Hey ROSETTA (or anyone else) people, what BOINC client do you use?

[Kelemvor]

Howdy,

I'm Kelemvor in Boinc for those that don't know...

Anyway, I'm just wondering which BOINC client or modified client everyone is using and what sort of average numbers you all are getting.

Here's my main units from the BoincSynergy site. I just upgraded them all from the 4.5 client to the 5.2 client or whatever the latest official one is. I'm not sure if using a different one will help my numbers. They are all installed and set to run as a Service. They also only run when the computer hasn't been in use for 1 minute. They are walk-up PCs in our building so they are only used sporadically throughout the day and usually not for very long at once.

I read on one page that any client version won't help you process more units because the processor can only run so fast but they do something that modifies your numbers so I'm not really sure what all that means.

Anyway, here's my info. The top 6 have been running Rosetta for a while (not sure why the one has such a higher RAC) and the others just got added last week.

CPU Vendor/Model ---------------------------------------- CPUs ---- Operating System - Total Credit - Recent Credit
GenuineIntel Intel(R) Pentium(R) 4 CPU 3.00GHz ------- 1 ---------- Windows XP ---------2,271 -------- 140.88
GenuineIntel Intel(R) Pentium(R) 4 CPU 3.00GHz ------- 1 ---------- Windows XP -------- 1,514 ---------- 91.98
GenuineIntel Intel(R) Pentium(R) 4 CPU 3.00GHz ------- 1 ---------- Windows XP -------- 1,426 ---------- 85.68
GenuineIntel Intel(R) Pentium(R) 4 CPU 3.00GHz ------- 1 ---------- Windows XP -------- 1,334 ---------- 88.92
GenuineIntel Intel(R) Pentium(R) 4 CPU 3.00GHz ------- 1 ---------- Windows XP -------- 1,190 ---------- 76.65
GenuineIntel Intel(R) Pentium(R) 4 CPU 2.26GHz ------- 1 ---------- Windows XP -------- 1,148 ---------- 68.85
GenuineIntel x86 Family 6 Model 8 Stepping 6 860MHz - 1 ---------- Windows 2003 -------- 215 --------- 86.65
GenuineIntel Intel(R) Pentium(R) 4 CPU 3.00GHz ------- 1 ---------- Windows XP ----------- 190 ------- 135.85
GenuineIntel x86 Family 6 Model 7 Stepping 3 548MHz - 1 ---------- Windows XP ----------- 182 -------- 62.16

 

[mondobyte]

Franky,

I would use CPUz (free utility ... just Google it) to determine your level of SSE support available for each of the computers.

Then I would go to Crunch3R's (Google Crunch3R and select the ALPHA entry - 3rd one down - then navigate to the BOINC clients) site and use the "optimized" BOINC client that is appropriate for your SSE support.

Should give you much better performance for most, if not all, of your computers.

I am no expert by a long shot but I do have at least a couple computers running BOINC these days. Check out my SETI, SIMAP, or PrimeGrid stats ...

mondo

 

[BofRA]

For the most part those of us in the Rebel Alliance are using clients from here. This client has atleast doubled if not tripled my benchmarks

 

[Kelemvor]

What exactly do the different benchmark numbers mean as far as how they affect my Boinc stats?

I slapped the Crunch3R on my laptop (Pentium M 1.6) here and went from:

01/14/2006 10:58:08 PM|| 1365 double precision MIPS (Whetstone) per CPU
01/14/2006 10:58:08 PM|| 2846 integer MIPS (Dhrystone) per CPU

to

01/14/2006 11:02:13 PM|| 1906 double precision MIPS (Whetstone) per CPU
01/14/2006 11:02:13 PM|| 6217 integer MIPS (Dhrystone) per CPU

Whet: 39% increase
Dhry: 118% increase

So obviously big numbers are better but I don't really know what the difference is....

 

[mondobyte]

Originally posted by: BofRA
For the most part those of us in the Rebel Alliance are using clients from here. This client has atleast doubled if not tripled my benchmarks

I tried those and found the Crunch3R a bit better. Have you tested both and what is your verdict?

