[Pythonmac-SIG] Pystone numbers for different Macs...
daniellord at mac.com
Mon Jan 21 08:05:36 CET 2008
My bus is 1.33 GHz--I think the 2nd Gen Quad-cores are bumped to 1.5
GHz maybe. I'll see if there is anything I can do to bump the numbers.
My system is just like yours but with a second Quad core chip that
only a few apps can take advantage of.
It helps me with Modo, Lightwave, Shake, Photoshop, and Final Cut
Studio (which is why I got it in the first place) but not with Python
and most other apps as well.
Skip is right of course--the Global Interpreter Lock doesn't play here
since the benchmark wasn't written to take advantage of multi-core
machines--silly me I thought any good benchmark would be. In the end
my error doesn't matter since, if one made the benchmark multi-
threaded in the true sense of the term, the GIL would clamp down and
limit the benefit anyway. Which is what I was alluding to and should
have been more clear about.
My point was that, as I understand it, thanks to the GIL--Python
cannot easily take advantage of multi-cores period even when the
program uses multiple threads--it it is a limitation of the
implementation of the language interpreter. I guess that tells us we
ought to write multi-core code in C/C++/ObjC instead. Either that or
Python's implementation needs to embrace threading more expansively.
Still Python is a great language as it is. No one language fits all--
as much as we'd like it to for simplicity's sake.
On Jan 20, 2008, at 11:53 AM, Jack Jansen wrote:
> On 20-Jan-2008, at 19:23 , Daniel Lord wrote:
>> I ran the test on my 1st Gen Quad Core ( 2 x Quad-core 3.0 GHz, 13GB
>> RAM) and was a bit surprised to see little improvement over the Core
>> Duo numbers.
>> 63019.7 pystones/second
>> I am assuming the GIL is limiting threading and therefore I am
>> running on one or two cores--hence the tangible improvement is just
>> CPU speed: from 2.33 GHz to 3.0 GHz and a bit of the memory bandwidth
>> increase as well.
> My first generation quadcore at 2.6 Ghz clocks at 62578.2.
> So there's another limiting factor: from my machine to yours is a
> 15% speed bump, but only a 1% increase in pystone numbers.
> Somebody told me recently that MacOSX is not very good for fast task
> switching with multiprocessors, because apparently (his words, and
> possibly misrepresented by me) the implementation of semaphores
> sucks. This seems to corroborate that.
> Hmm, what is your bus speed? Mine is 1.33 Ghz, is yours that as well
> is it 1.5 Ghz? If the former it could be that semaphores somehow run
> at bus speed and semaphore overhead dwarves any processing done. If
> your bus runs at 1.5Ghz there must be yet another bottleneck...
> Jack Jansen, <Jack.Jansen at cwi.nl>, http://www.cwi.nl/~jack
> If I can't dance I don't want to be part of your revolution -- Emma
More information about the Pythonmac-SIG