| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: Bogus Bulldozer BOINC Benchmarks - blooking blor Badvice  |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 38
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
For what it's worth, mine is running 2743 and 9902 at stock speeds. Win 7 64bit with 16 gigs of ram.
|
|||
|
||||
|
David Autumns
Ace Cruncher UK Joined: Nov 16, 2004 Post Count: 11062 Status: Offline Project Badges: 
|
Cheers petehardy
----------------------------------------It's this bit that is key -- A priori job size estimates and bounds For each job, the project supplies an estimate of the FLOPs used by a job (wu.fpops_est) -- ... this explains everything It looks like when the WU's are sliced and diced the points allocation for that task is now fixed based on the expected amount of floating point operations it is going to take I was close with this guess. "My bet would be that BOINC has a lookup table for each processor from which it allocates points" The lookup table is used prior in allocating points to the task. It's probably what the Beta testing stage establishes It doesn't matter about your benchmark result Apologies I'm still old school You live and learn Dave  [Edit 1 times, last edit by David Autumns at May 7, 2012 2:59:39 PM] |
|||
|
||||
|
ryan222h
Senior Cruncher Joined: Sep 4, 2006 Post Count: 425 Status: Offline |
My actual results on CFSW:
----------------------------------------1100t @ 3.5 ghz = 2.75 h/result FX-8120 @ 3.3 ghz = 4.3 h/result 24 hour X 6 core (1100t) / 2.75 = 52.36 results/day 24 hour X 8 core (fx8120) / 4.3 = 44.65 results/day Each result on both machines yields approximately 100 boinc points. 1100t system power consumption = 170 watt fx8120 system power consumption = 150 watt Performance per watt on these two machines are nearly identical. I will eventually crank the clock speed on the fx-8120 and see what happens to efficiency  [Edit 1 times, last edit by ryan222h at May 8, 2012 3:58:34 PM] |
|||
|
||||
|
Hardnews
Senior Cruncher England Joined: Oct 11, 2008 Post Count: 151 Status: Offline Project Badges: 
|
Interesting: I'm on Ubuntu 11.04 64bit. The correlation between benchmarks and points still baffles me. Morover, on a dedicated crunch box, 4 gigs RAM seems to be more than enough, (I used to run it with 2 gigs). I've given BOINC 50% of that.
Here's mine. boinc 6.1.0.59 Ubu 11.04/64b. I7-2600K stock speed, 4gig Ram Wets: 3200 Dhrys: 12,829 |
|||
|
||||
|
kateiacy
Veteran Cruncher USA Joined: Jan 23, 2010 Post Count: 1027 Status: Offline Project Badges:
|
There's a thread from maybe a year ago about BOINC benchmarks under Windows vs Linux (so of course it only dealt with BOINC versions earlier than 7.xx.xx). The gist was that, for CPUs that hyperthread, BOINC calculates the benchmarks (I believe just for floating point) differently under Windows than under Linux. For Windows, it runs a single thread and gives those benchmark numbers as if the same was achieved by all threads running simultaneously with HT. For Linux, it actually runs all the threads simultaneously, and reports the average for each one. So unless the HT is perfectly efficient, the Linux benchmarks come out lower.
----------------------------------------This wouldn't affect the 905e of course, since it has 4 completely independent cores and doesn't hyperthread. But I wonder whether it affects Bulldozer benchmarks.  [Edit 1 times, last edit by kateiacy at May 9, 2012 10:07:19 PM] |
|||
|
||||
|
David Autumns
Ace Cruncher UK Joined: Nov 16, 2004 Post Count: 11062 Status: Offline Project Badges:
|
kateiacy
----------------------------------------from memory I think the boinc benchmark on Linux ran a hybrid multicore/ single core benchmark in 2 stages I'll have another go at a comparison with Linux this weekend on the Bulldozer I will also see if I can spot the difference between the WCG version of the BOINC Benchmark and the latest from Berkeley SiSoft Sandra also has a nice Whetstone/Dhrystone benchmark which we can use as a reproducible comparison  Just completed a big "Microsoft Tuesday" successfully - phew! - it's Wednesday of course  Dave |
|||
|
||||
|
David Autumns
Ace Cruncher UK Joined: Nov 16, 2004 Post Count: 11062 Status: Offline Project Badges:
|
for the record
----------------------------------------Using SiSoft Sandra 2012.06.18.45 as the benchmark 1055T 95W TDP running at 3Ghz non-turbo core the 6 Cores gives 51.58GFlops / 65.67 G int Ops FX-8150 at 4Ghz 125W TDP without turbo core the "8" Cores gives 70.84 GFlops / 120.35 G int Ops The PhenomII is giving 2.86GFlops/Ghz whereas the FX Bulldozer is giving 2.21GFlops/Ghz so the 1055T is per tick more efficient than the BD But for 125W TDP you get 32Ghz from a BD and for 95W TDP (the effiicient and green) 1055T you get 18Ghz you do the maths I get the feeling the BD is warmer but just as efficient on the Flops front On the integer maths way ahead |
|||
|
||||
|
ryan222h
Senior Cruncher Joined: Sep 4, 2006 Post Count: 425 Status: Offline |
To me, benchmarks are irrelevant, anyway. I look at actual computing performance on these projects, especially the ones that have very equal runtimes to each other. Judging by that,BD is more of a step sideways than forward from Phenom II x6. Its got 2 extra "cores", but is less powerful core for core for clock. I do own both, btw.
----------------------------------------A 32nm Phenom II x8 would be da bomb! [Edit 1 times, last edit by ryan222h at May 18, 2012 2:31:17 AM] |
|||
|
||||
|
|