"Ask ATI" dev thread

[EagleDM]

Hello devs!!

Here are my questions:

1- will you fix the "vsync" issue where NVIDIA vsync perfectly with compiz and ATI fglrx drivers cannot ? (this I hope will be fixed soon)

2- Is there any change you could work with a Xvmc implementation or any sort of work to use the UVD and UVD2 on linux? or else, is such a waste of power sitting there.

3- Any change to fix via UVD2 or Xvmc the flashing and popping while playing videos?

4- Are you going to implement OpenGL properly (how soon?) If you rotate the cube while rendering OpenGL on screen, the windows rendering OpenGL will stay there while all the rest of the desktop moves in the cube, making obvious that the OpenGL rendering is not syncing with the rest of the desktop.

5- Is Crossfire active on Linux a priority? are you going to implement the rest of Catalyst for Windows in Catalyst for Linux?



That's all for now, since I do Gaming with Vista and in that matter Catalyst for Vista really works VERY good, my only concern now is with Ubuntu Linux , Vsync and Video.

Video is totally unaceptable as of right now

Thank you very much for your time and keep up the good work!

[Zhick]

I'm not an Ati-Dev, but regarding your 3rd and 4th question: This is not a problem with the fglrx-driver and not due to a faulty OpenGL-Implementation (though fglrx's OpenGL-Implementation isn't perfect anyway) but this is a limit of the X Server's and DRI's structure. nVidia has got a workaround in their proprietary driver, but every other driver has got the same problems as well (except for video-playback with Xv. This seems to be correctly redirected on most other drivers except on fglrx. I have no idea why though...).

[Melcar]

I'm not an Ati-Dev, but regarding your 3rd and 4th question: This is not a problem with the fglrx-driver and not due to a faulty OpenGL-Implementation (though fglrx's OpenGL-Implementation isn't perfect anyway) but this is a limit of the X Server's and DRI's structure. nVidia has got a workaround in their proprietary driver, but every other driver has got the same problems as well (except for video-playback with Xv. This seems to be correctly redirected on most other drivers except on fglrx. I have no idea why though...).

Darned, beat me to the punch . I don't know why there isn't some sort of sticky regarding the current limitation of X and DRI regarding this issue, so people would just stop asking about it and be better informed. Everyday I see at least 2 threads on the subject when lurking on the many distro support forums.

[EagleDM]

Sorry Melcar, IT IS how you say it is because there is NO sticky explaining the limitations of DRI and I am not a super uber-geek on Linux (yes I am an uber-geek on windows xp/vista) So, I had to find out about it.

Should be good to prepare a simple Sticky explaining all these so you don't have to suffer from it, but you can't blame me for asking something that is not being said in the open in the forums and I don't know everything.

by the way, I made the assumption that this could be fixed because I came from NVIDIA (I recently purchased a pair of 4850) and in the NVIDIA world, the vertical sync is PERFECT with the current standards and every 3D application I opened was in synch with the cube so I assumed this could be do-able on ATI hardware.

blame me for not knowing !


Daniel

[Christian_L]

Fedora support, new xorg & new kernels

I'm quite disappointed that ATI has not managed to support the current Fedora release for the third time in a row... Yes, I did downgrade to xorg 7.3, and fglrx works now, but I really would love to have my system up-to-date. After watching the fglrx releases for about 1.5 year now, I have the impression that ATI is constantly surprized by and not at all prepared for new distro releases and new xorg or kernel versions.

So I'd like to get some insight whether this is really the case: Do you have some internal policy to not look at any beta versions/release candidates at all? Or are the changes so profound that it just takes you so long to adapt the drivers? For example, will you be able to support xorg 7.4 once it comes out, or will that only be the time when you guys will have a first look at it and try to figure out what's new? Do you work with kernel development to know what's coming up ahead?

[RealNC]

My questions are quite simple: Why do the drivers suck so much? Why is 2D so slow? Why is 3D so slow too? Why does it crash when going full screen with video? Why does it crash when switching to a console? Why is ATI porting the Blue Screen of Death to Linux? Why does the screen get transformed into a gazillion of random pixels when switching to fullscreen in a YouTube video? Why should a Linux user even consider buying a new ATI card instead of an NVidia one?

Some of those questions are rhetorical. OK, most of them are. Let's say all of them are. But here's a real question:

Why on earth can't you deliver a working Catalyst driver yourself and instead ask others (Novell) to do it for you? No knowledge? This is AMD, how can you have no Linux knowledge?

[_txf_]

My questions are quite simple: Why do the drivers suck so much? Why is 2D so slow? Why is 3D so slow too? Why does it crash when going full screen with video? Why does it crash when switching to a console? Why is ATI porting the Blue Screen of Death to Linux? Why does the screen get transformed into a gazillion of random pixels when switching to fullscreen in a YouTube video? Why should a Linux user even consider buying a new ATI card instead of an NVidia one?

Some of those questions are rhetorical. OK, most of them are. Let's say all of them are. But here's a real question:

Why on earth can't you deliver a working Catalyst driver yourself and instead ask others (Novell) to do it for you? No knowledge? This is AMD, how can you have no Linux knowledge?

I'm not an ati dev but I'll answer some of your questions seeing as they show a slight misunderstanding of the issues at hand.

2d slow? I currently own a nvidia card and I have to say that on 2d nvidia sucks more than fglx (a lot more...comparing a radeon 2600 vs geforce 8600), the free ati driver is much faster at 2d.

3d? It is worse than nvidia but I haven't recently seen massive performance delta's between ati and nvidia.

