Slashdot Mirror
AMD's 'Crimson' Driver Software Released (anandtech.com)
An anonymous reader writes: Yesterday marked the launch of AMD's 'Crimson' driver software. It replaces the old Catalyst driver software, and represents a change in how AMD develops bug fixes, improves performance, and adds features. AnandTech took a detailed look at the new driver software. They say, "By focusing feature releases around the end of the year driver, AMD is able to cut down on what parts of the driver they change (and thereby can possibly break) at other times of the year, and try to knock out all of their feature-related bugs at once. At the same time it makes the annual driver release a significant event, as AMD releases a number of new features all at once. However on the other hand this means that AMD has few features launching any other time of the year, which can make it look like they're not heavily invested in feature development at those points." On a more positive note, the article adds, "Looking under the hood there's no single feature that's going to blow every Radeon user away at once, but overall there are a number of neat features here that should be welcomed by various user groups. ... Meanwhile AMD's radical overhaul of their control panel via the new Radeon Settings application will be quickly noticed by everyone."
Phoronix comes to the same conclusion:
http://www.phoronix.com/scan.php?page=article&item=amd-crimson-linux&num=1
Maybe this is why Windows 10 removed the old Catalyst Control Center a few days ago? A bit early, sure, but still...
Not the driver, that's out, but that they are going to change how they do drivers. They've said that numerous times before, and always the situation is the same. They are very slow at getting actual release drivers out (they are forever beta versions) and their OpenGL performance and support is garbage (to the point that HFSS would fail to run on systems with AMD cards).
So AMD: Less talk, more good drivers. I want to support you, I really do, but I've been burned too many times.
It wouldn't surprise me if at some point the open source driver would catch up to the closed source one. AMD would do well to direct their driver-related efforts in converging the two. That is: if they feel a closed source driver is needed, base it on open source components as much as practical. And put improvements back into those open source parts where possible.
From what I've read, AMD has already made some moves in that direction. Which is a good thing. Shared effort (community <-> AMD), limited resources, etc. Regardless of what products AMD kicks out, software is a significant part of making those products successful. And the open source crowd should NOT be ignored in that process (luckily that's not the case, but hey there's always room for improvement :-). Even if it were just a way to offload some of the work to 3rd parties.
The problem is that the "driver" nowadays isn't really a driver. The hardware still pushes the same shaders etc. to the card, over a standardised bus.
The problem is that the "driver" nowadays is a bunch of shortcuts and re-optimised shaders for particular operations, which are heavily dependent on how the games operate and basically "overrules" what the game wants the shaders to do, for the sake of per-game performance increases by sacrificing things that are sub-optimal on that particular card / game combination. Why else do you think that "new game X" suddenly needs a driver update to work when the game is using DirectX and the card is compliant with that level of DirectX?
In essence, this is tied quite tightly to DirectX. So the reason that the "drivers" often suck is that they are Windows-specific bodges to increase performance for individual games. That won't translate to even a Linux/OpenGL port of the same game on the same hardware, let alone for EVERY OpenGL game on EVERY Linux on ALL supported hardware.
And the investment is not in making a particular hardware faster, or pushing more texels over the standardised buses than before, but in optimising the hardware response for a particular game - which is labour-intensive and has to be redone for every game on every platform for each supported card.
AMD gets knocked for their drivers, but you have to wonder how much is due to intervention from the competition? I still remember Unreal Tournament 2003 would start the game with a character breaking through an nVidia logo. If games are optimized for nVidia hardware, can we really tell if performance of AMD is that much worse? Is there a test that is guaranteed not to favor any card?