Effective speed of Phenom X3 is faster than Phenom X4 !!

I recently upgraded my AMD Phenom X3 8650 to an X4 9600 with the same
clock speed (2.3GHz).

Imagine my surprise when the X4 turned out to be dog slow on my
workload, which fundamentally is several (~5) firefox browser with a
few hundred tabs to load up. The X4 could not even keep my DSL going
at full rate, something which the X3 never had any problem
accomplishing.

I then switched back to the X3, and everything was much faster again,
and the DSL is maxed out. I even tried booting the X4 with the
maxcpus=3 (linux, Fedora10/64b) kernel option. Still slow.

I'm dumbfounded. Is there some problem with the 2MB L3 cache being too
small for the X4 case, and causing some wild trashing of the cache
that just kills performance?

I should add that X4 is 4 cores and X3 is 3 cores, in case anyone did
not realize.

The big question is whether the speed gets worse because of a Phenom
hardware
effect, or because Firefox somehow runs WORSE with 4 cores than with
one.
Right now I wonder whether the latter may be the case.

I recompiled Firefox 3.5.1. on the 4-core machine but that does not
seem to make any difference. Firefox also uses more memory
(at least temporarily) on the 4-core machine, up 2x as much easily.

reikred writes:
I recompiled Firefox 3.5.1. on the 4-core machine but that does not
seem to make any difference. Firefox also uses more memory
(at least temporarily) on the 4-core machine, up 2x as much easily.

This all sounds very unlikely to have anything to do with the CPU...

-Miles

--
「ã�™ã�£ã�”ã�„ã€�ã�¨å‘Ÿã�„ã�¦ã‚‹ã€‚「ã�¸ã‚“ã�¦cã �Šã�ªã‚‚ã‚“ã�°ã�£ã�‹ã‚Šã€�
「�ん���ん��？�
「�ん���ん���
ã��ã�†è¨€ã‚�れるã�¨è¦‹ã�Ÿã��ã�ªã�£ã�¦ã�—ã�¾ã�†ã€ ‚

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Miles Bader pravi:
I recompiled Firefox 3.5.1. on the 4-core machine but that does not
seem to make any difference. Firefox also uses more memory
(at least temporarily) on the 4-core machine, up 2x as much easily.

This all sounds very unlikely to have anything to do with the CPU...

IE8 spawns additional processes, depending on the number of cores in the
system (1 core = 1 process, 2 cores = 2 processes). Perhaps firefox has
something similar?

LP,
Jure

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFKjYmWB6mNZXe93qgRAlN6AJ9n5RgnAXT8vUmjV/LaKxu5owXezQCfc4Zl
GshIL8hQEiEhRDA1ODmGZPU=
=WEDf
-----END PGP SIGNATURE-----

I agree, it is likely not the CPU per se. More likely there is
something strange about firefox that causes it to trip up in massive
fashion on 4-core cpu.

The behavior is extremely reproducible. I can switch the cpu from X3
to X4 and back to X3 and then X4 again, and the slowdown is HUGE
whenver I plug in the X4.

I agree, it is likely not the CPU per se. More likely there is
something strange about firefox that causes it to trip up in massive
fashion on 4-core cpu.

The behavior is extremely reproducible. I can switch the cpu from X3
to X4 and back to X3 and then X4 again, and the slowdown is HUGE
whenver I plug in the X4.

Firefox is still single-process (but multiple threads) as of 3.5.x,
unlike Google Chrome and IE8.

There is a plan to make Firefox have one process per tab or maybe
window,
more like Google Chrome, but it has not yet been implemented.