AMD Puts Out An OpenGL 4.2 Linux Driver

[DaemonFC]

I see you basically ignored everything I wrote....

No I didn't. But let's say you're right and FGLRX is being the proverbial princess and the pea (after it taints your kernel, and uses X to drive the hardware and uses DRI 1 which is from the late 90s and oriented around using one OpenGL application at a time?) Let's say it's possible for FGLRX to be too damned good after doing all of that. At some point it decides that everything has to be 100%.

That still sucks for the user. You're saying that Nvidia is bad because they run the application the user has executed and the user never has to be any the wiser. Explain to a user how their system just working is a bad thing?

Though I'm not at all sure I believe this is what is going on. If they're content with a driver stack the molests and mutilates the kernel and the X server just to run, and frequently take more than a year to respond to crash reports which are irrefutably caused by their drivers, I really doubt they care how OpenGL compliant they are. I'm no expert but if you see a house with siding that's falling off, a roof that's caving in, and car parts all over the lawn, you don't make good assumptions about what could be going on inside. But....I'm just a user.

[mirv]

No I didn't. But let's say you're right and FGLRX is being the proverbial princess and the pea (after it taints your kernel, and uses X to drive the hardware and uses DRI 1 which is from the late 90s and oriented around using one OpenGL application at a time?) Let's say it's possible for FGLRX to be too damned good after doing all of that. At some point it decides that everything has to be 100%.

Actually, I never said fglrx was perfect. I've even given examples of issues I've personally had with it. I have been trying to point out, however, that just because bugs exist with fglrx, doesn't magically make all other programs error free. Please stop trying to twist my words around - if nothing else, you're bad at doing it.

That still sucks for the user. You're saying that Nvidia is bad because they run the application the user has executed and the user never has to be any the wiser. Explain to a user how their system just working is a bad thing?

Because it runs outside of the OpenGL spec and locks the user in to using only nvidia drivers, with nvidia hardware. And if you can't grasp why that's extremely bad, then you really shouldn't be using linux. You should probably be running something from Apple.

Though I'm not at all sure I believe this is what is going on. If they're content with a driver stack the molests and mutilates the kernel and the X server just to run, and frequently take more than a year to respond to crash reports which are irrefutably caused by their drivers, I really doubt they care how OpenGL compliant they are. I'm no expert but if you see a house with siding that's falling off, a roof that's caving in, and car parts all over the lawn, you don't make good assumptions about what could be going on inside. But....I'm just a user.

So....you have problems with nvidia's blob too? Because everything you just wrote applies equally to that.

[Kano]

Who did try that driver yet? For me it was unstable, any other reports? I hope the "real" 11-8 driver, which should be 8.88x too is more stable. What i really hate most is that a system crashes due to a stupid gfx driver. That can happen with nv too with old xserver + vdpau flash accelleration...

[pingufunkybeat]

That still sucks for the user. You're saying that Nvidia is bad because they run the application the user has executed and the user never has to be any the wiser. Explain to a user how their system just working is a bad thing?

I wish it were just working. I've spent weeks battling on and off with a recent nvidia card under Debian at work. apt pinning, pulling things from experimental, cuda breakage. The driver works now, but I still have trouble building stuff using their sdk.

This is yet another of those myths.

[energyman]

people complain that AMD does not put out BETA drivers

When AMD does release BETA drivers, people complain and bitch around that they are not perfect.

Conclusion:

no matter what you do, people bitch around.

[Kano]

It has absolutely nothing to do with beta or release drivers. If AMD wanted they could stop their 12 driver/year crap. Basically when major changes are needed you have to wait for a minimum of 3 month, new xserver support about 5-6 mounth (just in time for the next U release). What do you gain with 12 half baked drivers? Absolutely nothing when you don't see an improvement, more likely you find regressions due to minimal testing. They should release drivers when they fix important things, no matter if the time between would be 1 week or 3 month, it just should work with current plattforms. Also the whole bunch of D3D fixes are completely unimportant for Linux, so why should all drivers released on the same day? There is absolutely no good logic behind that. When you look at Nvidia it does not really matter if a driver is tagged as beta, rc or final. In most cases Nv drivers are stable for general use (vdpau is tricky on old xservers), but not even thats the case for AMD...

