LHC@home

Hi there,

could someone tell me what happend to
canonical result 3039723 -> my resultid: 3039726 by Computer 53613
15 Sep 2005 5:58:50 UTC 16 Sep 2005 4:57:38 UTC Over Success Done 6,520.75
claimed 27.08 granted 12.82 - all others granted 25,64

Why am I receiving just half the amount of credits the others do? Can someone pls. explain this behaviour to me?
Best regards
5th.rider

---

<blockquote>
Why am I receiving just half the amount of credits the others do for???? Can someone pls. explain this behaviour to me?
</blockquote>

This is described here and should
go into the FAQ.

HTH

Michael

---

Team Linux Users Everywhere

The WU said "Client Error", mean your client miss calculated some data, your PC might doing something else harsh for CPU scrambled the process data. This happens to high overclocked CPUs as well.

Hi GX1020, Michael,

thx. Didn´t see the "invalid state".
There was cdpn project which blew up (already did find the reason - my antivir). It seems that this also destroyed this particular lhc... :-(

@Michael -yes, it´s a overclocked machine - but quit stable - no problems til now (athlon64 2Ghz@2,6 right now, still trying to reach 2.8 for comparing with my 2.8 P4 - altough the amd is already much much faster).
Thx for your fast help!
5th.rider

---

<blockquote>it´s a overclocked machine - but quit stable - no problems til now (athlon64 2Ghz@2,6 right now, still trying to reach 2.8 for comparing with my 2.8 P4 - altough the amd is already much much faster).
Thx for your fast help!
5th.rider</blockquote>

I tried to overclock FSB from 166Mhz to 181Mhz (2500+ to 3200+) But the computing power is the same. Also for my overheating problem i use Thread Master and work very well.. which reserved some CPU resource for me access on some websites.

I set Thread Master to use 80% CPU power to process WUs. And finished WU in the same time as the 100% CPU in use mode. Good thing is, it reduced the heat without slowing process!

For full test of P4 and Athlon64 performance. Suspend the WU processing. Go to Client, File -> Run Benchmark. And view the mill-flops via your account. As i know, my AthlonXP 2500+ beats some Pentium 4 3.2Ghz .. woohoo~!

@GX1020
HI, I can see a huge rise of my computing power by increasing fsb (and not touching multiplier).
I benched follwing results:
athlon64 wichestercore def.@2GHz, ram ocwear pc3200 def.@200Mhz
cpu ram BoincBench
2000 200 -> 1887/3531
2400 200 -> 2270/4229
2600 217 -> 2502/4673
tbc ;-)
in fact it seems for computing power that the cpu doesnt really care if it is triggered i.e. 300*9 or 270*10 - benchresults are about the same.
thx for threadmaster hint - didn't know this. I'll try later on.

5th.rider

---

5th.rider,

It's not recommended to overclock your processor and work on FPU-intensive BOINC projects. You may have never noticed a problem until now because your OC doesn't affect integer calculations as much as it does FPU calculations. Some of it has to do with the registers, the size of data being stored in them, and the amount of time to accomplish those tasks before moving on. Windows and nearly all standard applications use integer functions in 16 or 32 bit mode and store data 16/32 bits at a time max before moving on. FPU data can be as long as 80 bits which makes it more sensitive to OC .. more than twice the amount of data has to be moved in the same amount of time. It's the reason why if you OC too much and play your favorite FPS like Half-Life 2 you'll see shearing and disappearing polygons (Same story for OC your GPU as well). The data being processed becomes corrupt while trying to store & retrieve and you see graphical glitches. Same thing happens to LHC (et. all) when you OC with these projects you'll be far more prone to calculation errors which if passed unchecked undermine the validity of the work your client sends back to the server and threatens the integrity of the project you're crunching.

There used to be info in the "Getting Started" section on LHC about why it's bad to overclock and attach to this project, but alas it's gone now. Threadmaster can help, but why throttle your CPU when reducing the CPU clock proportional to the throttle setting you use would be a more sound and stable solution?

This project isn't about "who has the fastest CPU" or "who can crunch the most WUs in a day" - it's about donating your resources to further research on beam stability for the Large Hadron Collider at the CERN Institute. I'm not slamming you or anyone who chooses to OC .. I'm saying it's best to not do it on this or any project - do it on your games but not here.

