Mesa 13.0 Released With Intel OpenGL 4.5, RADV Radeon Vulkan Driver

[phoronix]

Phoronix: Mesa 13.0 Released With Intel OpenGL 4.5, RADV Radeon Vulkan Driver

Mesa 13.0 was released today as what is likely the most significant release to this 3D graphics driver/library implementation ever! Mesa 13 is huge for open-source driver uses particularly with Intel, Radeon, and GeForce hardware driver improvements. During development this release was known as Mesa 12.1

http://www.phoronix.com/vr.php?view=23709

[Holograph]

The RADV driver works in conjunction with the AMDGPU kernel driver meaning support by default for GCN 1.2 GPUs and newer or GCN 1.0/1.1 if you manually enable the experimental AMDGPU support in the kernel.

Am I the only one who thinks it completely unreasonable that CIK still needs an experimental flag enabled? Why does AMD not seem to realize how awful the state of Linux graphics with their cards are? Owners of these cards should be able to pick any distro with a reasonably recent kernel (say 4.8) and not have to download (or make) any special builds of the kernel to use the AMDGPU driver. Saying to use Radeon is just not good enough. We paid for these cards and we deserve for things to just work with a reasonable driver with reasonable performance.

[eydee]

Am I the only one who thinks it completely unreasonable that CIK still needs an experimental flag enabled? Why does AMD not seem to realize how awful the state of Linux graphics with their cards are? Owners of these cards should be able to pick any distro with a reasonably recent kernel (say 4.8) and not have to download (or make) any special builds of the kernel to use the AMDGPU driver. Saying to use Radeon is just not good enough. We paid for these cards and we deserve for things to just work with a reasonable driver with reasonable performance.

GCN1 will probably go legacy instead of gaining official support. They're 4 years old now, some rebranded Terascale was only 1 year old when it happened. At least there's "experimental" support for those that need it.

[CrystalGamma]

Am I the only one who thinks it completely unreasonable that CIK still needs an experimental flag enabled? Why does AMD not seem to realize how awful the state of Linux graphics with their cards are? Owners of these cards should be able to pick any distro with a reasonably recent kernel (say 4.8) and not have to download (or make) any special builds of the kernel to use the AMDGPU driver. Saying to use Radeon is just not good enough. We paid for these cards and we deserve for things to just work with a reasonable driver with reasonable performance.

Is the radeon kernel driver “unreasonable” or does it significantly impact the performance of your apps?
And as for Vulkan, I don't think asking for a flag to use an API (Vulkan) that has basically zero applications using it on the relevant platform is unreasonable.
Also, you could make the argument that these cards came out long before Vulkan even had a name.

[andre30correia]

well time to test to see if works better in skylake than mesa 12.0..3 with sna and/or glamor

[schmidtbag]

Am I the only one who thinks it completely unreasonable that CIK still needs an experimental flag enabled? Why does AMD not seem to realize how awful the state of Linux graphics with their cards are? Owners of these cards should be able to pick any distro with a reasonably recent kernel (say 4.8) and not have to download (or make) any special builds of the kernel to use the AMDGPU driver. Saying to use Radeon is just not good enough. We paid for these cards and we deserve for things to just work with a reasonable driver with reasonable performance.

I doubt you're the only one who thinks this but you're certainly in the minority. As an owner of a GCN 1.1 card, I personally am grateful at the rate of which AMD has been keeping up with these GPUs. Remember, you're whining about a company that is 10% the size of Intel, does not expect to profit much from Linux users, and yet has drivers that are arguably better than Intel's.

So if there is anything that is unreasonable, it's your expectations. Your argument of "we paid for these cards and deserve for things to just work..." is invalid. For example, take a look at a box of a R9 270X:

http://www.pcgameware.co.uk/images/MSI-270X-GAMING-OC-box-back.jpg

Notice the stark lack of Linux being mentioned anywhere, but it does explicitly state support for Windows. Therefore, they are not obligated to support us. I have never seen a box to a Radeon GPU that stated official Linux support, though I would like to be proven wrong. However, they are obligated to support FirePro users, which they do just fine (fglrx OpenCL performance is great).

[OneTimeShot]

RE: Nouveau and CTS compliance...

...surely if someone finds that Mesa is failing a CTS compliance test, but is passing Piglet tests, all they need to do is to produce an equivalent Piglet test to demonstrate the bug. This is no different to other software that not everyone has a license for...

[imirkin]

RE: Nouveau and CTS compliance...

...surely if someone finds that Mesa is failing a CTS compliance test, but is passing Piglet tests, all they need to do is to produce an equivalent Piglet test to demonstrate the bug. This is no different to other software that not everyone has a license for...

Of course that'd require someone with access to spend time and test CTS on nouveau, and do debugging to figure out how to write a piglit test to exercise that particular failure.

Or just make an apitrace, that has been successful in communicating issues thus far.

[R00KIE]

GCN1 will probably go legacy instead of gaining official support. They're 4 years old now, some rebranded Terascale was only 1 year old when it happened. At least there's "experimental" support for those that need it.

That would be reasonable if new laptops weren't being sold with GCN1.0 cards. At the rate new models are introduced you can't in good conscience say that these are old machines. I own one of these machines and right now support for the AMD dGPU is broken with the radeon driver, so I'm pinning my hopes on the problem getting addressed sometime in the future in the amdgpu driver and getting official support.

[bridgman]

Am I the only one who thinks it completely unreasonable that CIK still needs an experimental flag enabled? Why does AMD not seem to realize how awful the state of Linux graphics with their cards are? Owners of these cards should be able to pick any distro with a reasonably recent kernel (say 4.8) and not have to download (or make) any special builds of the kernel to use the AMDGPU driver. Saying to use Radeon is just not good enough. We paid for these cards and we deserve for things to just work with a reasonable driver with reasonable performance.

Other than radv (which we had not planned on) what is it that you want from amdgpu ? You are talking about performance and "just work"ing but performance is essentially the same between radeon and amdgpu kernel drivers and all the test coverage is on radeon.