[mirv]

It has absolutely nothing to do with beta or release drivers. If AMD wanted they could stop their 12 driver/year crap. Basically when major changes are needed you have to wait for a minimum of 3 month, new xserver support about 5-6 mounth (just in time for the next U release). What do you gain with 12 half baked drivers? Absolutely nothing when you don't see an improvement, more likely you find regressions due to minimal testing. They should release drivers when they fix important things, no matter if the time between would be 1 week or 3 month, it just should work with current plattforms. Also the whole bunch of D3D fixes are completely unimportant for Linux, so why should all drivers released on the same day? There is absolutely no good logic behind that. When you look at Nvidia it does not really matter if a driver is tagged as beta, rc or final. In most cases Nv drivers are stable for general use (vdpau is tricky on old xservers), but not even thats the case for AMD...

The decision for AMD to put their drivers in a baking pipeline actually likely gives them more than "minimal testing".
Don't forget as well, that OpenGL changes are made to the drivers, large areas of which are common to windows and linux, so releasing them on the same day makes sense.
1 driver per month is also a commitment by them - changes may be slower than your liking, but they do happen, and I challenge anyone to say that the fglrx drivers haven't seen rapid improvements since that was instigated.
I see a lot of people bashing AMD here, but without any real substance. If you don't like their drivers, at least give some meat to the argument. Also accept it when people bash nvidia (because they really aren't perfect either).

[xpressrazor]

Does it install and work on sandy bridge processors? Sandy bridge processors have its own graphics card, with ATI it is dual. It installed on Ubuntu 11.04 but had no 3D support. So I had to go through Natty Installation Guide to remove fglrx and reinstall the open source mesa version. I even tried the xorg-edgers but no real improvement on 3D support. Even on open source drivers the computer seems to ignore the ati card. I found similar bug report on the kernel so I commented on that. The most interesting part of the dmesg is

[ 15.949997] radeon 0000:01:00.0: Invalid ROM contents
[ 15.950088] radeon 0000:01:00.0: Invalid ROM contents
[ 15.950116] [drm:radeon_get_bios] *ERROR* Unable to locate a BIOS ROM
[ 15.950120] radeon 0000:01:00.0: Fatal error during GPU init
[ 15.950124] [drm] radeon: finishing device.
[ 15.950126] [TTM] Memory type 2 has not been initialized.
[ 15.951164] vga_switcheroo: disabled
[ 15.951383] radeon: probe of 0000:01:00.0 failed with error -22

At AMD's site they seem to have shut their mouth regarding this issue. From above message Linux kernel seems to dislike this dual graphics thing, so is it any help installing amd's driver. If you guys know some tricks, I can use to make the kernel believe I really have this card, would be of great help.

[PsynoKhi0]

What do you gain with 12 half baked drivers?

Priceless comments on Phoronix
Plus they are far from half-baked for everyone. *waves*

[smitty3268]

At some point it decides that everything has to be 100%.

That still sucks for the user. You're saying that Nvidia is bad because they run the application the user has executed and the user never has to be any the wiser. Explain to a user how their system just working is a bad thing?

Interestingly, this is the exact same argument Microsoft used to explain why it was a good thing that IE6 rendered horribly broken websites.

And to a degree, it is correct - if there is a minor error on the page/shader, why not have the smarts to automatically recognize and correct it.

The downside is that it's then impossible to implement the same hacks across competing software, and websites/shaders get worse and worse until they are completely locked in to a single solution. Just take a look at WINE, and how much effort it is taking them to try and support anything other than NVidia drivers. It's not because the other drivers are buggy, it's because WINE started making a lot of assumptions about how the drivers would work over the years that were never valid but just happened to work on one set of drivers.