[TiloProbst]

some PC buddy just informed me that the much talked about Calibrating BOINC Core Clients by Trux recently had their mac release.

take a look here: http://boinc.truxoft.com/core-cal.htm
-> Downloads

there have been quite some noise about Trux clients (at planet3dnow forums for example), since his attempt to make sure that every WU achieves the Credit it deserves, looks like cheating at first glance.

I have not tested them yet but maybe some of you want to do so

[OneMacGuy]

Originally Posted by TiloProbst

some PC buddy just informed me that the much talked about Calibrating BOINC Core Clients by Trux recently had their mac release.

take a look here: http://boinc.truxoft.com/core-cal.htm
-> Downloads

there have been quite some noise about Trux clients (at planet3dnow forums for example), since his attempt to make sure that every WU achieves the Credit it deserves, looks like cheating at first glance.

I have not tested them yet but maybe some of you want to do so

I am testing it on two computers right now, an Athlon 64 3200 and a P4 - 2.4 HT. I will let you know how they do. I can tell you that the benchmark numbers are about 10-15% lower than with crunch3r's optimized clients, but according to the Trux web site that may not matter,

[TiloProbst]

I know quite precisely about the RAC increase on PCs, I'd find it more interesting to see Trux clients on Macs

[Shaktai]

Originally Posted by TiloProbst

some PC buddy just informed me that the much talked about Calibrating BOINC Core Clients by Trux recently had their mac release.

take a look here: http://boinc.truxoft.com/core-cal.htm
-> Downloads

there have been quite some noise about Trux clients (at planet3dnow forums for example), since his attempt to make sure that every WU achieves the Credit it deserves, looks like cheating at first glance.

I'm testing on my iMac G5 with Einstein. I have a pretty good sense there on what it should do. Will let you know the outcome. Intial benchmarks are lower, but will see how the claimed credit works out.

[OneMacGuy]

Originally Posted by TiloProbst

I know quite precisely about the RAC increase on PCs, I'd find it more interesting to see Trux clients on Macs

While you may know quite precisely, many here do not, including me. So I was not just answering to you.
However, my initial observations show a slightly lower RAC on my windoze crunchers than I had gotten used to using the crunch3r optimized client.

[Todd Madson]

So what is the benefit if RAC is lower? Is claimed credit exactly what it should be or more?
How does this affect your account over time?

[OneMacGuy]

Originally Posted by Todd Madson

So what is the benefit if RAC is lower? Is claimed credit exactly what it should be or more?
How does this affect your account over time?

Well, some claim to have a much higher RAC when using this client, and maybe they do depending on what hardware they run it on. On my two test machines, I lost points on both benchmarks and there was no time added to my actual times to make up the difference in points. I think that this works best on really high end machines, but I am not certain of that since I do not have any of them to test it on.

Hey guys!

I was the one who compiled the trux calibrating client for the Mac.

it really does work!

Here is my mini http://setiathome.berkeley.edu/show_...hostid=2250966

And this is a 750mhz g4 at work http://setiathome.berkeley.edu/show_...hostid=2257091

To check and see if it is really working when you open boinc manager and click the message tab (or if you start the command line client) you should see something like this on startup:

2006-03-19 00:01:01 [---] General prefs: from seti@home (last modified 2006-03-11 20:35:36)
2006-03-19 00:01:01 [---] General prefs: using separate prefs for school
2006-03-19 00:01:01 [---] Remote control allowed
2006-03-19 00:01:01 [---] Listening on port 31416
2006-03-19 00:01:01 [---] truXoft add-on: return_results_immediately
2006-03-19 00:01:01 [---] truXoft add-on: calibrate_credit = all
2006-03-19 00:01:01 [---] truXoft add-on: check_max_time = 2.5

Then when the client is actually calibrating you'll se something like this:

2006-03-19 00:17:47 [seti@home] CC calibration: 5.05 >> 30.70 (time: 3327s >> 5378s / Gfpops: 0.73 >> 7.97)

The big thing to remember when using this client is to put the truxoft_prefs.xml (when using the included boinc manager) into the Library>Application Support>BOINC Data folder of your startup disk.

it will take crunching between 20-50 work units before you really start to see the effects of the calibration getting you up near the claimed of 28-32 credits.

Hope this helps!

[Todd Madson]

Boog: is there a benefit to running a faster machine with this or slower?

