~39,000-44,000 sec. (across about 7-8 long units) with all getting 177.20 cr's. Curious that I am getting a 5,000 sec. variation with the same credit granted (especially given that this machine has been doing mostly word processing only over the last week as I work on a manuscript revision). Should these workunits be varying this much?
Anyone expecting to get the same inflated credit they did with an optimized app when switching to a new standardized app is dreaming. Regardless of how much faster a new standard app is, it still has to give the same credit to maintain boinc's cross project equivilance.
Since akos's apps were 5-7x faster than the 4.37 app, the drop in credit/hour going from akos4.37 to s5 stock should be about 80-85%. IOW if you only saw a 50% drop, you're getting ~2-3x more credit than you should be. This could be because the non x86 apps scaled less with the improvements taken from akos's design, it could mean that 4.37 was claiming lower than it should've or that the current s5 app is claiming high. The latter is possible if the team is expecting farther refinements in the app, but doesn't want to do a midstream change in granted credit.
Since akos's apps were 5-7x faster than the 4.37 app, the drop in credit/hour going from akos4.37 to s5 stock should be about 80-85%. IOW if you only saw a 50% drop, you're getting ~2-3x more credit than you should be.
Yep. I was getting 40 credits per hour with S4 optimised apps, and I am down to about 20 with S5 stock app, which is still nearly twice what I get on the few other projects I have tried.
Quote:
This could be because the non x86 apps scaled less with the improvements taken from akos's design, it could mean that 4.37 was claiming lower than it should've or that the current s5 app is claiming high. The latter is possible if the team is expecting farther refinements in the app, but doesn't want to do a midstream change in granted credit.
Times for my 4 systems:
Athlon 1333mhz system: 109k sec per long WU (ave of 3) = 5.8 cs/hr (non-SSE)
AMD XP1700+: 11.3k sec per short WU (ave of 5) = 6.35 cs/hr (non-SSE)
AMD XP1600+: 52.9k sec per long WU (ave of 4) = 11.91 cs/hr (SSE enabled)
AMD XP2600+: 38k sec per long WU (ave of 5) = 16.96 cs/hr (SSE enabled)
AMD X2 3800+ @2.4GHz Long:
)
AMD X2 3800+ @2.4GHz
Long: 6.2-6.4h
AMD XP 2500+ 1.8GHz Barton
Long: 8.5h
Short: 0.9h
Both on Linux.
Pentium D
)
Pentium D 830
~39,000-44,000 sec. (across about 7-8 long units) with all getting 177.20 cr's. Curious that I am getting a 5,000 sec. variation with the same credit granted (especially given that this machine has been doing mostly word processing only over the last week as I work on a manuscript revision). Should these workunits be varying this much?
Intel Celeron D
)
Intel Celeron D 2.93Ghz
Short S5:
5218.34 sec (1hr,27min)
20.03 credit claimed
13.81 credit/hr
Long S5:
45700.92 sec (12hr,42min)
174.20 credit claimed
13.73 credit/hr
Mac PowerBook G4: 1.0 GHz
)
Mac PowerBook G4: 1.0 GHz (Ancient Lizzie)
Short Hanford
8,300 secs (2h 20m)
Credit claimed/Credit granted: 19.36/19.36
credit/hr: 8.42
MacBook Pro: Intel Core Duo 2.16 GHz
Long Hanford
20,234 secs (5h 40m)
Credit claimed/granted: 178/178
credit/hour:32.96
PowerMac G5: Dual 2.0 GHz
Long Hanford
32,500 secs (~9h 0m)
Credit claimed/ granted:178/178
credit/hr: 18.78
Short Hanford
3,900 secs (1h 05m)
Credit claimed/granted: 21.25
credit/hr: 19.68
Ok, I have now compared
)
Ok, I have now compared credits before with the
optimized 4,37 version compared to the new S5 version.
AMD 64 3000+ = You loose 54% credit per WU.
before 1CS per 90 sec.
now 1CS per 197 sec.
AMDXP 2600+ = You loose 51% credit.
Compared CS per sec:
before 1CS per 104 sec.
now 1CS per 212 sec.
RE: Ok, I have now compared
)
Disappointing.
Athlon
Stay tuned and keep crunching
Anyone expecting to get the
)
Anyone expecting to get the same inflated credit they did with an optimized app when switching to a new standardized app is dreaming. Regardless of how much faster a new standard app is, it still has to give the same credit to maintain boinc's cross project equivilance.
Since akos's apps were 5-7x faster than the 4.37 app, the drop in credit/hour going from akos4.37 to s5 stock should be about 80-85%. IOW if you only saw a 50% drop, you're getting ~2-3x more credit than you should be. This could be because the non x86 apps scaled less with the improvements taken from akos's design, it could mean that 4.37 was claiming lower than it should've or that the current s5 app is claiming high. The latter is possible if the team is expecting farther refinements in the app, but doesn't want to do a midstream change in granted credit.
RE: Since akos's apps were
)
Yep. I was getting 40 credits per hour with S4 optimised apps, and I am down to about 20 with S5 stock app, which is still nearly twice what I get on the few other projects I have tried.
I'd like to think that last thought is correct.
Dead men don't get the baby washed. HTH
Two Linux boxen: AMD XP-M
)
Two Linux boxen:
AMD XP-M @2.2GHz:
Long: 7:03
AMD Opteron 170 @2.6GHz:
Long: 5:36
Short: 0:38
Times for my 4
)
Times for my 4 systems:
Athlon 1333mhz system: 109k sec per long WU (ave of 3) = 5.8 cs/hr (non-SSE)
AMD XP1700+: 11.3k sec per short WU (ave of 5) = 6.35 cs/hr (non-SSE)
AMD XP1600+: 52.9k sec per long WU (ave of 4) = 11.91 cs/hr (SSE enabled)
AMD XP2600+: 38k sec per long WU (ave of 5) = 16.96 cs/hr (SSE enabled)
Seti Classic Final Total: 11446 WU.