02-16-2012, 11:16 AM
Regardless of whether or not mental health issues are an interesting topic of discussion, they're not relevant to this thread.
So, how 'bout those rendering differences, eh?
02-16-2012, 12:57 PM
If anybody is still reading this thread, would be great to repeat the test with nouveau/nVidia blobs/intel/intel driver on Windows/llvmpipe to compare the Enemy screenshots. I think it's difficult to establish which one is wrong between Catalyst and r300g/r600g without other data.
Originally Posted by mangobrain
02-17-2012, 12:16 PM
Supporting hybrid graphics well is mostly limited by a lack of infrastructure in the graphics stack to properly support decoupled rendering and display. The driver side is pretty trivial. The infrastructure to support it without ugly hacks is starting to fall into place:
Originally Posted by Sonadow
1. The initial v1 dma_buf patches landed in 3.3. This allows sharing buffers between drivers
2. Dave's "Prime" work builds on the dma_buf patches to allow buffer sharing between KMS drivers for things like hybrid graphics on laptops
3. Reworking the X server to handle decoupled rendering and display
Until all of that falls into place, it will always be a hack. There isn't really any magic involved. One GPU renders, the other GPU composites the result to the screen. Anyone can help out.
Last edited by agd5f; 02-17-2012 at 12:19 PM.
02-17-2012, 12:23 PM
AMD got it somehow working with their 12.1 blob by doing all hybrid stuff in userspace - the X server still thinks one of the two cards does everything. It's still a hack, but it seems a good one (however, nowhere as good as hybrid graphics should be).
02-18-2012, 07:42 PM
jpeg and small pictures. There is no need to comment on this article. Every difference can be explained by this two things alone.
Next time lossless and huge, please.
03-09-2012, 08:06 AM
Looks like Catalyst rendering is the wrong one: nVidia blobs vs. Catalyst. As you can see, it's nVidia blobs + r300g/r600g vs Catalyst.
03-09-2012, 10:40 AM
Catalyst is indeed rendering differently. But like I said before it's also possible that the bloom was originally written on Catalyst, without noticing that it relied on wrong driver behaviour.
In that case a combination of a Nexuiz bug and a Catalyst bug causes it to be rendered correctly on Catalyst only.