Browser v8 versus shell in Kraken speed difference.
35 views
Skip to first unread message
Joe Millenbach
Oct 19, 2012, 4:09:09 PM10/19/12
to v8-users
[Cross posting between Chromium and v8 as related to both]
I wondered if anyone knows why running some Kraken sub tests in a
browser go faster than if you run the same test from a command line
shell. Like by 20-30% sometimes. We expected the browser to only add
overhead, like it does for the web based SunSpider.
I should note we don't have exactly the same version of shell and
browser v8 engine (shell is 3.8.9 and browser says 3.8.10).
I wondered if anyone knows why running some Kraken sub tests in a
browser go faster than if you run the same test from a command line
shell. Like by 20-30% sometimes. We expected the browser to only add
overhead, like it does for the web based SunSpider.
I should note we don't have exactly the same version of shell and
browser v8 engine (shell is 3.8.9 and browser says 3.8.10).
Jakob Kummerow
Oct 20, 2012, 8:41:39 AM10/20/12
to v8-u...@googlegroups.com
Hard to say without knowing more about what you did and what you observed. E.g. if your browser was an x64 build and your command-line V8 was an ia32 build, that would explain all sorts of differences. Telling us *which* tests exactly were faster in the browser would also add quite a bit of relevant information.
That said, V8 3.8 is ancient. I don't think anyone really cares about its performance any more. Please repeat your tests with the current development version (V8 trunk or bleeding_edge, Chromium M24 trunk or Chrome Dev channel).
--v8-users mailing list
v8-u...@googlegroups.com
http://groups.google.com/group/v8-users
Joe Millenbach
Oct 20, 2012, 3:05:40 PM10/20/12
to v8-u...@googlegroups.com
I originally assumed both would be x64, but I'm not sure if Chromium installed through Ubuntu's software center is 32 or 64 bit. The shell was 64 for sure. Thanks for the additional parameter to check.
We picked these versions because that is what installed on a regular user machine, but on my Windows box it has 3.12.19.15, so maybe a newer version isn't such a bad test. And as you say nobody there likely cares about the difference in this version.
The subtests that differ don't follow me across different CPUs (Ivy Bridge vs. Llano vs. ...), but there often seem to be a couple that are faster in the browser. Originally we found this on an AMD Bobcat (E 350) CPU, and the two tests were ai-astar and audio-oscillator. Ai-astar was about 25% faster and audio-oscillator I think was even higher. I don't have the machines or data in front of me today to say more conclusively, but maybe this can spark something. I think audio-oscillator might have been common to the lists.
Reply all
Reply to author
Forward
0 new messages