolfenstein here, but the game cant open /dev/dsp so I get no sound, anyone know about any workaround?
Kano: you can use alsa with it
Roberth1: how do I set that?
Kano: you can do two things
Kano: first you can try
Kano: echo "et.x86 0 0 direct" > /proc/asound/card0/pcm0p/oss
Kano: and then execute the game
Kano: echo "et.x86 0 0 direct" > /proc/asound/card0/pcm0p/oss
Roberth1: that didnt give much sound
Kano: well the first is the executealbe name
Roberth1: they are both the same
Roberth1: but isnt et another game?
Kano: hi evocallaghan
evocallaghan: Hows things?
evocallaghan: michaellarabel: You about buddy ?
Roberth1: Kano: yes it is
Roberth1: Kano: i had to use echo 'wolfsp.x86 0 0 direct' > /proc/asound/card0/pcm0p/oss
Kano: ok, but you the idea ;)
Roberth1: uh the zombie/demon is scary:S
Kano: michaellarabel: did you ever test nvidia drivers on U 9.04?
Kano: for xserver 1.6 only ignoreABi is needed, just added that hack to my script
Kano: as my ati card is damaged, i could not test a similar thing for ati
Kano: if somebody has a u 9.04 install with ssh running i could test it
Kano: i put that part into a script, anybody likes to try?
Kano: or send me a working ati pci-e card ;)
Kano: mine has gfx errors even in bios
Kano: flash is still really slow in fullsize with 64 bit...
Kano: anybody awake?
michaellarabel: Kano: I don't have any ATI systems running Ubuntu 9.04 right now
Kano: then try it live
Kano: should not matter
Kano: or install ssh on one system in live mode and let me connect
Kano: well most easy way would be a modded fglrx script i guess