Newbie Questions - Dual Core, and WU Distribution
29 Dec 2005 10:58:17 UTC
Topic 190468
(moderation:
I have a Pentium D 3.20 in a Dell 9100 under XP Pro, 2 gigs of RAM. I’ve run einstein@home 24/7 for nine days and have processed 60 WUs. TaskInfo tells me that WUs are getting exclusive use of the two cores, with the exception of the odd blip from Explorer. Each WU takes about 7hrs25mins and consumes about 26.7K of CPU per second. Does that level of performance sound about right?
Some co-crunchers with Athlon 64 X2 Dual Core 4200+ consume about 16.5K of CPU per second vs. my 26.7K. It that chip really that much faster than mine??
I have 16 WUs pending credit. The four most recent, and the two now crunching, have only been sent to me. Does this mean anything other than that the speed of WU creation exceeds the total crunch power available?
Thanks for any insights…
Newbie Questions - Dual Core, and WU Distribution
)
None of this seems really unusual. The speed difference is partially due to E@H liking AMD's shorter pipeline. It may also be partly due to services that windows and dell have installed on your computer at the factory.
E@H uses locality scheduling. This means it trys to send work to hosts that already have the large data file. A side effect of this is that is sometimes delays the issuing of results since no hosts with that data file have checked in recently.
BOINC WIKI
BOINCing since 2002/12/8
I've noticed that because of
)
I've noticed that because of the optimized SETI app, my Intel boxes tend to make outrageous credit claims while taking the longest time to run an Einstein WU. Conversely, AMD boxes running the same software configuration tend to claim the median amount of credit and crunch the WU in the shortest amount of time (relative to processors designed for x86/Win).
Stewie: So, is there any tread left on the tires? Or at this point would it be like throwing a hot dog down a hallway?
Fox Sunday (US) at 9PM ET/PT
Interesting, your system does
)
Interesting, your system does almost consistently the same time per work unit. Tell me, are you running only Einstein@Home?
Anyway, 26K seconds per work unit is close to what i get on my 3.2 GHz computers.
As far as the comparison with the AMD chips, also note that the AMD chip is rated with an effective increase in performance of 1 GHz over your chip, nearly 25% better throughput. Other factors that come into play that you did not note are the cache size of the two chips ...
RE: Interesting, your
)
I fired up a web site with specifications and statistics for your perusal. :o)
All active units are running both applications. I had been running a 5/1 (Einstein/SETI) resource, but shifted down to 3/2 today after the latest optimized SETI client shaved 28% off the average WU time.
Fascinating, since my Celeron is running around 23.5K and 1.8 Duron has been in the upper teens... :o/
Stewie: So, is there any tread left on the tires? Or at this point would it be like throwing a hot dog down a hallway?
Fox Sunday (US) at 9PM ET/PT
RE: Interesting, your
)
It's the only BOINC-driven application I run, but it is my only PC and I'm at it 3-4 hours, though I don't run other CPU-intensive work.
RE: RE: Anyway, 26K
)
As John said while AMDs chips don't clock as fast as P4s the accomplish significantly more per cycle. Broadly speaking AMDs per cycle performance is equivlant to a p3 (both are shallow pipeline designs), Pentium-M's are derived from the p3 archtecture and have AMDish performance rather than P4ish.
Without knowing what your celeron is, I can't comment on it's relative performance (intels used that brand for both p3 and p4 style chips).
I am wondering about your Duron 1.8. My old Athlon 1.4 ran ~36-40k (10-11h) per WU, and my dualcore gig is taking ~20k/core. I'm not sure why your 1.8's running faster than my 2.4's, although I've got a few thoughts. This is my main machine, but I don't do any heavy gaming and am typically >95% e@h, so while this might be a small part of it if you're duron's strictly a cruncher it can't be the whole story. My second thought is that it's more to do with the difference between dual core and true dual proccesor designs. Namely that while I have 2 cpus they're sharing the same memory bus so depending on how many cache misses I have the ram bottleneck might be what's holding me back.
As assides, the difference between P4/athlon archtectures performance/clock and absolute clock speeds, is just as true for the athlon/PowerPC comparisons although IIRC the reason's are somewhat different. Both Athlons and PPCs use shallow pipelines, but the PPC has a much larger vector (massively parallel) proccessing capability (called Altivec for hte PPC and mmx/sse for x86s).