Slashdot Mirror
AMD Offers a Performance Boost, Over 20 New Features With Catalyst Omega Drivers
MojoKid writes: AMD just dropped its new Catalyst Omega driver package that is the culmination of six months of development work. AMD Catalyst Omega reportedly brings over 20 new features and a wealth of bug fixes to the table, along with performance increases both on AMD Radeon GPUs and integrated AMD APUs. Some of the new functionality includes Virtual Super Resolution, or VSR. VSR is "game- and engine-agnostic" and renders content at up to 4K resolution, then displays it at a resolution that your monitor actually supports. AMD says VSR allows for increased image quality, similar in concept to Super Sampling Anti-Aliasing (SSAA). Another added perk of VSR is the ability to see more content on the screen at once. To take advantage of VSR, you'll need a Radeon R9 295X2, R9 290X, R9 290, or R9 285 discrete graphics card. Both single- and multi-GPU configurations are currently supported. VSR is essentially AMD's answer to NVIDIA's DSR, or Dynamic Super Resolution. In addition, AMD is claiming performance enhancements in a number of top titles with these these new drivers. Reportedly, as little as 6 percent improvement in performance in FIFA Online to as much as a 29 percent increase in Batman: Arkham Origins can be gained when using an AMD 7000-Series APU, for example. On discrete GPUs, an AMD Radeon R9 290X's performance increases ranged from 8 percent in Grid 2 to roughly 16 percent in Bioshock Infinity.
Why are they comparing it to 13.12, which is a year-old driver package? Is is because it isn't actually that much quicker than 14.9 or 14.11?
I don't need any of those new fancy features.
What I need is their OpenGL driver complying with the OpenGL specification. Whenever I do anything advanced, it reliably works on my Nvidia system and reliably needs annoying and performance-degrading workarounds on my AMD/ATI system. We're talking about stuff like simple branching in shader code causing the optimizer to emit returns, or unnecessarily having to feed in vertex data when the geometry could be deduced from gl_VertexID alone, or the Uniform Buffer Object layout specified in the shader not being preserved when using the binary shader format (means I have to recompile it every time), or atan in shaders yielding results that are half a degree off under some circumstances, or the builtin attributes not working if you use your own attributes with names that would alphanumerically be sorted before gl_*.
Through the ATI support forum back then I once got in touch with a technician who looked up these things in the driver sources and confirmed a few bugs with me, but later on I only got automated responses stating that he is leaving the company, and then the forum was trashed and a new AMD forum put in its place.
Yes. This is driving me nuts. On Nvidia, it all just works as it should.
The rendering at higher resolution then down-scaling without the game being aware of it is a pretty dreadful idea, you're just going to get tiny interfaces in most games or, as apparently pictured, a massive field of view which makes it harder to see smaller details. Microsoft's DirectX12 (or was it 11?) for mobile devices allows you to render the game world at higher or lower resolution and the interface at native, then merges them when displaying it; requires hardware support, apparently, but that seems like the best approach to scaling.
I'll this seems like the introduction of Eyefinity/Surround/Stereoscopic 3D/hardware PhysX. They'd be cool if games supported them properly, but since each implementation is different (and they have to wrestle with the Windows display system) it becomes easier to just ignore them. You're only likely to find Eyefinity/Surround in racing games, and physx where nVidia has paid to add it. The chances of many games going to the effort of supporting an upscaling hack seem pretty low.
AMD just dropped its new Catalyst Omega driver package
Is this a new meaning of the word drop, that I was until now unaware of? To my ears it sounds like they're not releasing anything.
-- Make America hate again!
it is understandable if Nvidia backs independent standards like OpenGL and Direct3D, rather than a vendor specific API that would always put them at a competitive disadvantage, because it is developed and controlled by a rival manufacturer.
I agree. From what I've seen, we're expecting Direct3D 12, and future OpenGL versions, to be such that they offer many of the benefits that Mantle and Metal offer now. Mantle seems to have been very effective in spurring innovation, but, as you say, there are legitimate reasons nVidia might not want to jump aboard.
If the tables were turned and nVidia were to make Physx available to AMD, AMD would always be playing catch-up: there's a conflict of interest if one party has total control. The frameworks from Khronos and Microsoft seem a better bet.
On top of all that, it's probably not great to have yet another graphics API, from either the driver-devs' perspective or the game-devs' perspective: devs already have to worry about Direct3D, OpenGL, OpenGL-ES, and console-specific APIs, without adding Metal and Mantle.
the difference rarely exceeds 10%, and is often just 5% or less. For developers that are not AMD "business partners", it might not be worth spending time on porting their engine to a 3D graphics API that makes only a minor improvement on the products of a single GPU vendor that has a market share of 29-30%.
Also good points. I don't know the numbers for the Metal API.
http://www.dsogaming.com/news/...
http://www.overclock3d.net/art...
Om, nomnomnom...
So it's their last driver release ever?
Escher was the first MC and Giger invented the HR department.
I am not even remotely an expert on the matter, but I believe the point is to render at higher resolution and then reduce AA and such.
If you can make the game engine render the content at such a high resolution then there less need to do post processing of the images to do things like smooth edges as long as you have a way to efficiently scale down the image.
For instance the new Final Fantasy XIII PC port has no graphics options (update to fix that is supposed to be out tomorrow). You can't even pick resolution let alone AA and other settings, so you are stuck at 720p. So the community released a tool (GeToSaTo) that forces the game to 4k resolution and then downsamples it, so that even without AA enabled everything looks as it would if it were.
So the question is, is it harder for a GPU to render the scene at 4k with less post processing or at 2k with lots of post processing turned on to equal image quality? Apparently AMD and nVidia both have it now so at least some games must either look better or are more efficient using downsampling.