| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Community
Forum: Chat Room
Thread: [FINAL RESULTS HT Test] The Solar System Initiative - SSI |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 81
|
|
| Author |
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3715 Status: Offline Project Badges: 
|
Hypernova,
----------------------------------------The benefit on returned WUs seems disappointing, indeed. I suspect that you have counted only valid WUs, and if pending WUs are not proportional in both machines this is affecting the measurement. You should count all returned WUs during the same period in both machines. You can do that retroactively from the BOINC file "job_log_www.worldcommunitygrid.org.txt" which you will find in the BOINC Data directory. The first field of each line is the timestamp on completion of a job, in seconds since 01/01/1970 00:00:00. To save your time, the timestamp for 22/03/2010 00:00:00 is 1269216000, and the timestamp for 29/03/2010 00:00:00 is 1269820800, for example. The name of the job is between the nm and the et tags as you would have quickly found yourself, I am sure. Of course it is recommended to work on a copy of that file in order to not disturb BOINC.  Good luck. Jean. Edit: I realize that my timestamp examples are one week too early, but you will certainly adapt them yourself without too much work. ---------------------------------------- [Edit 1 times, last edit by JmBoullier at Apr 7, 2010 1:22:45 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I would like to suggest that you make sure there are no crossover tasks between the old version of HCC and the new version included in your results.
----------------------------------------Pure speculation... I think another reason that the numbers are closer than we all expected may be project specific, meaning that perhaps intermediate calcs for HCC fit in the L2/ L3 cache if HT is off, helping to improve performance but once you move to 8 threads the shuffle in/ out of cache to main memory slows things down. Even a 13% improvement is still quite nice and likely not easily achievable by increasing the OC by 13% at the same voltage. Certainly with HT off, an i7 cpu runs much cooler and draw much less power so you can afford to crank up the volts a bit to get a better OC. Taking that path I have found in the end that I am typically limited by the overall OC of the CPU before I can reach a break even point with HT off compared to HT on. At the upper edge of stability I can only get about 100-150 MHz more out of my CPUs (i7-920 C0 and i7-920 D0) with HT off. HT ON FTW  [Edit 4 times, last edit by Former Member at Apr 7, 2010 1:43:04 PM] |
||
|
|
Jack007
Master Cruncher CANADA Joined: Feb 25, 2005 Post Count: 1604 Status: Offline Project Badges: 
|
Ok,
----------------------------------------So I can't remember where I mentioned this, but I had found some of the moons of Saturn in Wiki. There was a nice collection of them in one combined jpg. I took that 4 meg file downloaded it to a thumb drive and took it to a printer and had them print up a 12" x 36" Poster (for lack of a better term) on a foam backing. It will now reside above my computer desks and i can look up and dream about Dione (the most beautiful one in my opinion). I miss my astronomy, since I live more in the country now, perhaps I'll break out the spotter scope and start checking out the night sky as it gets warmer. How does this fit in with this thread? Well Hypernova has the planets locked down, so I'ma gonna go after the moons of Saturn (there are like 70 of them so plenty to go around) I'm taking Dione (densest moon covered with ice) Rhea largest airless moon, and Hyperion because it's shaped like a potato. Next comp, perhaps a I7980x, or I've seen a quad socket motherboard (for $679 just for the moboard), will be TITAN.  |
||
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges:
|
Jack 007, I love the idea. Your post fits perfectly in this thread. Saturn satellites are yours go ahead and keep us posted.
