Cheers Batchoy

5 out of 6 of mine are Athlon XP's (A couple are early Sempron's but they are just XP's in disguise) most have 512Mb Ram and my best UD overall rating on any box is 158 (126 - 158 is the range) so I wasn't against the end stops with UD either

I'm now 442nd in the world up 37 places in the last 3 days (and that's a huge gain in 450th Land)

Just look at this with no hardware change at all!!! That last points block still has another 6 hours worth to be added at midnight



The toothless gap was a particularly bad day for me

Folks your thoughts and observations please

Dave

There's nothing like a bit of backwards compatability and this is nothing like a bit of backwards compatability

I must admit I'm getting the same effect Dave with the results but we have been assured that the FAAH.exe bit of the program is the same in both UD and BOINC and that the different agents are just a wrapper for that code so there shouldn't be any change in the output.

However the FAAH exe's don't have the same filenames in UD and BOINC agents so they may be different. We always knew that the UD exe's were not optimised for the processor they were running on and so they didn't have some of the potentially useful number crunching extensions such as all those MMX, SIMD, SSE and 3dNow that Intel and AMD have added to the X86 codebase since 1996. Maybe in BOINC these features have been enabled giving rise to the improved efficiency which would be great news and we have really earned those points

As yet we've had no feedback from above in this thread to say differently.

The FAAH code in BOINC returns 2 files back to base after each work unit maybe these are both being counted as results and having 2 lots of run time and points being allocated as a consequence, a little bug'et.


I think I have 3 maybe's in this post already.

It would be good to have some input from those who are really in the know from IBM.

Take care Dave B

Dave A

I attempted to point out what Dave Bell has illustrated above in the quote below


Ady reported the same findings in his quote below

Your reply in the following post in the MOT thread was "NOT"

I am very pleased to report that I getting an average of 70% increase in results returned, since installing BOINC

There is another possibility

The work unit slicing and dicing

The work units are prepared by a "heuristic" to be around the 10 hour mark on your average PC by IBM for the UD agent

The BOINC agent claims to be able to download an amount of days worth of Work Units in advance (how it does this when the WU's are non-deterministic is beyond me - those estimated run times are never correct sometimes the end time goes down faster than time other times you gain 2 seconds to the end for each second of crunching - this we have always known)

Maybe (that's my 4th) the hopper of work to be done by the BOINC client is chopped up in a different fashion - although even as I write this my experience (so far) with BOINC doesn't suggest they are any shorter

I would put my money on double accounting for the FAAH work units on account that 2 files are uploaded to IBM from BOINC instead of the gzip'ed up single file from the UD agent

I'm at about 133%, at a guess. Nifty.