Ati Mobility Radeon HD 4670 DDR3 and Linux

[jonnycat26]

Now i am going to give ATI 1m chance. Is it worth it ? or should i go After lastest GF 9600GT so to not to have any move problems with grafic drivers ?

I picked up an ATI 4650 because I wanted to support ATI. I ended up picking up a GF9400GT because the ATI drivers were so bad.

Take my advice, stick with NVidia until ATI gets the driver situation sorted.

[TASADAR-F]

I have ATI HD4670 desktop, I bought in August. 2D,3D and wine work very well. The main problem is video tears. (I not test with composite)

I choose ATI for 2 reasons. 1- ATI documentation (drivers radeonHD and radeon). 2- Price, less power than 9600 and acelerated video (XvMC).

But at this moment Only Nvidia has video acelerated (VDPU).

Open drivers(RadeonHD and radeon) for HD4000.3000.2000,X1000 improve very fast. Now have EXA 2D, tears free in X1000,HD2000, basic 3D in X1000 and 2000.

[cutterjohn]

3D isn't working because the card is too new for OSS driver support.

Catalyst 9.2 seems to be working OK with my ATI Radeon Mobility 4850, but I had to turn off compiz as it was driving me nuts and causing other problems.

On a side note, you MAY have to resort to a basic text mode install of whatever linux you choose, as Ubuntu 8.10 x86-64 wouldn't even work with the VESA drivers on my nb (MSI GT725-074US). I did a manual text install, got networking up, grabbed cat 9.2 (plus some other needed packages), installed everything, rebooted to a working X11. (Blast from the ancient past, as while I can do manual text installs, I still don't enjoy doing them and had hoped to see the last of them YEARS ago. Also got to do some manual actual file editing configs... joy... felt like 1993 all over again excepting that I didn't have to bootstrap gcc then build X11 from source...)


I'm not even going to bother trying the OSS drivers until they support 3D on the 4XXX series Radeons... hopefully before the endof the year.

Also Ubuntu 9.04 has a special catalyst build with support for X.org 1.6, so I suspect that that beta release may just work out of the box w/o resorting to manual text install.

[krazy]

Setting video out to X11 - is that a joke? That's software rendering for video, who wants that?

I do. Unless you're watching HD, it doesn't take much CPU on a modern computer and currently offers the best performance for me on ATI's drivers.

[grigi]

I don't know why people are so hard on ATI's drivers (fglrx). I actually have more frustrations on my Nvidia based desktop than my ATI based notebook.

Yes, wobbly windows has issues, but the Nvidia also has issues. (The Nvidia's issues are only there after using them for a while, they still did not get rid of the memory leaks.) And the Nvidia 180.xx series drivers don't work for me at all

So on all systems I run without Compiz/Kwin. I can live without some of KWin's usability optimisations... The only system that works correctly is an Intel Q33 desktop with not-quite-released-yet DRI2/GEM/KMS. Although performance on that box leaves much to be desired

Powermanagement, Suspend/Resume, Wine, Native Games, Videos... I have no troubles with that.

Also note: Nvidia cross/re/backward brands their products so much, that you don't know what is going on.
The Mobile GF 9600GT is almost 3 times slower than a desktop GT 9600GT. It is also a tad slower than a Desktop 9500GT.

[cutterjohn]

Actually my experiences with nVidia are the exact opposite. Desktop has 7600GT(and had a 9600M GT nb for a while) no problems with memory leaks, actually uses less memory than my nb w/4850 AFTER the desktop has been up for weeks!

VESA mode had always been detected fine on nVidia cards, while, again, it failed with the 4850 forcing a text install plus a bunch of manual config to allow obtaining and installation of catalyst drivers.

Suspend/resume: worked fine(EVERY time) with nVidia cards, but so far w/the ATI nb 3 times out of 4 the nb failed to resume correctly from suspend.

9600M GT: Really only true for the GDDR3 version, as the GDDR3 version(which I had) significantly outperforms the GDDR2 version, placing it somewhere between the 9500 & 9600GTs, c. 5500 3DMark06 3500 or so Vantage IIRC. (4850 does c. 9500 & 4900, but the ATI mobile parts are pretty much the same as the desktop while nVidia mobile parts are scaled down versions, fewer shader blocks ow they both reduce GPU & memory clocks for mobile. Users with tweaked settings and P9500s are getting 11500 3DMarks, and 6k or so Vantage.)

Compositing: AWFUL(did I mention AWFUL?) w/Catalyst drivers(only thing that I can realistically use ATM) while it seemed to work beautifully w/nVidia cards, although I really don't like it.

Temps: 4850 seems to run cooler than my 9600MGT did at least for Stalker SoC which got the 9600 to low 90sC while the 4850 topped out at c. 76C.

Beyond that both nbs had the same specs, 1680x1050 LCD panel, P8600, 4GB DDR2-800, 320GB WD hdd, Intel WiFi 5XXX series, PM45M + ICH-9M chipset. (One thing I must say is that the 725 is 17" while the 620 was 15.4" and it makes more of a difference in portability than I would have thought sizewise. The 725 is fairly light for a 17" nb.)

So, ATM for performance the 4850 is great, but nVidia is releasing some updated mobile parts which will lift them ahead again in raw performance. (4850 basically matches 9800GTX ATM...)

[EDIT]
One problem that was unique to the 620 though was that read-edid was only picking up VESA to 1280x1024, but the nVidia driver picked 1680x1050 some how which was the native resolution, but nothing else so it seemed to be incapable of changing resolutions. I never got to fully explore this as I had keyboard problems with that nb and ended up getting refunded for it, so I'm not at all sure if it was an nVidia problem or something fishy that MSI was doing w/BIOS/firmwares. Similar non-detection problems have been reported with other GPUs/LCD Panels/OEMs. GT725 is, seemingly, picking up all supported resolutions fine at least by the catalyst drivers which have the full list of selectable screen resolutions, although ATI may be hacking a workaround that nVidia elected not to do. Something else that came out of this is that Ubuntu is, apparently, using a very old version of read-edid. (This is all so that we don't have to manually configure modelines in the conf files, which is nice since some of these LCD panels are hard to find necessary info to generate modelines for.)
[/EDIT]

[cutterjohn]

OK, nVidia drivers, here's my desktop with AMD 4800+ X2 (socket 939) + 2GB DRAM DDR-400:

 15:08:18 up 7 days, 18:49,  8 users,  load average: 0.00, 0.08, 0.11

 6301 root      20   0  104m  49m  12m S    0  2.5   20:47.16 Xorg

And here's my nb with the 4850, 64b ubuntu

 15:08:46 up  1:34,  2 users,  load average: 0.59, 2.22, 1.43

 5854 root      20   0  540m  74m  25m S   12  1.9   0:36.66 Xorg

Notice the obvious differences?

desktop is
8.04 LTS Ubuntu 32b, 169.12 nVidia driver. Had similar results with newest betas on a nb with 9600M GT, compositing enabled.

nb has cats 9.2 + Ubuntu 8.10 64b, the numbers may not look so bad now, but both shared and resident will go over 100m after a few more hours. (compositing disabled since ever windowed opengl app will flicker like mad with it on, and I can't say that I truly miss it though...)

[Kano]

@krazy

xv has usually filters active when you use scaling to fullscreen for example. Usually a dvd uses always scaling when the video is anamorph encoded - unfiltered it looks blocky for good eyes with default scaling. Also i don't think that anybody wants to see 720p or 1080p/i with software scaling...

[macmus]

how it behave now with new 9.3 driver (and improved composite support)?
Is ATI 4670 feasable now ?

Anyone tried 1366x768 on this one ?