FWIW: with wine from git from a couple days ago, here are wine build times on various Intel CPUs in seconds on Ubuntu 8.10's gcc-4.3.2 (or 8.04):
sec cpu wine notes 676 e7200 June http://kegel.com/new-computer-2008.html ubuntu 8.04 615 e7300 Nov 585 e7400 Nov (estimated, supposedly 5% faster than e7300) 540 e8400 Nov 360 q9400 Nov
In all cases, this was with the default optimization level, -O2. Compiling with -O0 was roughly twice as fast (e.g. 196 seconds to build on the q9400). The dual core e7300 and e8400 were tested with -j3; the quad core q9400 with -j5.
The current crop of CPUs is that they run cooler than older ones. Replacing the old Athlon 64's in the patchwatcher cluster with new CPUs and 80+ power supplies makes a big difference in the room's temperature!
The quad CPUs are coming down in price. Once you get used to fast builds, it's hard to go back, so don't try a q9400 or faster CPU unless you want to keep it :-)
Older benchmarks:
http://techgage.com/article/intel_core_2_quad_q6600/4 (is gentoo really that fast, or was he building with -O0?)
http://www.winehq.org/pipermail/wine-devel/2008-June/066121.html (q6600 386 seconds).
On Friday 07 November 2008 10:01:53 Dan Kegel wrote:
FWIW: with wine from git from a couple days ago, here are wine build times on various Intel CPUs in seconds on Ubuntu 8.10's gcc-4.3.2 (or 8.04):
sec cpu wine notes 676 e7200 June http://kegel.com/new-computer-2008.html ubuntu 8.04 615 e7300 Nov 585 e7400 Nov (estimated, supposedly 5% faster than e7300) 540 e8400 Nov 360 q9400 Nov
In all cases, this was with the default optimization level, -O2. Compiling with -O0 was roughly twice as fast (e.g. 196 seconds to build on the q9400). The dual core e7300 and e8400 were tested with -j3; the quad core q9400 with -j5.
Does your measurement include running configure? If I run make -j5 twice after 'make clean' on my q6600 2.4GHz system and take the second measurement I get 223 seconds.
This is 64bit Ubuntu 8.10 with gcc-4.3.2 and default optimization. Or perhaps disk IO is playing a role here? I'm using a stripe set of two disks.
-Hans
On Fri, Nov 7, 2008 at 1:35 AM, Hans Leidekker hans@codeweavers.com wrote:
sec cpu wine notes 676 e7200 June http://kegel.com/new-computer-2008.html ubuntu 8.04 615 e7300 Nov 585 e7400 Nov (estimated, supposedly 5% faster than e7300) 540 e8400 Nov 360 q9400 Nov
In all cases, this was with the default optimization level, -O2. Compiling with -O0 was roughly twice as fast dual core e7300 and e8400 used -j3; quad core -j5.
Does your measurement include running configure?
No, nor make depend. Nor is it cold boot, it's after 'make clean'. I didn't control whether there was a previous run, so the cache may be warm or cold.
If I run make -j5 twice after 'make clean' on my q6600 2.4GHz system and take the second measurement I get 223 seconds.
This is 64bit Ubuntu 8.10 with gcc-4.3.2 and default optimization. Or perhaps disk IO is playing a role here? I'm using a stripe set of two disks.
Yes, could be disk I/O. This system has a crappy old parallel IDE drive. (Could also be the video; this system has an internal graphics card.) - Dan
On Fri, Nov 7, 2008 at 11:04 AM, Dan Kegel dank@kegel.com wrote:
On Fri, Nov 7, 2008 at 1:35 AM, Hans Leidekker hans@codeweavers.com wrote:
sec cpu wine notes 676 e7200 June http://kegel.com/new-computer-2008.html ubuntu 8.04 615 e7300 Nov 585 e7400 Nov (estimated, supposedly 5% faster than e7300) 540 e8400 Nov 360 q9400 Nov
In all cases, this was with the default optimization level, -O2. Compiling with -O0 was roughly twice as fast dual core e7300 and e8400 used -j3; quad core -j5.
Does your measurement include running configure?
No, nor make depend. Nor is it cold boot, it's after 'make clean'. I didn't control whether there was a previous run, so the cache may be warm or cold.
If I run make -j5 twice after 'make clean' on my q6600 2.4GHz system and take the second measurement I get 223 seconds.
This is 64bit Ubuntu 8.10 with gcc-4.3.2 and default optimization. Or perhaps disk IO is playing a role here? I'm using a stripe set of two disks.
Yes, could be disk I/O. This system has a crappy old parallel IDE drive. (Could also be the video; this system has an internal graphics card.)
- Dan
Dan,
Could you post photos and complete specs of the PW cluster somewhere? I'm curious :)
-Zach
On Fri, Nov 7, 2008 at 6:27 PM, Zachary Goldberg zgold@bluesata.com wrote:
Could you post photos and complete specs of the PW cluster somewhere? I'm curious :)
Sure, once I bring the third slave online. In the meantime, see http://kegel.com/new-computer-2008b.html which should give you a partial idea.
On Sat, Nov 8, 2008 at 9:33 AM, Dan Kegel dank@kegel.com wrote:
Sure, once I bring the third slave online. In the meantime, see http://kegel.com/new-computer-2008b.html which should give you a partial idea.
I get the sense your basement is just covered in case-less computers. Awesome.
-
Dan Kegel -
Hans Leidekker -
Zachary Goldberg