RichardC (tich.delete@this.pobox.com) on May 15, 2013 11:13 am wrote:

>
> > Page rendering is limited 1st by network by far, 2nd by I/O. CPU is a far 3rd order effect.
>
> It shouldn't be doing any (non-network) IO at all, so your 2) is questionable. And I

Uh? Browsers use memory and *disk* for caching.


> don't think it's all about network bandwidth. Even on fast local network some kinds of
> pages take noticeable time to render. And I've read of various efforts to speed it up,
> using threading or GPGPUs, which suggests that other people than me think it's a real

For those people who have CPUs like an Atom, a Bobcat or an A7, it is an issue.
Ie, my Atom can choke quite a bit on some pages. And since it has SMT, a threaded HTML rendered would be quite welcome.

But we were discussing CPU cores with top single threaded performance, like an Ivy Bridge, were we not?


> issue. Present some evidence for your analysis.

I can only present anecdote.
My Atom can get jerky while rendering some web pages. Ie, scrolling freezes, etc.
Among the worse case offenders, Slashdot articles with lots of comments and going very down on Facebook's timeline.

OTOH, on the same pages my Sandy Bridge simply never hits 100% CPU for a measurable time and it never feels less than butter smooth.
Any JS CPU profile say 90% of the time is spent idling.

Even on the cheapest AMD A4 (dual core Bobcat) I could buy, it always feels smooth.

Can you present evidence otherwise?