Preliminary results, looks like at least 20% speedup over 4.20, and possibly more than 25%.
"Fast." ... you ain't kiddin'!
Speaking of fast. . .
It appears that you're overclocking the living daylights out of this Core 2 Quad. For curiosity, what do you have it overclocked to? And, are you doing anything special for cooling?
I think that one was still crunched exclusively by 4.20, see the debug output:
634, 635, 2007-12-17 15:08:32.0246 [normal]: Built at: Nov 29 2007 15:00:43
2007-12-17 15:08:32.0248 [normal]: Start of BOINC application 'einstein_S5R3_4.20_i686-pc-linux-gnu'.
2007-12-17 15:08:32.0249 [normal]: Found 'EAH_DEBUG_DDD' file, trying debugging with 'ddd'
CU
H-B
Okay then. I don't usually do my app changes when I still have half-crunched WUs (in fact I only did it this time because a friend of mine tried and it worked okay for him) so I wasn't sure.
Thanks a lot for looking into this. Maybe it's just this laptop acting up again... BOINC and Ubuntu never quite seemed to like each other on this box. But it obviously isn't a power user app problem. I can't wait to see the speedup for that, even if this is a Core 2 I should see quite a difference I think...
EDIT: Did some research; obviously "signal 6" means an "illegal character" in a WU. So maybe I was just unlucky, or my less-than-reliable internet connection messed the WU up during download.
It appears that you're overclocking the living daylights out of this Core 2 Quad. For curiosity, what do you have it overclocked to? And, are you doing anything special for cooling?
While I'd like to take credit for the performance of the machine, the accolades belong entirely to the engineers at Intel. The motherboard is a Supermicro C2SBA+II, the BIOS for which has limited overclocking ability. You can select the stock clock speed +5%, +10%, or +15%. This machine is running +5%. It wasn't stable at +10%, but there are voltage adjustment knobs available and I know other people out there are running their Q6600's much faster. I think a lot of the highest achievers in that arena are using ASUS motherboards.
Nothing out of the ordinary for cooling. It's air-cooled with a Zalman CNPS9500 AT. CPU temperature hovers between 32-35C. Temperature in the room where it's located is generally between 20-22C.
Check this out, it directly quantifies the speedup of 4.21 over 4.20.
Processed from start to finish on my machine using 4.21: work unit 35942797. I finished it in 22,978 CPU sec., wingman finished it (with 4.20) in 25,530 CPU sec.
A recent result with the same wingman, where both machines used 4.20: work unit 35931305. I finished it in 30,480 CPU sec., wingman finished it in 25,722 CPU sec.
Based on that, the expected time for the first workunit (35942797) under 4.20 on my machine would have been 30,252 CPU sec, but with 4.21 instead it was 22,978 CPU sec., which means that 4.21 is over 24% faster than 4.20.
Once again, nice work Bernd!
(BTW, I'd like to have this wingman all the time! :-)
Check this out, it directly quantifies the speedup of 4.21 over 4.20 .....
For your architecture. :).
In view of Bikeman's comments, I've installed the new 4.21 app on four SSE2 capable machines, leaving aside for the moment a large number of SSE only machines similar to Bikeman's dual PIII 1400. The machines I've installed on are 2 X Amd64, 1 X P4 2.4 dual Xeon HT and 1 X P4 1.6.
So far most of these machines have completed work started with earlier versions without incident. They are now on work that was commenced with the new version and so I'll get a more accurate picture of the speedup (if any) soon.
My preliminary assessment is that all four machines listed are not going to show any significant speed increase at all. The work that was completed with the new version took pretty much what was being estimated with the old and the new work now partially complete also seems to have pretty much an unchanged estimated crunch time as well. Obviously, I'll be trying to assess the natural variation between different tasks in the sequence but so far it doesn't look very hopeful for much of a speedup on the architectures I'm using.
Bikeman, thanks for pointing me to the sse/sse2 issue.
I'm not sure how much speedup is actually due to the SSE2 code. I'll fix the description on the Power Apps page, but won't build a new SSE-only App before I fixed some other problems in the code.
BOINC and Ubuntu never quite seemed to like each other on this box. But it obviously isn't a power user app problem. I can't wait to see the speedup for that, even if this is a Core 2 I should see quite a difference I think...
Strangely, BOINC 5.10.28 with standard GUI is only available for Ubuntu and Debian. I am using SuSE 10.1 and shall stick to 5.10.21. But why?
Tullio
I already asked myself the same and actually took the more conservative approach, meaning I'm also using 5.10.21 atm. No idea why there would be a special app for Ubuntu/Debian, except it only referred to it being offered as a .deb package, but then there would be another download in .tar.gz or similar format for other distros. So there must be sth special about the app itself. But don't ask me what. Ubuntu and Debian normally tend to have very different kernel versions (due to the different development cycle) so that can't have to do with it...
BOINC and Ubuntu never quite seemed to like each other on this box. But it obviously isn't a power user app problem. I can't wait to see the speedup for that, even if this is a Core 2 I should see quite a difference I think...
