Welcome to MilkyWay@home

Posts by Gecko

1) Message boards : Application Code Discussion : @ Cluster Physik: Re FFT libs for ATI (Message 22594)
Posted 18 May 2009 by Gecko
Post:
Hiya Cluster Physik,

Do you know of any any FFT libs compatible w/ ATI GPU?

Raistmer @ Lunatics could use some input from you on this.

Thx!!!

As far as I know there is nothing freely available. If one don't want to write that on his own, one would have to ask ATI for it. Maybe they can provide him with something. Sorry.


Thanks for the reply!

One should really ask AMD about it. I've seen some FFT performance comparisons from them. They claimed a HD4870 to be 75% faster than a 9800GTX and about 15-20% faster than a GTX280. I wonder what code they have used for that comparison.


Hummm...this is VERY interesting. I'll pass it on to Raistmer. Thanks again : > )
2) Message boards : Application Code Discussion : @ Cluster Physik: Re FFT libs for ATI (Message 22576)
Posted 17 May 2009 by Gecko
Post:
Hiya Cluster Physik,

Do you know of any any FFT libs compatible w/ ATI GPU?

Raistmer @ Lunatics could use some input from you on this.

Thx!!!

As far as I know there is nothing freely available. If one don't want to write that on his own, one would have to ask ATI for it. Maybe they can provide him with something. Sorry.


Thanks for the reply!
3) Message boards : Application Code Discussion : @ Cluster Physik: Re FFT libs for ATI (Message 22415)
Posted 16 May 2009 by Gecko
Post:
Hiya Cluster Physik,

Do you know of any any FFT libs compatible w/ ATI GPU?

Raistmer @ Lunatics could use some input from you on this.

Thx!!!
4) Message boards : Number crunching : ati 4770 (Message 20983)
Posted 30 Apr 2009 by Gecko
Post:
Asus 4770 TOP looks to be insanely fast w/ on board voltage tweak and smartdoctor performance control application for voltage mod to 1.2V.
Stock is 800 MHz (core), and 850 MHz (memory). Using the above, voltages can be upped from 0.95 V to 1.2 V w/ freqencies as high as 971 MHz, and 1150 MHz for the memory. That's 1.24 Tflps for @ $100 IF a person can find one.
Essentially same performance as reference 4870, but w/ a single 6-pin power connector.

Short enough to fit on the new mini-itx board and mini-itx case from Silverstone ; > ) I'm getting the itch to grab one when I can.
5) Message boards : Application Code Discussion : milkyway released under GPLv3 (Message 14912)
Posted 11 Mar 2009 by Gecko
Post:
I've seen these type of GPL and license discussions multiple times (as have you Paul) with other projects.
It's clear that GPL isn't very well understood by many projects that are based on it & there are differences in understanding between devs. & even less understanding w/ casual crunchers.
I think a general "assumption" in many cases is that when source is released by admin of project that "is" or intends to be "GPL", the gates are open. So, overall understanding of GPL re: to DC projects is generally lacking across the spectrum.

That said, GPL requirements may not actually align very well with the typical kind of volunteer development efforts we're used to seeing here or other DC projects at all.
A dev. has responsibility for the results of his released binaries to the project & also to the users.
Unlikely that a single dev. has all the multiple platforms (Intel, AMD, & all CPU flavors w/ various SIMD limits) to test & verify proper function of each. So, in strictest interpretation of GPL, he can only keep source private based on what he can test himself. As soon as he gives a binary to someone else to test on a rig he doesn't have, he's obligated to make the source available "upon request".

Over the years, I've seen well tested optimized binaries that were thought to be suitable for public, have to be recalled or updated when unknown issues surfaced after running on enough hosts to produce & gain visibility to them. Have also seen several "alpha" binaries that did not perform to expectation or provide anticipated results, even when a simple SIMD compile variant of the same source. Projects wouldn't want untested or bad opt. binaries released & devs wouldn't want the responsibility of bad binaries or bad source being in the wild. So, GPL doesn't seem to align well w/ volunteer optimizing on the scale necessary for multi processors and platforms to support a community.