If I'm already verging on a 2000 RAC with my biggest machine is there a benefit
to moving to this?

I have both a G4/400, a G5 2.5 dual, an Athlon 2400 box, and a Pentium 1.7 laptop.

Also, did you start your mini and G4 at the same time with the calibrating client?

Looks interesting, I might try it on one or two machines in my farm.

[sdubz]

Originally Posted by Todd Madson

Boog: is there a benefit to running a faster machine with this or slower?

If I'm already verging on a 2000 RAC with my biggest machine is there a benefit
to moving to this?

I have both a G4/400, a G5 2.5 dual, an Athlon 2400 box, and a Pentium 1.7 laptop.

Also, did you start your mini and G4 at the same time with the calibrating client?

Looks interesting, I might try it on one or two machines in my farm.

It seems to me, if your using an optimised seti client, and claiming less than 20, this will benefit your RAC, slower machines can end up claiming this just because they are slow (my rac has went up from almost 1700 to over 2k!).

I started the mini well before the G4 at work, the work machine has only been crunching since the 14th of March. The mini has been crunching a lot longer.

Putting trux on all your machines will make a big difference in your granted credit, you won't always get what you claim, but you will see a difference, unless your machine is already claiming close to 32 credits with an optimised seti client.

[Todd Madson]

K - here's my findings so far.

I put it on my little Pentium M laptop here at work and my first block
submitted shows the corrected time of 1544 seconds as opposed to
the usual 2200 or so. Claimed credit was 5.44 but granted credit
was 31.66 for the first block submitted. A friend on my team set up
his PIV machine to run with this and the first three blocks submitted
with this have similar findings.

Here's that machine:
http://setiathome.berkeley.edu/resul...hostid=1897710
As of right now, that CPU has 543.76 RAC, 22105.61 total credit,
with a daily result number of 292. I should know later today just
how well this has done.

The thing that I don't think people quite realize is that:
-you can run with Alex Kan's optimized crunching client along with
this so you get the speed and the credit and more credit than you
should actually receive.

Most interesting - it does sound like a cheat in some respects but if
it is within the system and Seti isn't upset about it I'm not complaining.

All of my CPUs including my big box will be running with this by tonight.

[sdubz]

Originally Posted by Todd Madson

K - here's my findings so far.

I put it on my little Pentium M laptop here at work and my first block
submitted shows the corrected time of 1544 seconds as opposed to
the usual 2200 or so. Claimed credit was 5.44 but granted credit
was 31.66 for the first block submitted. A friend on my team set up
his PIV machine to run with this and the first three blocks submitted
with this have similar findings.

Here's that machine:
http://setiathome.berkeley.edu/resul...hostid=1897710
As of right now, that CPU has 543.76 RAC, 22105.61 total credit,
with a daily result number of 292. I should know later today just
how well this has done.

The thing that I don't think people quite realize is that:
-you can run with Alex Kan's optimized crunching client along with
this so you get the speed and the credit and more credit than you
should actually receive.

Most interesting - it does sound like a cheat in some respects but if
it is within the system and Seti isn't upset about it I'm not complaining.

All of my CPUs including my big box will be running with this by tonight.

Give it about 12 to 30 work units to start to see the big difference of your claimed going up.

[Todd Madson]

I ran the Trux client on the G5 for two days and was not achieving performance
as well as I was previously with Alex and Rick's client so I've pulled it off.

However, my slower machines are working well with it (2.2 ghz Athlon on Linux,
Pentium M 1.7 ghz and G4/400).

[sdubz]

Originally Posted by Todd Madson

I ran the Trux client on the G5 for two days and was not achieving performance
as well as I was previously with Alex and Rick's client so I've pulled it off.

However, my slower machines are working well with it (2.2 ghz Athlon on Linux,
Pentium M 1.7 ghz and G4/400).

If you were going by the times listed on the berk site, then it will look like you lost performance. The trux boinc "tweaks" the time as part of it's "calibration" method, so, if you were using the same optimized seti worker as when using the one from Alex and Rick, then your actual wu output should be the same.

The actual time can be seen if you click on one of the "Result ID click for details" links of a finished work unit. Then you will see something like this:

<core_client_version>5.3.12.tx36</core_client_version>
<real_cpu_time>6528</real_cpu_time>
<corrected_cpu_time>8040</corrected_cpu_time>
<corrected_Mfpops>3985.6</corrected_Mfpops>
<stderr_txt>

the actual time to complete the work unit was 6528 seconds, the "reported" time that it took was 8040 seconds, to achieve a Claimed credit of 30.11.

