XRender on ati catalyst

[Dandel]

The page uses a lot of features, and causes a huge strain on the rendering code... right now firefox 3.6.9

1280x1024 windowed - 14fps
1920x1200 windowed - 5fps

I would honestly expect at least 30fps on 1920x1200 with a current generation video card ( HD5770).

[bridgman]

OK, I get "error on page" with IE8 on Vista

[DarkFoss]

Using Firefox 3.6.8 I get 10 fps, Opera 16 fps, Iron 4 fps. All 64 bit browsers Cat 10.8 HD 5770 1920x1080@60Hz kde 4.5 composite on opengl accelerated w/vsync on.

[bridgman]

If anyone is running the open source drivers I guess it would be interesting to see what kind of numbers we see with those.

[jrch2k8]

I guess the first thing I would do is check to see which Render hooks were pointing to driver code and which (if any) were still pointing to software rendering code in the X server... then hook those locations and see which ones were being called and how often.

Not like I remember how to do that or anything...

i don't wanna be mean or something since i know that fglrx priority is some workstation apps mostly but since the driver tend always to be heavily bugged at most levels(opengl feature/shaders/uvd/etc) it wouldn't make sense to put a button on catalyst control center to activate verbose hell like debug option?

i know your priority is workstations but remember your PR department that ppl buying firegl or quadro always tend to use in their PC's the respective consumer level radeon/geforce cards, so many users out there looking for a workstation card won't think too good about firegl cards if they have a pc with a radeon using fglrx these days(ofc i assume this type of users wont go to radeon until is stable enough), so an easy way to debug the driver should help in the workstation market in the long run since that would help on the quality of the driver and give a better image to your hardware on linux

[bridgman]

Sorry, just to be clear we have debug/verbose mechanisms for the developers, I thought you were asking for a way to see for yourself which functions were accelerated.

Sounds like you are suggesting a verbose debug mechanism for regular users... other than "it works on our hardware but breaks on yours" scenarios how do you see that being useful ? Doesn't it need a developer to analyze the results anyways ?

[ergo]

Hey guys, so whats the current consensus on this? will we be able to benefit from rendering accelleration in browsers?

Im an end user - not a developer ( at least not requiring gpu for my web app programming), so this topic really interests me, since HTML5 IS around the corner.

[rohcQaH]

notebook, core2duo, 2.1GHz, Quadro NVS 140M, binary blob
seamonkey: 2 fps
firefox: not installed
opera: 14 fps

workstation, Athlon 64X2, 2.6GHz, Radeon HD 5770, fglrx 10.7:
seamonkey: 11 fps
firefox: 14 fps
opera: 10 fps

as above, OSS drivers:
seamonkey: 1 fps
firefox: 1 fps
opera: 12 fps

While seamonkey 2.0.6 and opera 10.61 were 64bit, I've used a 32bit-install of firefox 3.6.8.
In all cases, kwin3 with compositing was running.

The fps were fluctuating quite a bit and I didn't do any repeat runs, so take the numbers with a grain of salt.

[nanonyme]

If anyone is running the open source drivers I guess it would be interesting to see what kind of numbers we see with those.

My results with open drivers and Firefox mostly were 3 fps until CPU peaked at 100%, everything stopped responding and I had to change to VT to kill Firefox.

[c0un7d0wn]

core i7 Q720 with 5830 with xorg-ati:evergreen_accel kernel 2.6.36-rc3

firefox: 1fps
chromium: 24fps