----------------------------------------Jupiter satellites are also free for grab if there is someone interested in the concept. For those who are interested there is however a small limitation: "ALL THESE WORLDS ARE YOURS EXCEPT EUROPA. ATTEMPT NO CRUNCHING THERE"  And there is another satellite which is not free and that is the MOON. Otherwise everybodyis welcome to follow the concept. Our objective is to build the WCG galaxy of crunchers.   |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Here are some numbers corresponding to a full 7 days period. This are taken from the daily device statistics. Just to remember Eris is without HT, and Pluto with HT. Points: Eris 171'523 Pluto 247'767 ( +44% ) Runtime in seconds: Eris 2'628'723 Pluto 5'539'205 ( +110% ) Returned Valid Results Eris 637 Pluto 722 ( +13% ) According to what is the most important criterion for you you are very happy or just happy. For Points and Runtime HT remains a must. For Returned Results HT is definitively better but not a killer. Jean you were right on that point. For crunchers HT is a boon, as Badges are based on Runtime and Ranks on Points. For the Scientists valid WU's are the most important criteria, and even for them HT remains a plus. The test will continue with a new and last update for a 10 days period. After that Eris will be back wit HT. This update will be on Sunday 11th April. It's quite probable the 6.08 optimizations took out allot of air that HT was previously filling, but at any rate at least for HCC there's a positive variance on the returned work unit count with the most steady run times of all WCG projects. Can't think of another WCG science that is suitable to do this... maybe hfcc, though there is large variability in run times over a stretch of a week the 2 rigs would pick up the longer and shorter in fairly equal ratio. This one is interesting as it got fair bit of Page Faulting, but a much lower Integer ingredient. edit: got 1 HFCC job at 39%, 5:28 hours CPU and 2.8 million PF, Delta nul. The memory needs are also 4 / 5 times that of HCC.
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Apr 11, 2010 9:57:44 PM] |
||
|
|
Jack007
Master Cruncher CANADA Joined: Feb 25, 2005 Post Count: 1604 Status: Offline Project Badges:
|
HO HO,
----------------------------------------Any news yet on your test Hypernova? |
||
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges:
|
Here are the final values for Returned Results over 13 days including Pending Validations.
----------------------------------------For Points and Runtime refer to my previous post above. ERIS (HT OFF) 1093 PLUTO (HT ON) 1249 (+14,2%) Adding PV has improved a little the picture in favor of HT. The test is now finished and HT is ON again. Conclusion: use allways HT ON for WCG crunching.  P.S. I may do a test in the future with two identical i7 980X rigs and see if we get similar results. |
||
|
|
Jack007
Master Cruncher CANADA Joined: Feb 25, 2005 Post Count: 1604 Status: Offline Project Badges:
|
Excellente!
----------------------------------------That bit of info should be in the back of everyone's mind, as they pick processors. I've been considering a workstation, and for the extra bit of dough to go intel over AMD, I guess a 14% price difference will be the critical factor. (not that the extra points/runtime wouldn't hurt) |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3715 Status: Offline Project Badges:
|
PLUTO (HT ON) 1249 (+14,2%) Thank you for this complement of information, Hypernova.I was still disappointed seeing only +14.2 % more work done with HT ON, but it is consistent with Movieman's tests which give durations per WU much higher than what I expected** at the clock speeds he is achieving ( 95 minutes running at 4152MHz ). So I thought a little more and I came to this conclusion: 1.. HT is meant to make use of the cycles that the processor is not able to use without HT 2. the benefit of the HT feature is thus lower when the processor is already more efficiently used, which is most probably the case in your machines with better Lx caches in their processors and top quality speedy RAM to feed them. Less losses = less to recover. ** The HCC average duration in my Q6600 at 3.16 GHz is 0.85 hour (51 minutes) (measured on 1526 results under Ubuntu 64). |
||
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges:
|
I decided to check the influence of DDR3 RAM speed and characteristics to see if there is a difference.
----------------------------------------I used two identical rigs with core i7 980X CPU with HT on and running at 4 Ghz. CERES has DDR3 RAM running at 1600 Mhz with settings at 8-8-8-24. MERCURY has DDR3 RAM running at 1860 Mhz with settings at 7-7-7-20. The RAM of Mercury is frequency wise 16% faster. Timings are also more agressive. They crunched both either the same mix or exclusively HCC. Over a twenty eight day average MERCURY produced 2.6% more points and 8.5% more results than CERES. The impact is much less than expected. [Edit 2 times, last edit by Hypernova at Apr 18, 2010 7:20:17 PM] |
||
|
|
|