Strangely, BOINC 5.10.28 with standard GUI is only available for Ubuntu and Debian. I am using SuSE 10.1 and shall stick to 5.10.21. But why?
Tullio
The 5.10.xx builds should be working with SUSE too.
My 6000+ machine just finished its first 4.21 WU, at just over 31,000 seconds.
It's hard to say yet if the SSE code is having an effect, since times for workunits on this machine range from just under 30,000 seconds to a bit over 40,000 seconds. I should know after a few more results come in, so that I can look at the new range.
RE: Preliminary results,
)
Speaking of fast. . .
It appears that you're overclocking the living daylights out of this Core 2 Quad. For curiosity, what do you have it overclocked to? And, are you doing anything special for cooling?
We aspiring speed-demons would like to know.
RE: Is it this one?
)
Okay then. I don't usually do my app changes when I still have half-crunched WUs (in fact I only did it this time because a friend of mine tried and it worked okay for him) so I wasn't sure.
Thanks a lot for looking into this. Maybe it's just this laptop acting up again... BOINC and Ubuntu never quite seemed to like each other on this box. But it obviously isn't a power user app problem. I can't wait to see the speedup for that, even if this is a Core 2 I should see quite a difference I think...
EDIT: Did some research; obviously "signal 6" means an "illegal character" in a WU. So maybe I was just unlucky, or my less-than-reliable internet connection messed the WU up during download.
RE: It appears that you're
)
While I'd like to take credit for the performance of the machine, the accolades belong entirely to the engineers at Intel. The motherboard is a Supermicro C2SBA+II, the BIOS for which has limited overclocking ability. You can select the stock clock speed +5%, +10%, or +15%. This machine is running +5%. It wasn't stable at +10%, but there are voltage adjustment knobs available and I know other people out there are running their Q6600's much faster. I think a lot of the highest achievers in that arena are using ASUS motherboards.
Nothing out of the ordinary for cooling. It's air-cooled with a Zalman CNPS9500 AT. CPU temperature hovers between 32-35C. Temperature in the room where it's located is generally between 20-22C.
So, yeah, the Core 2 design really is that good.
Check this out, it directly
)
Check this out, it directly quantifies the speedup of 4.21 over 4.20.
Processed from start to finish on my machine using 4.21: work unit 35942797. I finished it in 22,978 CPU sec., wingman finished it (with 4.20) in 25,530 CPU sec.
A recent result with the same wingman, where both machines used 4.20: work unit 35931305. I finished it in 30,480 CPU sec., wingman finished it in 25,722 CPU sec.
Based on that, the expected time for the first workunit (35942797) under 4.20 on my machine would have been 30,252 CPU sec, but with 4.21 instead it was 22,978 CPU sec., which means that 4.21 is over 24% faster than 4.20.
Once again, nice work Bernd!
(BTW, I'd like to have this wingman all the time! :-)
RE: Check this out, it
)
For your architecture. :).
In view of Bikeman's comments, I've installed the new 4.21 app on four SSE2 capable machines, leaving aside for the moment a large number of SSE only machines similar to Bikeman's dual PIII 1400. The machines I've installed on are 2 X Amd64, 1 X P4 2.4 dual Xeon HT and 1 X P4 1.6.
So far most of these machines have completed work started with earlier versions without incident. They are now on work that was commenced with the new version and so I'll get a more accurate picture of the speedup (if any) soon.
My preliminary assessment is that all four machines listed are not going to show any significant speed increase at all. The work that was completed with the new version took pretty much what was being estimated with the old and the new work now partially complete also seems to have pretty much an unchanged estimated crunch time as well. Obviously, I'll be trying to assess the natural variation between different tasks in the sequence but so far it doesn't look very hopeful for much of a speedup on the architectures I'm using.
If anything changes, I'll report again.
Cheers,
Gary.
Bikeman, thanks for pointing
)
Bikeman, thanks for pointing me to the sse/sse2 issue.
I'm not sure how much speedup is actually due to the SSE2 code. I'll fix the description on the Power Apps page, but won't build a new SSE-only App before I fixed some other problems in the code.
BM
BM
RE: BOINC and Ubuntu never
)
Strangely, BOINC 5.10.28 with standard GUI is only available for Ubuntu and Debian. I am using SuSE 10.1 and shall stick to 5.10.21. But why?
Tullio
I already asked myself the
)
I already asked myself the same and actually took the more conservative approach, meaning I'm also using 5.10.21 atm. No idea why there would be a special app for Ubuntu/Debian, except it only referred to it being offered as a .deb package, but then there would be another download in .tar.gz or similar format for other distros. So there must be sth special about the app itself. But don't ask me what. Ubuntu and Debian normally tend to have very different kernel versions (due to the different development cycle) so that can't have to do with it...
RE: RE: BOINC and Ubuntu
)
The 5.10.xx builds should be working with SUSE too.
My 6000+ machine just
)
My 6000+ machine just finished its first 4.21 WU, at just over 31,000 seconds.
It's hard to say yet if the SSE code is having an effect, since times for workunits on this machine range from just under 30,000 seconds to a bit over 40,000 seconds. I should know after a few more results come in, so that I can look at the new range.