As to my other machines, the results are far more favorable to S41.06.
700 MHz Pentium III
has been running short units from Z1_0232.5 using S40.12 in about 3732 seconds. First full S41.06 from same set took 3074 seconds--CPU ratio 0.82.
Early indications from my 933 MHZ Pentium III strongly indicate substantial improvement, possibly better than the other two hosts reported here.
1.4 GHz Pentium M (Banias)
has been running long units from Z1_1333.0 using S40.12 in about 5520 seconds. First full S41.06 from same set in 4500 seconds--CPU ratio 0.82
First completion from the 933 MHz Pentium III took .77 as much CPU time as the preceding WU's from the same major datafile.
As this particular machine had already reached a .2 ratio to distribution at the S-39L level, it is now somewhere better than .154--so is over 6.5 x the science productivity as before akosf.
Now if only the virulent disease which seems to have robbed large-cache hyperthreaded P4's of their speed starting with the change from S40.04 to S40.12 could be cured, I'd be even more blissful. I mention this point not as a criticism but as an opportunity--I think this category of machine makes a substantial contribution to the Einstein output, though not the the favored machine of many of frequent posters here.
IIRC to fix validation problems akos had to significantly enlarge a lookup table. This hurts the p4 badly since it has a very small l1 cache 8k data, 11k code. This is why S39L which squeezed the working set size down so much benefited p4's greatly, l2 is far better than system ram, but significantly slower than l1. The working set size of akos's apps is in the low teens, so the huge l2 cache is almost meaningless*, the addition of a few k of data won't hurt an amd machine with 32k l1 data cache. IIRC p3's also have larger l1 caches and are unaffected as well.
*only gain is that it reduces the odds of a context switch dumping everying back to system ram.
One is to set "cpu affinity" by project. It should do the trick if you write 1 into the seti scope and 2 into the einstein scope of the truxsoft_prefs.xml. To generate a project scope add:
...
I'm using Trux v.36. I tryed this, but when boinc starts it erases the following part of the truxsoft_prefs.xml file.
...
Do you know where I can download v.40? Anyway, thanks for the effort...
One is to set "cpu affinity" by project. It should do the trick if you write 1 into the seti scope and 2 into the einstein scope of the truxsoft_prefs.xml. To generate a project scope add:
...
I'm using Trux v.36. I tryed this, but when boinc starts it erases the following part of the truxsoft_prefs.xml file.
...
Do you know where I can download v.40? Anyway, thanks for the effort...
For a while I got back to S40.04
Are you looking for the way to run Einstein@Home & SETI@Home everytime the program start & will never run 2 same project (e.g. 2 SETI at once)??
Then you just need to add the line
Einstein@Home,SETI@Home
to truxost_prefs.xml if that also you just runing 2 project in your BOINC.
Thx C-SL!48. Actually my post was a little bit too quick. I thought my idea would work while reading through the trux homepage, but never tried it as I'm not running seti any more.
And archae86 is right: the project specific affinity needs release 5.3.12tx40 or higher, while the newest available beta for win is tx37.
Switched from U4101 to S4106. results are valid (8 valid, one pending).
Cpu is an Athlon 64 3200+ Socket 939 Venice Brand-ID: 4 90nm Stepping 2 Rev DH-E6. According to CPU-Z it supports MMx(+) 3dNow(+) SSE SSE2 SSE3 x86-64.
To get rid of the 32WU/d problem I run BoincStudio as boinc client. That's the reason why this computer is listed with 2 cpu.
Crunch time fell from 3600 to 2700 but the data file is different.
(compare S4009 and S4106)
One is to set "cpu affinity" by project. It should do the trick if you write 1 into the seti scope and 2 into the einstein scope of the truxsoft_prefs.xml. To generate a project scope add:
...
I'm using Trux v.36. I tryed this, but when boinc starts it erases the following part of the truxsoft_prefs.xml file.
...
Do you know where I can download v.40? Anyway, thanks for the effort...