It will also correct the Mfpops, to get you closer to the claimed 32.32.

[sdubz]

Originally Posted by Todd Madson

I ran the Trux client on the G5 for two days and was not achieving performance
as well as I was previously with Alex and Rick's client so I've pulled it off.

However, my slower machines are working well with it (2.2 ghz Athlon on Linux,
Pentium M 1.7 ghz and G4/400).

I just looked up your G5, I see why your getting a performance "hit", your machine is actually overclaiming on alot of wu's with the client your using, getting "overinflated" results.

[Todd Madson]

I am running the 5.2 alpha version of Alex Kan's optimized client.

I have two options then to try and make it work:

I suppose I could try going back to Alpha 5 which is much slower but not sure if I
would get a high enough RAC score to make up for the loss of speed.

Or, try Alpha 6 but that one is messed up in other ways. Sigh.

[sdubz]

Originally Posted by Todd Madson

I am running the 5.2 alpha version of Alex Kan's optimized client.

I have two options then to try and make it work:

I suppose I could try going back to Alpha 5 which is much slower but not sure if I
would get a high enough RAC score to make up for the loss of speed.

Or, try Alpha 6 but that one is messed up in other ways. Sigh.

if you use the same optimized seti client from here at macnn, then use the trux boinc, you will still be crunching work units just as fast as before, just not overclaiming. But, remember the times on the stats for your computer will say different than the actual times that your computer is actually crunching.

Or just keep it the way it is, as long as berk isn't complaining, why worry?

The main reason for the trux boinc is to get your rac/claimed credit up so that your not claiming only 4-10 credits per wu.

I just end up feeling funny about overclaiming on work units.

[Todd Madson]

The 5.2 Alex Kan/Rick Berry client is from Macnn, it's alpha 5.2,
what is typically available is Alpha 5. But checking that webpage
you refer to it isn't any of those:

The link you refer to in your original post is Alpha 4 - I've not tried
that one, that is an older test version of the optimized Mac client.

5.0, 5.2 and 6.0 were all test versions.

5.0 produced wu's in an hour or just under on a DP 2.5.
6.0 produced wu's in 28-38 minutes but had issues which caused it to be pulled.

5.2 was an updated version of 5.0 that utilized vDSP that was in 6.0 without
the issues. It's generally about as fast as 6.0 (maybe 100-200 seconds
slower) but produces more "correct" results than 6.0 did.

We've seen the results with 5.2/5.0 - RAC dropped and production did
as well. I even tested Alpha 6 for about 12 hours and it went like a
scalded rabbit but again the overclaiming issue I suspect.

Next up: Alpha 4. As long as I'm testing I might as well try it with what
is known to work with Trux.

[Todd Madson]

Update: After a few days working with Trux I can vouch for the fact
that it works very well on the correct clients.

I can verify that if you have the Trux client on a PC running Crunch3rs'
optimized clients you will experience amazing production.

I theorize this is the case because the author of Trux had some
communication with the authors of the optimized clients.

I looked at it running on the Mac G4 400 I have and verified that
though production levels have remained about the same, the RAC
has slowly risen to just under 100 higher than it usually runs.

The Linux box is just over 100 higher than it normally runs as well.

The Pentium M laptop is about 125 higher than previous.

More later. I think this is really interesting how by going slower
in some cases you get can more credit/production.

[sdubz]

Originally Posted by Todd Madson

Update: After a few days working with Trux I can vouch for the fact
that it works very well on the correct clients.

I can verify that if you have the Trux client on a PC running Crunch3rs'
optimized clients you will experience amazing production.

I theorize this is the case because the author of Trux had some
communication with the authors of the optimized clients.

I looked at it running on the Mac G4 400 I have and verified that
though production levels have remained about the same, the RAC
has slowly risen to just under 100 higher than it usually runs.

The Linux box is just over 100 higher than it normally runs as well.

The Pentium M laptop is about 125 higher than previous.

More later. I think this is really interesting how by going slower
in some cases you get can more credit/production.

Cool! I'm glad to see your having some luck with it!

I wish I would have figured out how to build it before now!

[Todd Madson]

I'm also trying it on the wife's 1.25 ghz G4 - we'll see how it works over there too.

More when I get a chance to try A4/Trux on the G5.