Optimized clients

[DDTUNG]

I've been experimenting with optimized clients for Rosetta, available here:

http://www.guntec.de/Crunch3r/boincx86.html

For AMD DC I'm using the 5.2.13 client, and for Intel DX 5.2.11. Pick the client that is appropriate for your rig.

Download the clients, unzip each into its own folder, then copy and paste into the BOINC folder, overwriting the standard files. Make sure you exit Rosetta before copying. Compare the cpu benchmarks before and after the switch, and you will see the difference.:)

I believe you should be producing roughly 3 times as many points per day in rosetta as you were in D2OL.

DDTUNG:cool:

[[XC] Duc]

I've been experimenting with optimized clients for Rosetta, available here:

http://www.guntec.de/Crunch3r/boincx86.html

I believe you should be producing roughly 3 times as many points per day in rosetta as you were in D2OL.

DDTUNG:cool:

So you'll be producing about 50k a day once you've switched over your farm completely?:slobber:
(Or even more with the added AMD DC rigs you're building right now?)

[DDTUNG]

Around 40k, plus new builds.

But my point is the XS D2OL team should be capable of 150k. Don't tell the cows.:)

DDTUNG:cool:

[Hardass]

Done. Still not sure if settings are correct. Where is that benchmark folks are running??

[k0nsl]

Where is that benchmark folks are running??

Commands> Run benchmarks

:)

-k0nsl

[Hardass]

Commands> Run benchmarks

:)

-k0nsl

So you run a benchmark while program is running at the same time?

[k0nsl]

So you run a benchmark while program is running at the same time?

That is how I figured everybody else did it. Maybe it is possible to perform it by using the command line, I don't know, because I did not try.

I just paus all the work and proceed with the benchmark.

-k0nsl

[Hardass]

Sorry DDT for taking up your thread. Thanks for help. I got:
1365
1941
Dual xeon 1.6 @ 3000mhz with 512mb Corsair BH-5.

[L0$t Pr0PhEt]

If you start the benchmark it will automatically pause the work. I believe most of us are allready using the optimized clients. I get 4 to 5 times the daily production of d2ol on my a64 rig :).

@ Hardass: that's a quite low benchmark you got on your DX rig, or is this on every DX rig? Check the benchmark thread.
I checked it and it's almost normal, but do you have to multiply it by 4 or not because of the 4 threads?

[DeltZ]

yeah..

benchmark finds that value for each core

including logical cores

so x by 4 to find to find max output of all logical cores i guess...

would have expected it to be higher though....
maybe these values arn't as straightforward as i thought. ..hmm.

[DeltZ]

also...concerning the optimized clients on that website

since they've changed from rar to zip so you come up with a file index..


anyone tried the version xxxxxxx.14 HT EDITION?

DDT?

or is the older version really actually better?

EDIT: nvm...they are Linux .out file types or something...so nvm.

[DDTUNG]

I just did a quick check and found that while some XS team members are using the optimized clients, many others are not. Production can double once these clients are used, so if you haven't already done so, switch over to the optimized clients now.

DDTUNG:cool:

[lv_dicedealer]

I just did a quick check and found that while some XS team members are using the optimized clients, many others are not. Production can double once these clients are used, so if you haven't already done so, switch over to the optimized clients now.

DDTUNG:cool:

Could you edit your first post using big bold type to tell us morons which client we should be using, this multi-client crap is annoying already!

ps... make it even easier, and use hotlinks for the direct d/l of the particular client depending on cpu

[afireinside]

Stevil links to these patches in the stickies ;)

[lv_dicedealer]

Stevil links to these patches in the stickies ;)

I am aware of that, I was trying to get DDT to make it as easy as possible for ppl to get and use the optimized clients best suited to their particular cpu.

On a side note: I have found the benchies to vary wildly... Integer MIPS can vary by almost 500 on my machine (Intel DX using 5.2.13)... Double floating point doesn't vary as much, but I did score 500 more using the 5.2.13 v. 5.2.11...

Summary: Integer higher on 5.2.13, double floating point higher on 5.2.11

So, which is more important: Integer or Double Floating Point?