For a while I got back to S40.04
Are you looking for the way to run Einstein@Home & SETI@Home everytime the program start & will never run 2 same project (e.g. 2 SETI at once)??
Then you just need to add the line
Einstein@Home,SETI@Home
to truxost_prefs.xml if that also you just runing 2 project in your BOINC.
Doesn't setting the project to only use 1 processor on a multiprocessor machine do the trick?
RE: As to my other
)
First completion from the 933 MHz Pentium III took .77 as much CPU time as the preceding WU's from the same major datafile.
As this particular machine had already reached a .2 ratio to distribution at the S-39L level, it is now somewhere better than .154--so is over 6.5 x the science productivity as before akosf.
Now if only the virulent disease which seems to have robbed large-cache hyperthreaded P4's of their speed starting with the change from S40.04 to S40.12 could be cured, I'd be even more blissful. I mention this point not as a criticism but as an opportunity--I think this category of machine makes a substantial contribution to the Einstein output, though not the the favored machine of many of frequent posters here.
IIRC to fix validation
)
IIRC to fix validation problems akos had to significantly enlarge a lookup table. This hurts the p4 badly since it has a very small l1 cache 8k data, 11k code. This is why S39L which squeezed the working set size down so much benefited p4's greatly, l2 is far better than system ram, but significantly slower than l1. The working set size of akos's apps is in the low teens, so the huge l2 cache is almost meaningless*, the addition of a few k of data won't hurt an amd machine with 32k l1 data cache. IIRC p3's also have larger l1 caches and are unaffected as well.
*only gain is that it reduces the odds of a context switch dumping everying back to system ram.
RE: One is to set "cpu
)
I'm using Trux v.36. I tryed this, but when boinc starts it erases the following part of the truxsoft_prefs.xml file.
...
Do you know where I can download v.40? Anyway, thanks for the effort...
For a while I got back to S40.04
Athlon XP 2000+ D41.12 - 4300
)
Athlon XP 2000+
D41.12 - 4300 sec on average (2 valid, 2 pending)
S41.06 - 3700 sec on average (still pending)
akosf, this is incredible!
RE: RE: One is to set
)
Are you looking for the way to run Einstein@Home & SETI@Home everytime the program start & will never run 2 same project (e.g. 2 SETI at once)??
Then you just need to add the line
Einstein@Home,SETI@Home
to truxost_prefs.xml if that also you just runing 2 project in your BOINC.
RE: RE: Are you looking
)
Very clever. Thanks!
Thx C-SL!48. Actually my post
)
Thx C-SL!48. Actually my post was a little bit too quick. I thought my idea would work while reading through the trux homepage, but never tried it as I'm not running seti any more.
And archae86 is right: the project specific affinity needs release 5.3.12tx40 or higher, while the newest available beta for win is tx37.
MrS
Scanning for our furry friends since Jan 2002
S4106
)
S4106 client
http://einsteinathome.org/task/27565814
Time: 3,749.72 (sec)
"Old" client
http://einsteinathome.org/task/27565808
Time: 18,352.72
http://einsteinathome.org/host/580438
Intel(R) Celeron(R) CPU 2.80GHz
1Gb memory
BOINC manager: 5.3.31
Switched from U4101 to S4106.
)
Switched from U4101 to S4106. results are valid (8 valid, one pending).
Cpu is an Athlon 64 3200+ Socket 939 Venice Brand-ID: 4 90nm Stepping 2 Rev DH-E6. According to CPU-Z it supports MMx(+) 3dNow(+) SSE SSE2 SSE3 x86-64.
To get rid of the 32WU/d problem I run BoincStudio as boinc client. That's the reason why this computer is listed with 2 cpu.
Crunch time fell from 3600 to 2700 but the data file is different.
(compare S4009 and S4106)
RE: RE: RE: One is to
)
Doesn't setting the project to only use 1 processor on a multiprocessor machine do the trick?