[Public WebGL] Re: webgl/swiftshader

Liam Wilson [email protected]
Sun Feb 19 11:20:22 PST 2012

But the majority of people don't have even a mid range discrete GPU. They have an integrated GPU (and if http://unity3d.com/webplayer/hwstats/pages/web-2011Q3-gfxcard.html is anything to go by, they don't even have a good integrated GPU). Even if the do happen to have an okGPU there is still the issue of out dated drivers/missing GPU features.There's stats on http://people.mozilla.org/~bjacob/gfx_features_stats/ (again, take these with a pinch of salt), only about 10% of win XP users currently stand a chance of getting decent WebGL performance (layers acceleration enabled).

It's not just a matter of raw performance. Rendering only needs to be fast enough to make applications usable. At a modest resolution and frame rates this is entirely possible with software rendering. There are far too many WebGL demos out there that attempt to render 60 fps at full screen resolution. 30fps at 640x360 would be perfectly good.

 From: Florian Bösch <[email protected]>
To: Liam Wilson <[email protected]> 
Cc: public webgl <[email protected]> 
Sent: Sunday, 19 February 2012, 16:06
Subject: Re: [Public WebGL] Re: webgl/swiftshader

On Sun, Feb 19, 2012 at 4:32 PM, Liam Wilson <[email protected]> wrote:

I think "software rendering is slow" is a bit of a self fulfilling prophecy. No one thinks it can be fast, so no one bothers to write a fast software renderer. So we then wind up drawing comparisons between unoptimised CPU renderers, and highly optimised GPUrenderers. Which then makes the performance gap look far worse than it actually is.

1) Software rendering *is* slow. Mid-end desktop GPUs output around 500 million triangles per second and about 30 billion texels per second (that would be 14'000 full-hd screens per second).
2) Swiftshader is a good/fast implementation of software rendering, but you cannot expect this to ever reach anywhere actual GPU performance.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://khronos.org/pipermail/public_webgl_khronos.org/attachments/20120219/fce345c2/attachment.html>

More information about the public_webgl mailing list