Dropping pre-i686 from the archive

Chris Jones cmsj at tenshu.net
Thu Jan 12 17:16:43 GMT 2006


On 4:52:55 pm 12/01/2006 Eric Dunbar <eric.dunbar at gmail.com> wrote:
> Why doesn't someone with the wherewithal to compile various packages
> do a benchmarked comparison of packages compiled for i686 vs i386?

Even the gentoo people have never justified all their compiling-for-speed.

> It shouldn't be the hardest thing in the world and it seems that the

how do you benchmark firefox? capturing timing data would be pretty hard,
so I suspect that the overall task could be quite hard.

It would be very easy to show that 686 optimised algorithms run faster, but
much of the time applications aren't running complex algorithms, they're
waiting for information from the user or from disk.

> i386 and i686 comparisons. So far, all I've seen in this thread are
> anecdotal statements regarding speed (and, given that FireFox and

I've seen a fair bit claiming that the firefox that ships with breezy is
slower than the mozilla.org ff1.0.7 binary. I'm not sure how anyone
verified that to claim it though.

> PS Do "Via C3" CPUs not run i686 compiled packages? (I know nothing

As I understand it they do not implement the full 686 instruction set and
perform poorly with other 686 instructions.

> about the Via C3) Also, how commonly are these Via C3 CPUs used? Are

I'm not sure how many there are tbh, but I would suspect it would be enough
to be statistically relevant.

> they in embedded solutions or in (formerly) high-end machines? If so,

They probably are used in embedded solutions, but generally I see them pop
up in mini computers (e.g. the mini-itx box connected to my TV).

> then what's the problem with _not_ supporting them out-of-the-box, IF

the box connected to my tv will stop workign ;)

Chris Jones
  cmsj at tenshu.net

More information about the ubuntu-devel mailing list