[Movieman]

Question:
Just installed Rosetta today using ver 5.2.13 as per their suggestion on the site. Do I read this correctly that the 5.2.11 version is what I should be using? And are the versions on the site that DDTUNG linked to "modified
in any way so as to optimise the client?
I'm running all Intel DX systems.
Thank!;)

[ex2cib]

there is a link in the getting started thread with links to optimized BOINC clients as compared to your system, ex. Athlon, athlon 64, pentium 4, pentium 3, etc.

you should download the optimized client, and unzip all the files in the download into your BOINC directory, this will optimize the boinc client specifically for your type of processor, and it does speed up production considerably

[Hua Luo Han]

*refering to P4-3.2C(HT) on Linux 2.6.12-10-686-smp & non-smp

though i'm using the optimized cleint, but still unsure of the following :

you run benchmark without smp option, result about 3k,6k
with smp result maybe 2k,2k
so if u start pc and opt non smp to bench a higher score, and restart in smp mode, but dun benchmark, the higher benchmark on single thread stays and you get higher credits while crunching 2 instances ? :confused:

sorry if i have confuse anyone :confused:

[ex2cib]

im confused

[WesM63]

Hua,
The way I understand it works is anytime the agent is re-started it runs the benchmark so it knows what kinda work/points to give for said computer. If your running HT the benchmark will only show 1 core so multiply that by 2 to get the actual benchmark.

I hope this helps.

[Hua Luo Han]

i read somewhere tat benchmark will auto run every 24hours but my rig is not 24/7, so ever since the last time i ran benchmark, the score stays for (non-HT) but i'm actually running with HT now

CPU type GenuineIntel
Intel(R) Pentium(R) 4 CPU 3.20GHz
Number of CPUs 2 <<<-------------- HT ON
Operating System Linux
2.6.12-10-686-smp <<<------------ smp selected
Memory 1011.06 MB
Cache 512 KB
Swap space 2957.24 MB
Total disk space 180.54 GB
Free Disk Space 166.61 GB
Measured floating point speed 3368.93 million ops/sec <<<----- bench with
Measured integer speed 5951.15 million ops/sec <<<----------- non-HT

still confused :confused: hahah but wth, just fold.

[L0$t Pr0PhEt]

The benchmark runs every 5 days. And as ddtung it's very important to use the optimized clients!!!

[STEvil]

Has anyone tried running only 3 threads instead of 4 to see if "freeing" the one HT will allow easier thread balancing and possibly increased performance for the xeons?

[Vapor]

Or how about turning off HT and running two instances?

[[XC]thewildblue]

That will also drop memory usage as well.

[lv_dicedealer]

OK... I ran six benchmarks with each of the 5.2.11 and 5.2.13 clients on my 3GHz Dual Xeon with HT turned on
All data is listed with Double Precision MIPS first and Integer MIPS second,
so sample one under 5.2.11 is 1406 Double Precision MIPS and 2002 Integer MIPS

5.2.11 5.2.13
1406 / 2002 1947 / 1693
1471 / 2085 2016 / 1541
1485 / 2294 1994 / 1810
1496 / 2120 2011 / 1776
1452 / 2044 2019 / 1772
1467 / 2039 2021 / 1733
AVG 1462.8 / 2097.3 AVG 2001.3 / 1720.8
Total 3560.1 Total 3722.1
DELTA (H-L) 90 / 292 DELTA (H-L) 74 / 269



So, as you can see in my case I have to make a compromise.... Do I gain 538.5 Double Precision MIPS and lose 376.5 Integer MIPS or just go for the Highest AVG Integer MIPS???

Added total as suggested by LP....

[L0$t Pr0PhEt]

OK... I ran six benchmarks with each of the 5.2.11 and 5.2.13 clients on my 3GHz Dual Xeon with HT turned on
All data is listed with Double Precision MIPS first and Integer MIPS second,
so sample one under 5.2.11 is 1406 Double Precision MIPS and 2002 Integer MIPS

5.2.11 5.2.13
1406 / 2002 1947 / 1693
1471 / 2085 2016 / 1541
1485 / 2294 1994 / 1810
1496 / 2120 2011 / 1776
1452 / 2044 2019 / 1772
1467 / 2039 2021 / 1733
AVG 1462.8 / 2097.3 AVG 2001.3 / 1720.8

DELTA (H-L) 90 / 292 DELTA (H-L) 74 / 269



So, as you can see in my case I have to make a compromise.... Do I gain 538.5 Double Precision MIPS and lose 376.5 Integer MIPS or just go for the Highest AVG Integer MIPS???

Go for the higest total (double precision + integer), that's what counts for your total score.

[lv_dicedealer]

Later today I will run tests with HT OFF to hopefully shed some light on this question....

[Rodzilla]

@ DDTUNG and all...

If you downloaded the 5.2.13 Boinc Installer from R@H should you be using the matching 5.2.13 Optimization files as well?? Is it safe to mix these?

And also... make sure you use the correct version (Intel P4/A64 or the AXP/P3) or your BOINC client will refuse to start! :lol: Learned from personal experience :D If you make this mistake just DL the appropriate Optimization client and move the files over... it will replace the files that are keeping your client from starting.

[WesM63]

I put the AXP/P3 ones on my regular old Athlon 900. Works fine lol.

[Rodzilla]

I accidentally put a P4 Op Client on a AXP and it tanked! :lol: Put the right one in its place and it was fine!