---

Hi Travis,

thx for your advice. I didn´t know about the problems OC could make to lhc.
In fact Boinc projects are (also) used by me for stability (and performance) tests because of their intensive work.

I thought that I would be able to recognize a "too high OC" by receiving "invaild results" if I'd reach a level cpu/ram/whatever hw-component couldn´t work properly. like all other software i use for testing, if it get´s unstable I am told. Isn't there a similar behaviour by boinc by validating the result? Pls correct me if I am wrong at this?

At last - yes, this is about donating my resources for cern. ( and I do not feel slammed at all) - but why donate less I can?

---

<blockquote>
I thought that I would be able to recognize a "too high OC" by receiving "invaild results" if I'd reach a level cpu/ram/whatever hw-component couldn´t work properly. like all other software i use for testing, if it get´s unstable I am told. Isn't there a similar behaviour by boinc by validating the result? Pls correct me if I am wrong at this?
</blockquote>

No, but this is exactly what happened to the
result
you mentioned. This result is INVALID. You only received
credit because of LHC's special policy regarding invalid results.

Michael

---

Team Linux Users Everywhere

<blockquote>5th.rider,

It's not recommended to overclock your processor and work on FPU-intensive BOINC projects......'
Thanks DJ for the info re. the difference between integer and FPU processing and the impact of OC. It's great to have clever people like yourself on the message boards to help the rank and file people like myself understand the limitations of their machines. Cheers, Rog.

---

Hi there,

maby I am making myself looking stupid now...
The invalid result I produced was caused by my AntiVirus-prog, which doesn't like CPDN and crashed the whole boinc-client.
I never had any troubles concerning invalid results - so I can't follow you right now.
I do agree Travis DJ, and I can follow ( and I do agree with) his argues. But (again) - as long as I am receiving validations - the computing on my side should be fine? Or is it possible to scrumble a result without lhc recognizing? Could they validate a scrumbled result?
sorry for asking this again - as I said, I also use Boinc-projects for being sure I don't OC too high and have a healthy system.
Thx to you all!
5th.rider

---

@5th.rider & ARmassey-

You're welcome. :)

There's all sorts of levels to where OC is good or bad. A "good" example buying DDR533 on a CPU whose FSB is 400 (200*2)- It would be a good idea then to OC the FSB to 533 (266*2) and then reduce the multiplier of your CPU to get it close to it's design spec.

Which AV do you use, 5th? It would be worthwhile to look into what files you can "exclude" from the realtime and on-demand scans.

ARmassey, if you wanna know something about general cpu & motherboard architecture, ask - If I don't know, someone else on here will. It's a shame we can't search the message boards, there's a wealth of useful info in it.

---

@Travis DJ
first - thx for your advices - I really do appreciate!

My AV was from H+B EDV, called AntiVirGuard - at http://www.free-av.de

After crashing of the boinc-client I was wandering through the CPDN fora and found that this one's non-compatible. After deinstalling the AV now CPDN seems to run - at least computing is further it was when the client crashed. The failure doesn't seem to be caused by filechecking, because disabling AV doesn't help. At least none at the cpdn fora did solve this - and I surely won't be able too. So a new AV will be next.

Of course I do not want to danger the results I provide - better say the computing I donate. And of course I do OC very smooth - 5Mhz up, testing, and so on. I didn't reach the limit of the cpu or the ram yet (tested cpu 2800, ram 250), but this was too hot and noisy because all fans needed to run fullspeed. And I didn't feel well about OCing the RAM that much. So I reduced to my current config - but this is still "under costruction". And again of course, all is tested with prime, superpi, memtest,..., and boinc projects.

One last thing I like to contradict - in my opinion it is a matter of "how many results". Results - not credits! Because it's all about providing CERN (and all the other boinc-projects out there) the best support I can offer. If I can help CERN scientists meeting their deadlines, making this gigantic machinery work - that´s what I crunch for. Not for beating statistics or personal records - although it's nice to compete sometimes ;-)
5th.rider

---

If OC works than is fine, but note that there's high chance to corrupt the data packets during processing. So best if you can "leave it alone" with no important background running programs (reserve all resources to BONIC client). Oh yea, I do OC my PC, with only added 1Mhz in FSB lol .. Not to memtion, there's also good change to build a very stable overclocked machines. I heard Asus mainboard has the quality of stable for overclockers.

