1 gig ram on pent 4 ht 3ghz,, btw the currnet job of "2 in 1" is 70% and 70% at 9 hrs each simultaenoulsy, so i am not convinced.. i would have to compare a few times ( ok i admit i got a bit too excited so have both pents doing 2 each , the other puter "2 in 1" is 3hrs 20% x 2 w/u .. so the w/u are different in difficulty. i would need to do a few more "head to head" comparisions and or "grab" equal ie same number work units. 512 mb ram pushing it i am guessing, 1 gb works fine for me but agaiin i have to stress the 1st apparent 1.5 increase in efficiency may have been due to chance.. would lie to get a bunch of others to have input

Okay, thanks. But you have to keep in mind that every job is different, so maybe you'll get more points for those jobs in the end.
I guess that IF you receive more points per hour for the HT device, that automatically means it runs more effective, or am I confusing something?

Keep us updated!

Hi Dagorath,
The WCG staff decided that, on average for our members, a BOINC point equals 7 UD points. But the UD scoring method was capped years ago for some reason or other. So a modern high-end CPU can earn more (uncapped) BOINC points than would be awarded by the UD client. But there must be a lot of older low-end CPUs that score better using UD than they would using BOINC. The 7 to 1 ratio was chosen because just about as many computers scored better using UD as the number of computers that scored higher using BOINC. Neither client actually speeds up the application. But BOINC does allow multiple threads, which might produce a real improvement and might also introduce a difference in points awarded. (Though BOINC tries to award points based on how long a thread runs. If done correctly, hyperthreading should not change points awarded per hour. Unless the cache thrashes. . . . I do not think that would change the BOINC points awarded per hour, though.)

Over on grid.org, snoekie complained that even the single-threaded UD client slowed down his P4 computer noticeably while running Rosetta because he ran a voice-input thread in the background. Running 2 monster floating-point routines in the background really slowed down his foreground tasks. Keeping this in mind, I never recommend running 2 distributed computer applications in the background on a hyperthreading computer that is also being used normally. But that does not mean that it cannot work well. I am just trying to avoid mysterious problems that might hurt our members. Sort of like keeping beginning swimmers in the shallow end of the swimming pool.

The rule of thumb for hyperthreading is that it speeds up some carefully programmed video-editing processes by about 45%, but for most programs it speeds up total throughput by about 14% - 24%. But the speedup comes from using different parts of the CPU simultaneously. Sounds great when we are running 1 floating point routine plus standard integer programs. Running a scientific program in the background is exactly what the doctor (or computer architect) had in mind when designing hyperthreading. But running 2 monster floating point programs through the same cache and the same FPU. . . . . . I doubt that anything like a 14% speedup is possible.

Just my opinion,
Lawrence

Hi Lawrence,

Hmmm. OK. Good food for thought. I did not consider threading and how threading might be used differently or influenced differently by the 2 agents (BOINC and UD). Indeed, there are lots of variables and I don't think I know enough about modern processor architecture to add much to what you have said. But I have an idea...

I have here 2 almost identical machines..... 2 HP Pavilion a1112n. I propose setting up a shootout between BOINC and UD, BOINC on 1 machine vs. UD on the other, Bad Boy BOINC Toe2Toe With The UD Kid.

I have a third machine that I would do my own work on. The 2 shootout machines would be dedicated strictly to crunching WUs. The only difference(s) between the 2 are that 1 runs on Win XP Home (32 bit) while the other is on Win XP 2003 (64 bit). However, if I ran BOINC on the 32 bit OS and UD on the 64 bit for a week, noted the results, then switched BOINC to the 64 bit OS and UD to the 32 bit... well... that should cancel OS influence as well as other variables. For example, 1 has ATA-100 HD while the other has ATA-133 HD.

Same amount and brand of RAM in both, BIOS is same revision level and configuration in both.


Any thoughts? Anybody? Any way I could make this shootout more accurate? interesting? easier to interpret?


--

Lawrence,

OK... BOINC and UD have been shooting it out for about 2 days now. Both are crunching only FAAH WUs.

BOINC has returned 8 results in total run time of 41 hrs. UD has returned 10 results in total run time of 27 hrs. (The cooling fan on the UD shooter started making a scary sound so I had to shut it down until I could replace the fan, that's why the difference in total run time).

On Average Run Time Per Result, the 2 are neck to neck, 5.08 hrs for BOINC, 5:13 hrs for UD. BOINC had a 30 minute advantage 24 hrs ago but the averaging effect over time has reduced that to a negligible lead. I predict BOINC's lead will approach 0 in the long run.

On Average Points Per Result, BOINC has a significant lead, 479 to 193. That difference has not changed much in the past 24 hrs.

One difference between the 2 shooters is that UD is running on 64 bit Windows XP whereas BOINC is running on 32 bit XP. UD is therefore running under WoW (Windows on Windows) which is a virtualization of the 32 bit environment on 64 bit platforms. I don't know for sure how that might skew the results but I should think it would slow UD down rather than speed it up. Yet, UD keeps up with BOINC. Of course we can only say that applies to my hardware and configuration.

I created 2 new WCG user accounts in addition to this account, 1 for each of the puters described in my earlier post up thread. If you are interested in verifying the results I have reported here then please email me at

$kim$bergman$at$shaw dot ca (remove the $ chars)

and I will give you passwords to the 2 shooter accounts and you can observe the results. If you know someone with account privileges connected to the project who might also be interested then the 2 shooter acounts are:

BOINC-XP-32
UD-XP-64

I plan to let the shooters have at it for another 48 hours or so then flip-flop. UD will run under 32 bit XP, BOINC under 64 bit XP and WOW, just to see if there is any advantage to either platform.

BTW, both machines are doing almost nothing other than crunching WUs. Both have virus protection running in background but weekly scans have been turned off for the duration of the shootout.

--