Thanks

mondo

 

[Dalephi]

I also use the Truxoft client. I am not sure about the other clients but with this one you can send back results as soon as they are completed.

 

[Rattledagger]

Orignially optimized BOINC clients was made so non-windows-users should claim roughly the same credit as windows-users. Later it's also to be paired with optimized seti-applications.

But, in other projects than SETI@Home that don't have optimized science-applications, the optimized BOINC clients only gives artificially inflated benchmarks, and by extension artificially inflated claimed credits. Since Rosetta@home isn't using redundancy, these inflated claimed credits is directly translated into granted credits, and can therefore be seen as a form of cheating...


Looking specifically on "Return results immediately", this was a debug-command in the official clients that was removed in v4.xx or something, and that Truxoft has added back in. Reporting results immediately gives both higher database-load and scheduling-server-load than letting client wait till next time asks for work...

 

[Smoke]

Originally posted by: BofRA
For the most part those of us in the Rebel Alliance are using clients from here. This client has atleast doubled if not tripled my benchmarks

It has definitely made a difference. TE may have to start running these opt versions.

 

[Freewolf]

Originally posted by: Smoke

Originally posted by: BofRA
For the most part those of us in the Rebel Alliance are using clients from here. This client has atleast doubled if not tripled my benchmarks

It has definitely made a difference. TE may have to start running these opt versions.

I thought you already was.
We seem to have a couple more host running than ya'll do right now also.

 

[Kelemvor]

So if I look at my benchmarks and compare them, is it basically the higher the benchmark numbers, the higher my credit rating within ROsetta will be? So I can put on crunch3r and run the benchmarsks, then put on the trux and run the benchmarks and whichever oen gives me better numbers is the one I should use?

And which benchmark is more inportant? whet or dhry? Not sure what each one signifies.

 

[Freewolf]

Yes and I get slightly higher benchmarks from the crunch3r . I don't know which of the two benchmarks have more importance.

 

[Smoke]

I've tried various opt vers on my home computers but have not decided upon which vers is best for the rest of the team (TE).

Which is more important: Whetstone or Dhrystone?

It seems on some vers, one is higher than the other and then on another vers - the reverse is true.

edit: That's not exactly what I meant to say. Dhrystone is always higher than Whetstone. I meant the improvement over the basic client seems to vary.

 

[petrusbroder]

And now a "monster" post by petrus (if you dislike long posts, skip it now! 😉 )

Ahumm: I quickly looked into this matter using an optimized BOINC-client (i.e. the boinc.exe and boinc.dll - files):
The standard client is version 5.2.12, the optimized client is from Crunch3R. The results of the benchmarks as run using BOINC itself.


Processor _______________ Standard BOINC-client _________ optimized BOINC-client__________ Change in % ___
_______________________ Whetstone ___ Drystone _______ Whetstone ___ Drystone ______ Whetstone ___ Drystone

Athlon XP 1900+ (SSE) ___ 1351 ________ 2259 ___________ 1581 ________ 4310 ___________ 17.0 ________ 90.8
Athlon XP 2400+ (SSE) ___ 1856 ________ 3141 ___________ 2169 ________ 5966 ___________ 16.9 ________ 89.9
Athlon 64 3000+ (SSE2) __ 1888 ________ 3503 ___________ 2739 ________ 8349 ___________ 45.1 _______ 138.3
Intel P4-M 2.0GHz (SSE2) _ 1770 ________ 3657 ___________ 2452 ________ 7955 ___________ 38.5 _______ 117.5


Note to the measurements:
The numbers are averages of 4 measurements. The standard error was less than 1%
I am using the computers for different things: multi media computer, a file server, communication computer, game computer for my children. They have different sets of drivers, memory resident programs etc, and are thus not easily comparable with other comps.


The results make me ponder:
When using optimized science-applications (such as for Seti, which uses the SSE or SSE2 functions, see here for a thread ), the use of SSE- or SSE2-optimized clients (i.e. the optimized boinc.exe) may be justified because with the (non-optimized) standard client you get too low credits.

When using the SSE- or SSE2-optimized BOINC-client with science-applications which do not use SSE or SSE2 you measure the efficacy of functions which the science application does not use.