BSODs are preferable to hard locks (haven't seen this bsod you're referring to tho...)

Youtube uses flash...flash sucks (and sucks massively under linux), I hear 10 is much better tho

Novell is making the free radeonhd driver not the Catalyst (which has improved massively lately).

Hopefully this will save an ati dev from answering the less enlightened questions

[RealNC]

I'm not an ati dev but I'll answer some of your questions seeing as they show a slight misunderstanding of the issues at hand.

2d slow? I currently own a nvidia card and I have to say that on 2d nvidia sucks more than fglx (a lot more...comparing a radeon 2600 vs geforce 8600), the free ati driver is much faster at 2d.

I'm not talking about the free driver. I'm talking about Catalyst. If the free driver is faster, why isn't Catalyst too? It's under AMD's direct control.

3d? It is worse than nvidia but I haven't recently seen massive performance delta's between ati and nvidia.

But it's still worse than nvidia. Doesn't have to be "massive".

BSODs are preferable to hard locks

I don't know why you prefer hard locks, or any locks (to each his own, I guess), but for me clean and stable operation is preferable, not hard locks or BSODs.

(haven't seen this bsod you're referring to tho...)

When I switch from X to a VT, it's a gamble if fglrx is actually gonna make it. If not, screen goes either black or blue and only reboot helps. There's no text message in "BSOD" though, but knowing ATI, they're probably working on it ("Blue Screen of Death for Linux (TM)").

Youtube uses flash...flash sucks (and sucks massively under linux), I hear 10 is much better tho

Flash sucks but works 100% with Intel drivers, open source ATI drivers, and closed nvidia drivers.

It sucks with Catalyst.

Novell is making the free radeonhd driver not the Catalyst (which has improved massively lately).

Why isn't AMD then making the Catalyst driver good? Why do they need Novell to deliver a working driver? What is it that Novell can do but AMD can't?

Hopefully this will save an ati dev from answering the less enlightened questions

As I said, the questions are rhetorical. "Enlightened?" They're supposed to be criticism. I'm not an AMD fanboy, so I guess that's why your idea of what's "enlightened" differs from mine.

[bridgman]

Thanks

I'll try to answer the rest, although I might skip the "why do the drivers suck so much ?" question for now.

We are chasing a fullscreen-related bug (reported here first I think). Not sure if we have been able to reproduce it in house yet but we know about it.

Switching to a console is always a pain to keep running since it depends on multiple drivers making the same decisions when programming the chip. The current X/Linux mechanism is not great. We are trying to support the transition to kernel modesetting, at least for the open drivers, which seems to be the best "real and permanent solution". Fitting kernel modesetting into fglrx is more cmoplicated so not sure what will happen there.

We are not porting the Blue Screen of Death to Linux. Under Linux it will be green and that part of the driver will be all new code

re: 2D, my understanding is also that we suck less than the competition. There were some driver-related performance issues but I think we've nailed most of them... most of the remaining performance pain seems to be with Firefox -- I don't know the details but reading between the lines it seems to be trying to do things in software rather than making use of acceleration.

As _txf_ said, the Novell effort is something completely different -- we worked with Novell to kick off the initial 5xx/6xx open source driver effort. We see the open source drivers as being important to aallow a good out-of-box experience with distros and to allow the X framework to evolve quickly (since the drivers and framework need to change at the same time).

I'm not talking about the free driver. I'm talking about Catalyst. If the free driver is faster, why isn't Catalyst too? It's under AMD's direct control.

The radeonhd driver is not faster, except in cases where shadowfb (software rendering into a frame buffer in system memory) is faster than hardware acceleration. The radeon driver probably is faster if you are running with EXA acceleration and the latest X server code which has some nice glyph cache optimizations.

But it's still worse than nvidia. Doesn't have to be "massive".

The current situation as I understand it is that NVidia drivers are relatively faster in a CPU-limited scenario while ATI drivers are faster in a GPU-limited scenario. That's what the benchmarks seem to say anyways -- and it fits the different hardware design approaches between the two companies.

Flash sucks but works 100% with Intel drivers, open source ATI drivers, and closed nvidia drivers. It sucks with Catalyst.

I hadn't heard this before, will look into it. Are you running the latest version of Flash (from a month or two ago) ?

Why isn't AMD then making the Catalyst driver good? Why do they need Novell to deliver a working driver? What is it that Novell can do but AMD can't?

Again, totally different scenario. The two drivers are aimed at different uses and markets. The radeonhd driver is primarily modesetting today with basic acceleration, although we are close to getting full acceleration in for 5xx and are working on 6xx.

[RealNC]

Well, as you can imagine, I own an ATI card (X1950XT 512MB PCIe). I wouldn't bother going on rants here otherwise.

When I bought the card back then it cost a fortune (top of the line for single-GPU cards for that generation). And then? It can't even play videos correctly. Tearing 'till your eyes hurt. And all the other bugs I mentioned.

And then, when I get out my ancient Matrox out of the closet, everything works. If an ATI card that is 15 times more expensive and 30 times more powerful gets beaten by an ancient, cheap Matrox, only one things comes to mind: ATI sucks.

PS:
All your analysis is good and fine. I respect that you're trying to achieve something. *But*, this is going on for *years* now. When I boot into Windows, the card flies. Everything is perfect. In Linux, only my Matrox can deliver that experience. What on earth was ATI doing all these years? I'll tell you what they did: they didn't care a bit. No, NVidia wasn't really better either. Wintendo has the 90%+ market share so who cares about the rest anyway.