Yes, there's been quite a bit of discussion about this here and in threads in some of the other WCG forums. BOINC is not for everyone and some machines do better with the UD agent. Those seem to be the more low-end machines. Yes, you can load and forget with BOINC which was the intent with the UD agent. However, BOINC is much more flexible so that you can maximize your crunching efficiency as conditions change (traveling, at home, not being able to connect to the net for a few days, etc. If you like to take a more active role to get the most crunching done on your machine, BOINC's the way to go. Anything has its plusses and minuses though.

Well objectively I was runnning 30 points per hour. Since BOINC 50 points per hour....prior running 1.38 results per day, now 1.7 results per day.....figure that on this 4 year old box, which on the day before BOINC installation lost half the ram due burn-out! Guess what's going to happen when that gets replaced in the next few hours

If CPU time on a given device translates directly to work as it seems to -since it does not necessarily reflect real-time (unlike UD's task time which does), then HT does allow more work to be completed by running two jobs concurrently (rather than simply dividing CPU time in half so that each job takes twice as long and thus only does the same total work).

I noticed this while experimenting with throttling and found that two jobs can each run in real-time or less (comparing to an independent clock). For example at 40%, each may accumulate 5s of CPU time in 5s while at 30% only 3s in 5s (total 6s in 5s so still better than running a single job).

It could be that more actual work is possible in the same CPU time as long as it is not less than real-time (that is, it is completed faster). Perhaps more testing over a longer period would be necessary to see how completion times are affected.

But it would be no surprise that HT is doing more work as I am already familiar with how much it benefits general multi-tasking and intensive video encoding/decoding &c.

Since this is new to me, perhaps someone can riddle me this: why is it that credits granted can exceed claimed (I could understand the reverse, especially in the case of errors)?

hmm , ok i tried very hard to STOp my ht pentium from doing 2 w/u 's, hmm maybe i should have left it at the "default" hmm i will change it back jusdt to check it out, did not have the option with ud of course so the question did not arise but now will give it a go and see wot happens.. hmm the dual core has sen 2 compputers only and seems very happy with that, i will toy with 2 jobs on one pentium sinlge core .

experiment continues,

puter 1 - 70% one w/u 6hrs

puter 2 - 34% and 35 % at 4hrs real "but 8hrs" combined with HT ie total of 70% worth of real work done in w/u equivalents.

projected is 8hrs for 100% on puter 1
projected is 6hrs for 100% (equivalent - ie 50 % times 2) on puter 2
(or 12 hrs for 2 completed w/u)

caveat - not the same w/u could be random allocation of easier work unit... will keep youse informed...

short version is that if w/u are comparable then i see about a 50% boost in actual throughput, of course dont count this as fact ( yet!) if i can do a "random" bunch in a row with the same parameters then it would be a "valid" experiment, or of course if i could get the exact same w/units etc would also be valid experiment.. cheers