No anti-aliasing in WebGL
Reported by steven.w...@gmail.com, Nov 3 2012
UserAgent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_8_2) AppleWebKit/536.26.14 (KHTML, like Gecko) Version/6.0.1 Safari/536.26.14 Steps to reproduce the problem: 1. Open one of the links below 2. Examine the edges of the renderings http://mrdoob.github.com/three.js/examples/webgl_geometry_colors_blender.html http://acko.net/files/tubecurves/ http://web.chemdoodle.com/tutorial/3d-structure-canvases/transformer3d-canvas What is the expected behavior? Anti-aliasing What went wrong? No anti-aliasing Did this work before? Yes Chrome Canary, couple days ago Chrome version: <Copy from: 'about:version'> Channel: n/a OS Version: OS X 10.8.2 I don't know where else to voice this, so I'm adding it here for whom may care: It is absolutely retarded that in an age of MSAA/CSAA/EQAA/FXAA/GBAA/GPAA/SRAA/.. we only have an "antialiasing: true/false" flag for WebGL and web developers are locked out of yet another feature that native APIs provide full control over. How can we expect web apps to rival native when this sort of dumbing down is completely routine? The developer is unable to control or even determine the quality of the rendering, for example the number of color / coverage samples. If a browser (e.g. mobile) were to decide to implement a post-process filter like FXAA as a cheap alternative, you'd never know, yet it would drastically reduce the quality of e.g. WebGL rendered text/UI unless you separate it into a another rendering pass. This is hugely important for example for rendering 3D vector graphics with WebGL, which I'm currently working on. Unlike photorealistic scenes games, you really need at least 8xMSAA (or CSAA/EQAA equivalent) for it not to look like ass. If I could turn on 16xMSAA, it would be even better. This is doubly problematic because the antialiasing flag doesn't just control multi-sampling, but has a a huge effect on the quality of rendered lines too: without anti-aliasing, the end caps are always horizontal/vertical, and the line width for diagonals is reduced by 40% compared to straights. Why are these two quality features linked? Oh right, because I'm treated like an idiot instead of a professional graphics developer :/.
Nov 3 2012,
Apologies, the browser crashed while entering the issue and I forgot to repaste in the version: This was on: 25.0.1314.1 canary
Nov 3 2012,
Nov 5 2012,
Nov 5 2012,
Unfortunately there is a severe bug in the NVIDIA OpenGL driver on Mac OS X which required disabling of multisampling for WebGL to avoid information leakage. I suggest you contact Apple Computer and push them to fix: Radar 12557679: Corruption of OpenGL Multisample Renderbuffers with NVIDIA chipsets which contains a more reduced test case for a previously filed bug: Radar 11873444: OpenGL texture corruption in low VRAM situations You will soon find that every WebGL implementation on Mac OS, not just Chrome's, is going to have to prevent the use of multisampling when running on NVIDIA chipsets. Please raise your concerns about the design of the WebGL API, in particular the context creation attributes, on Khronos' public_webgl mailing list. I'm closing this as a duplicate of Issue 137303 , which originally uncovered the need to disable multisampling.
Jan 11 2013,
Issue 169450 has been merged into this issue.
Jan 29 2013,
I have AMD card. Why should I care about security issue in Nvidia's drivers? Is there any switch to turn it on? Something like "I understand about security issues, I don't care, turn webgl AA" in Chrome settings?
Jan 30 2013,
You can goto about:flags and turn off software rendering list.
Mar 11 2013,
Apr 5 2013,
Apr 6 2013,
Sign in to add a comment