Optimized Clients for BOINC - Yes or No?

[Kelemvor]

Originally posted by: Malak
The science may not suffer, but the competition does. I'm sorry, but knowing that one person doing less science can get more points just because they use a special client... that doesn't sound like cheating to any of you? It doesn't matter if it is open source, it doesn't matter if anyone can get it... it's still cheating. In any and every competition in the world that is cheating. Even if the administration doesn't care, I will still think it's cheating. Obviously the administration doesn't care about the competition itself, only the science.

Exactly. The competition is just a game. It's for fun. That's why I don't care what anyone else runs. I just try to tweak to get the best preformance numbers I can from my machines.

By your rules, overclocking your video card to get better performance would be cheating. OC'ing your processor would be cheating. It's just a performance advantage that anyone is welcome to take advantage of if they choose to.

 

[Insidious]

Originally posted by: Malak
... that one person doing less science can get more points just because they use a special client... that doesn't sound like cheating to any of you? ..

Has anyone called YOU a CHEATER because you are intentionally sandbagging your points by running system with a client that is slower (edit: as in point accumulation) than it can be?

Set your system up any way you like.. you own it. I'll do the same. But you dasm well better be prepared to show me the rule I am breaking before you call me a cheater.

-Sid

 

[networkman]

Originally posted by: Malak
The science may not suffer, but the competition does. I'm sorry, but knowing that one person doing less science can get more points just because they use a special client... that doesn't sound like cheating to any of you?

That underlined section is where your arguement is flawed. The whole point here has been that the people with optimized clients are NOT doing less science, they're doing the same amount as anyone else. Where the issue lies, is at the point of the "claimed credit" being returned; the optimized client will likely have a larger "claimed credit" value over that of non-optimized clients. In the Rosetta project where the admins have apparently chosen NOT to make use of redundancy checking for pending credits, that is where the notion of cheating arose. Without that added redundancy, ie. a security check for the validity of the workunit, that is where the notion of cheating arises. In the Einstein@Home project where that redundancy IS implemented, an optimized client is rarely likely to give one any additional credit(over a non-optimized client) unless the others returning the same workunit also using optimized clients.

So perhaps the whole point is that we should determine which projects are or are not using redundancy checking before making any more blanket decisions on who may or may not be "cheating".

 

[mondobyte]

And again, my sentiments exactly Insidious ... well said ...

mondo

 

[rise]

By your rules, overclocking your video card to get better performance would be cheating. OC'ing your processor would be cheating. It's just a performance advantage that anyone is welcome to take advantage of if they choose to.

i don't think thats what he's saying at all. clearly, oc'ing a cpu will allow you to do more work and therefore should earn more credit.

a closer comparison would be your video card example, if you imagine running 3dmarks at "high performance" and claiming it was "high quality" settings. but its still not really the same.

the truXoft author's notes even say flatout that we should not be using this if we are not using optimized project apps.

n the other hand, if you do not use any optimized project application, or spend a considerable part of your CPU time on projects with official unoptimized software, you should not install the optimized BOINC client either, because it would add just more inaccuracy to the credit calculation. Please use these optimized BOINC client only if you dedicated most of your computation time to projects with an optimized client or at projects that do not use the BOINC benchmarks at all (i.e. CPDN).

anyway, i just wish rosetta, who lacks the redundancy check other projects use, would just fix their client if indeed this optimization is used to "level the playing field". that way everyone would be on the same level.

until then, i will continue to use the optimized client as its not returning bad results, i do enjoy the stats competition and (my mother will strike for this) everyone else is doing it

 

[Insidious]

Rise, your points are well taken.

the only place I disagree is that your post makes it appear that the developers of the BOINC project(s) have requested we not use this client. I believe the quote you have put up is from the TruXoft site which is not affiliated with the developers we are refering to when we say 'DEVs' or 'Project Developers'. TruXoft has no input to project policies or rules.

