4K TVs Continue To Rise Despite TV Shipments Dropping

[BoFox]

Oh yes, that's right - shaders!

On a 980 574ms for the merge stage alone depending on the scene. Tri SLI with perfect scaling would get you just under *six* frames per second. Keep acting like you got a clue kid.

GPUs can do amazing things with optimizations in architecture. Threaded parallelism and compression, rather than snail-paced serialization here... and you got it flying like in some cases where a GTX 280 crawls around at 8fps compared to a GTX 480 zipping at 100+fps - it's not hard to find a game with certain settings where that is the case.

Looks great, except for the artifacts as explained in Figure 10.  I wish the article would have made it more clear how often those artifacts would appear in games.

Simply a factor of granularity used in stage one and not truncating too much in the merge stage. In other words, you have a dial you can use- reducing artifacts close to zero would probably be several orders of magnitude more intensive then having a small margin of error- edge cases, as always, will be your undoing here.

The question is - by how much?
The paper tries to paint a positive picture here:
"Due to the fragment synchronization issues noted above, overdraw in areas of high depth complexity introduces serialization that has a large performance impact. We minimize this with a z-prepass to cull triangles. Our prototype uses an 8xMSAA pass to find the maximum z per pixel, and we discard more distant fragments during our merge. Because of the difference in sampling rates, this is not fully conservative and can cause the background to leak along geometric cracks (see Figure 10(d)). A more conservative z-prepass would solve this issue; either higher hardware MSAA, analytically computing the maximal depth per-pixel, or matched sample rates would work."

Blurred texture, incorrectly merged, incorrectly discarded, and z-prepass artifacts probably won't be a common enough thing but I expect Nvidia to spend a lot of time making sure it's not an issue in the games we play (the z-prepass artifacts the most of all, especially).

This isn't from nVidia, they are hosting a doc that was written by what appears to be some University students. Nothing wrong with that of course, but it isn't something nV's developers are likely to be actively working on. RAM is cheap, TFLOPS of spare processing power, not so much.

Hmm, NV hosting it on their site probably means that NV is indeed interested and looking into it right now. Of course, they'd love to develop a new AA method that keeps the spotlight away from AMD.

Nevertheless, I think both NV and AMD will be seriously focusing on bigger resolutions this generation.  5K is insanely demanding though (unlike how 5K sounds compared to 4K - a kinda lame naming scheme) but AMD and NV will have to try their best to stay ahead in case 5K monitors take the spotlight with review sites.

5K will likely pick up about as much traction as 2K- in other words, very little. The reason 4K is such an easy swap is it is a clean pixel perfect downsizing for 1080p content. 5K won't fail, but 4K will be *the* standard until 8K picks up.

Good point, agreed. Well, 5K will likely be as "important" as 4K is right now, at least, lol (as 4K is still a "niche"). Review sites love to bench at the bleeding edge, and fanboys love to use the bleeding edge as an indicator of the potential.