This may be a problem when running several projects (with the standard BOINC-client) at the same time as running Seti with optimized seti-application: the seti-stats get somewhat too low credits.
On the other hand: If the optimized BOINC-client is used Seti gets the correct credits, but the other (non-optimized) projects get inflated credits.

The use of optimized clients (boinc.exe) with non-optimized science applications makes the comparison between teams, individuals or even computers very difficult because the measurement with the optimized boinc.exe (using SSE or SSE2) does not reflect the work of the not-optimized science applications which do not use SSE or SSE2.

On a personal note (and this is my opinion): I have decided to use the standard BOINC-client because I run 3 - 4 other projects at the same time and since the penalty is small: in Seti the shortest and longest crunching time is dropped and the two middle values are used for credits. I almost always get 10% - 200% more credits than I claim - and that without using the optimized BOINC-client; I never get less.
I also understand that many crunchers use the optimized BOINC-clients. That does not bother me the least. 🙂

I have the impression that a better method for calculating credits is being developed, a method which takes into account if a processor has SSE or SSE2 or SSE3 or MMX or what not. :thumbsup:

In due time the credits will really reflect the calculations performed. Today the credits have (mostly) the function to stimulate crunching and competition. Since probably most of us, as a team and as individuals, compete and race with each other and with other teams for fun, micrometer-justice serves no purpose. It is much more important that we enjoy the crunching, racing, knowing and posting. As you all can see, I at least enjoy all of this. 😀

 

[petrusbroder]

Originally posted by: Smoke
I've tried various opt vers on my home computers but have not decided upon which vers is best for the rest of the team (TE).

Which is more important: Whetstone or Dhrystone?

It seems on some vers, one is higher than the other and then on another vers - the reverse is true.

edit: That's not exactly what I meant to say. Dhrystone is always higher than Whetstone. I meant the improvement over the basic client seems to vary.

As far as I know the credits ar determined by calculating CPU Time multiplied by the CPU Benchmarks (which is Whetstone + Dhrystone).

From the BOINC-wiki:
One cobblestone is is 1/100 day of CPU time on a reference computer that does
* 1,000 double-precision MIPS based on the Whetstone Benchmark.
* 1,000 VAX MIPS based on the Dhrystone Benchmark.

This does not make things easier, but the Dhry- and Whetstones are equally important. If either of the "stones" increases the sum (see above) will be bigger.

 

[Smoke]

If I understand you correctly, then another way of saying this would be: Whichever version produces the highest SUM (of Whetstone + Dhrystone) would be the most efficient to use on any particular system?

 

[petrusbroder]

Yes. But that is still only benchmarking - does not affect the crunching of the WUs and does not affect the crunching efficiency of the system.
The optimized client only affects how the credits claimed are calculated. The WUs still take the same time to crunch with the standard client and with the optimized client.
The efficiency of the crunching is affected only by an optimized science-application, and those exist only for Seti@home as far as I know.

 

[Kelemvor]

Well I guess I'll go through on Monday and put the Optimized versino on my PCs and see what happens to the numbers.

Most of them are P4 3.0Ghz machines so we'll see what happens.

 

[networkman]

Hmm.. having a larger fleet means upgrading to "optimized" clients is going to take some time in doing. 😉

Any hints to save me some time? Currently, I'm running solely the Einstein@Home project, and I know that they do utilize redundancy in granting credit.

The makeup of most of fleet can broken into just four groups:

a) a couple Intel dual p3-800 rigs
b) a couple Intel P4 2.1 to 2.8 rigs
c) many AMD XP1800 to XP2500 rigs
d) a few AMD Dual MP1800 to Dual MP2200 rigs

Currently, I am using Boinc Manager 5.2.2 for all of them.

Thanks in advance!

 

[mondobyte]

Use ORCA to transform the BOINC MSI and create a separate MSI transform for each of SSE, SSE2, SSE3 ...

I use active directory to distribute BOINC and/or update it ... like TV Magic Cards, once you know the secret ... it is so stinking easy ...

One could also distribute BOINC and/or updates via script ... I do that too for the project clients ... again easily done.

a, c, and d above are all SSE
b are probably SSE2 but they would benefit significantly from running the SSE client.

mondo