I think the quote from TruXoft is a "CYA" attempt to deflect potential criticism that might be directed at their site. (similar to the criticisms we have in this thread)

-Sid

 

[Crazee]

Very well said rise4130

The redundancy check is not the end all and be all for the stats otherwise they wouldn't be looking for different methods of scoring. The redundancy check may limit the effect of an optimized client but it will not eliminate it. If the highest score is turned in by an optimized client and the second highest score is as well then you will have inflated scores for that workunit.

The only way this is going to really be fixed is when they change the way boinc awards points. Until then I will crunch with optimized clients. One, because every other team is doing it. Two, because I know they will have some method of conversion when they do go to the new scoring system and I don't want to have to work twice as hard to make up a deficit that shouldn't exist.

 

[rise]

sid-oops, my bad man. thats what i get for not linking

fixed it.

 

[Kelemvor]

Honestly, why does the client have anything to do with my Credit anyway. It should just look at the information I send back and give me credits based on that. What computer I run it on shouldn't have any effect on it at all. The system should be configured differently.

Anyway, here's how my clients changed after using the crunch3r files

P4 2.26
W 2024.75 +73.28%
D 3564 +47.52%

P4 3.0
W 2463.75 +65.05%
D 4211.5 +66.69%

P4 3.0
W 2470.5 +65.92%
D 4263.25 +80.00%

P4 3.0
W 2467.75 +66.51%
D 4137 +68.63%

P4 3.0
W 2467.25 +65.87%
D 4236 +69.27%

P4 3.0
W 2461 +65.95%
D 4174 +69.42%

P4 3.0
W 2466.25 +66.44%
D 3989 +71.48%

 

[rise]

the only thing that changes for me is the dry, and that changes wildly.

heres todays edition

1/16/2006 1:40:50 PM|| 2633 double precision MIPS (Whetstone) per CPU
1/16/2006 1:40:50 PM|| 8887 integer MIPS (Dhrystone) per CPU

the whet are always around the same. the dry yesterday was ~9800. i've seen it as low as 4500, which is what it was with the original clent, and as high as 10.800. but even with the optimized, i can bench it and see ~4500 so its weird.

this is using the truxoft one.

 

[Kelemvor]

Originally posted by: rise4310
the only thing that changes for me is the dry, and that changes wildly.

heres todays edition

1/16/2006 1:40:50 PM|| 2633 double precision MIPS (Whetstone) per CPU
1/16/2006 1:40:50 PM|| 8887 integer MIPS (Dhrystone) per CPU

the whet are always around the same. the dry yesterday was ~9800. i've seen it as low as 4500, which is what it was with the original clent, and as high as 10.800. but even with the optimized, i can bench it and see ~4500 so its weird.

this is using the truxoft one.

What kind of processor? I got much higher numbers on my laptop with a Pentium M 1.6 than I'm getting on these P4 3Ghz machines. Maybe I should try the truxoft one on one as a test.

 

[rise]

oops, sorry rig in sig. x2 3800 @ 2700.

 

[Insidious]

I've seen the same thing when I benchmark my X2 (using stock or truxoft). It's like sometimes it doesn't notice that I have 2 processors (erroneous benchies almost always ~1/2 what I expect)

At present, I have

two X2 3800+ computers: W=6.3 D=10.2 (appx. both machines 2.5GHz)
two AXP 2500+ computers: W=2.1 D=6.1 (appx. both machines 2.15GHz)

-Sid

 

[rise]

hey sid, thats the other wierd thing.

your x2 at 2500 should never bench higher than my x2 @ 2700, 2x 1024 @ 2.5-3-2-8 @ 245. (not that memory should have much impact, just showing i'm not running cl 4 or something)

but as we've talked about, it seems almost random what they come up with. i gotta look at what sandra says and see if theres some kinda correlation.

 

[Kelemvor]

Well, here's my results on a P4 2.26Ghz machine.