Last thing, the main objective for me is to finished WUs as fast as my PC could. Coz i want to see how will this LHC work in later year 2007. Any delay will not be appreciated (for me) // Mostly same objective as 5th.rider

Don't forget that most AV packages also have a process running to try to stop in memory activity. I am not sure if Norton AV was causing my CPDN crashes as I have never uninstalled it to check ... hmmm, if/when I get time I guess I will have to try to remember to do that ... OS-X is pretty robust so the risk should be low I suppose ...

---

<blockquote>Don't forget that most AV packages also have a process running to try to stop in memory activity. I am not sure if Norton AV was causing my CPDN crashes as I have never uninstalled it to check ... hmmm, if/when I get time I guess I will have to try to remember to do that ... OS-X is pretty robust so the risk should be low I suppose ...</blockquote>

My solution, do not use AV.

Do not visit unknown websites. Do not use IE. Do not download software using filesharing utilities. Use webmail instead of a client.

I check for viruses every 6 months or so (a quick install of Norton usually does the trick). I haven't had a virus since I got rid of my Amiga 1200 (>6 years ago).

Just be careful out there1

---

<img src="http://www.boincstats.com/signature/user_267407.gif">

<blockquote>
Do not visit unknown websites. Do not use IE. Do not download software using filesharing utilities. Use webmail instead of a client.
</blockquote>

Don't use Windows...

Michael

---

Team Linux Users Everywhere

Trend Micro offers (IMHO, from experience) a fantastic Antivirus, Antispyware, and Firewall program called PC-Cillin Internet Security 2005 and can be had at Newegg.com for about $33 USD via digital delivery.

Unlike Norton or McAfee AV, Trend's home product(s) don't rely on Internet Explorer or ActiveX technologies to administer, install, activate, or otherwise use the program. It among other technologies makes PC-Cillin very lightweight, mostly user-transparent, and durn fast. Also, as in 5th.rider's case with his bad AV/CPDN experience, he could exclude anything in the BOINC folder from realtime and manual scans so it won't even bother it at all; in practice that's hardly a necessary measure with PC-Cillin.. never had a moment's problem with it and BOINC.

:)

---

Hi there,

@Travis DJ - thx again for advicing me - haven't decided yet which AV next one will be -one out of kapersky/trend micro/AVK.

@VinylPusher
Oh you are so right! But this is somehow not practicable to me ;-)

@Michael:
I really don't want to start this Windows/linux flame war - I am using both and both have their advantages ( and their problems).
But a somehow astonishing detail:
I do use two absolutely identical machines - one based on linux, one running win xp.

take a look at the benches:

GenuineIntel Intel(R) Pentium(R) 4 CPU 2.80GHz
Number of CPUs 1
Operating System Linux 2.6.11.4-21.7-defaultdell p4 2.8 linux
Measured floating point speed 706.54 million ops/sec
Measured integer speed 2157.9 million ops/sec
--
GenuineIntel Intel(R) Pentium(R) 4 CPU 2.80GHz
Number of CPUs 1
Operating System Microsoft Windows XP Professional Edition, Service Pack 1, (05.01.2600.00)
Measured floating point speed 1419.5 million ops/sec
Measured integer speed 2736.3 million ops/sec

That´s quite interesting, isn't it?

---

There's already a story said Linux process slower. As it shown in measured floating points, so i recommend stick to WindowXP ;)
Story~~ http://lhcathome.cern.ch/forum_thread.php?id=1700

For my PC security i use Norton Internet Security - Anti Spyware Edition. Real-time in memory virus/spyware/adware detector. personal firewall, outbreak alert, intrusion detection, privacy control, anti-virus and anti-spam. Yet it didn't caused any problem lately. It does launched virus scanning automaticly every friday night but i stop it everytimes to prevent corrupt process data or slowing process down.