BOINC 5.x.x standard install---------Crunch3r patch--------------------Truxoft patch
W: 1168.5 D: 2416-------------------W: 2024.75 D: 3564--------------W: 1819.25 D: 4200
Percent Incrase-----------------------W: 73% D: 47%------------------W: 56% D: 74%

Hmm. Oh well. THey are both higher than the original so I guess I'll just stick with Crunch3r since that's the one that's already on there.

 

[Insidious]

Rise

Yeah, it is strange that I can bench above you.

maybe it is our multipliers

I have FSB=250MHz
CPU=X10=>2500MHz
HT = X4 => 1000MHz
RAM = (.83)(250)=208MHz (4x512MB, 2.5,3,3,9 2T)

(I worked really hard to have every component running maxed with no Underclocks.)

Do you see anything here that might explain it? (I'm thinking HTT)

-Sid

 

[rise]

nothing that i see. mines at-

270x10
HT= 3x ( i always set it and forget it as it means nothing)
and my ram is a little tighter and faster and at 1T ( i got lucky as hell, thanks tankguys )

i did a fresh boot benchmark after my last post so it shouldn't be any extra processes either.

i guess the only possible difference would be the HT/LDT but as you can see here its very unlikely it would have any effect at all.

i'll bench my sons machine later and see. his is the same x2 3800 but at stock.

 

[mondobyte]

I discovered at high OC on my opteron that it was producing errors ... hence a bit less overclock and stable / repeatable high benchmarks ...

mondo

 

[rise]

not to be one of those "don't tell me its my oc, my rig is UBER STABLE!!!" guys but i'm very confident its not my oc.

i know this chip can do 2900 if i didn't mind the extra vcore and temps. as is its 24 hour+ prime stable on all 3 tests, occt stable, s&m stable and the final test of 3 hours prime with rthdribl full screen stable.

maybe its not boinc stable wouldn't that be a kick in the pants. it would be back to f@h for me

 

[Insidious]

Originally posted by: mondobyte
I discovered at high OC on my opteron that it was producing errors ... hence a bit less overclock and stable / repeatable high benchmarks ...

mondo

I was thinking the same thing. those timings and 245MHz sounds mighty high for PC3200 RAM.

(I have to use 2T because I have 4 sticks)

@Rise, you might try going down to the next RAM divider and see what kind of impact it has on your benchies

another afterthought: You did turn off Cool-n-Quiet and spread spectrum right?

 

[rise]

I was thinking the same thing. those timings and 245MHz sounds mighty high for PC3200 RAM.

nah, thats just good ram. memtest 12 hours at 2.5-3-2-6 @ 258mhz. the best sticks i've ever owned, better than my ocz tcc5.

mushkin changed the binning of it shortly after i got these sticks to pc3700 cl 2.5 iirc. i know i felt like dirt after recommending them to fellow AT'ers only to see them change the pcb and kill the overhead.

edit- yeah, cool n quiet is bios disabled.

 

[Insidious]

I'm totally jealous of your setup.....

don't understand the benchmark thingy though?

 

[rise]

thanks for the rig compliment. i haven't been so lucky with my sons rig though, oc wise anyway. i think its actually an ati driver issue but i haven't had the desire to get back to it until i get him a real card anyway. so now its just a video and crunching box.

just ran my sons and its 1933/6751. so i guess it scales properly.

 

[mondobyte]

Memory is one thing. The entire system is another.

My "Gold Standard" of stablility is http://users.bigpond.net.au/cpuburn/ CPU Burn-IN.

If it will go 24 hours without an error ... great ...

You will need to run 1 instance for Each Processor or 2 per Processor if HT is enabled ...

Just my 2 cents worth ...

mondo

 

[rise]

looks like the same thing as primes or occt or s&m. i don't usually mention memtest as i know memtest stable and windows stable are two different things. i wouldn't be stability testing in windows if memtest didn't pass. heck, i don't even install windows if memtest doesn't go a few loops at stock.

btw, does that burn in have xp support? only lists 2000.