However, it seems like a vol dev. could certainly modify source, test on his own platforms and then provide source BACK to project ADMIN for project testing/incorporation back into project source & possible distribution as a Beta or whatever. This is what Travis appears to want to happen anyway ; > )

Of course we all know there's also another approach akin to the saying "If a tree falls in the woods & no one hears it....." (Not endorsing this option, just acknowledging reality)

It's clear however that public involvement in alpha/beta development is a slippery & risky slope w/ GPL requirements and the current dev & public interests of MW (or any other open source DC community). A slope that isn't very "motivating" towards encouraging volunteer efforts unfortunately.
6) Message boards : Application Code Discussion : milkyway released under GPLv3 (Message 14858)
Posted 11 Mar 2009 by Gecko
Post:
As mentioned earlier by another poster, there's a certain amount of quid pro quo going on anyway.
Not a coincidence that stock app is orders of magnitude faster than a few months ago ; >)
Also, source was previously released so that another volunteer dev. could do a OSX port that the project lacked software & licenses.
7) Message boards : Application Code Discussion : milkyway released under GPLv3 (Message 14746)
Posted 10 Mar 2009 by Gecko
Post:
This is what happens when alpha test applications are tested via public community vs. private.
Almost always, someone wants the source & questions/challenges come-up re: GPL.
With developmental alphas, there's risk to distributing code that may not be 100% sorted that could end-up in mass public distributed binaries compiled by someone else.
A developer's interest & motivation for controlling the source of a alpha dev. app is valid.
Question is whether it's legal.

Legal considerations aside, it seems poor form to request alpha dev. source from a developer while they are actively developing the application. Comes across as rather rude ; > )

Hat's off to you Cluster for your amazing work and accomplishment!

Unfortunately, as we can see, no good deed goes unpunished :>(
8) Message boards : Application Code Discussion : GPU app teaser (Message 11549)
Posted 19 Feb 2009 by Gecko
Post:
Does CPU speed matter in re: to GPU application? or does GPU app not have any significant dependency on CPU? Would one expect a major difference running Pent D or AMD X2 for example vs. a Core2 CPU?

As not the whole computation is done on the GPU (just 99.5% or so) it may have a small effect. But I would say it is less than one second difference per WU between a P4 and a Core2 for instance.
Quite a bit of this CPU computation time can be hidden when running 2 WUs (or more) concurrently. When one WU waits some milliseconds for the GPU to complete the current task, another WU can calculate its CPU part.
That's actually the reason I've integrated the possibility to run several WUs concurrently. It slightly increases the efficency. For slower CPUs it just pays off more.


Thanks for answering my question. Excellent! Also...this is good news as I have a PentD 945 & ECS C19A MB that are brand new, but it never seemed worth building after Core2 came out.....until now.
Just need to add an ATI 4850, load-up xp64 Professional & it will have a legitimate new purpose.
Cool!

Edit: Last question. How many WUs can a card run w/o any performance degredation w/ 512MB GPU memory? I understand throughput remains the same untimately, but I'd be inclined to run a greater # of WUs vs. individually if there's not a performance penalty.
9) Message boards : Application Code Discussion : GPU app teaser (Message 11510)
Posted 18 Feb 2009 by Gecko
Post:
@ Cluster Physik:

Does CPU speed matter in re: to GPU application? or does GPU app not have any significant dependency on CPU? Would one expect a major difference running Pent D or AMD X2 for example vs. a Core2 CPU?

Thx.

10) Message boards : Number crunching : 3rd.in - optimized apps (Message 10900)
Posted 15 Feb 2009 by Gecko
Post:
It looks like Alex Kan was using the Intel compiler to do the Intel builds of his V8 app on SETI.

I don't know what he used for the PPC builds though.


Correct, Alex used(s) OSX version of ICC.
PPC was built via GCC.
FYI, pretty sure V8 PPC is the "last" of the line.

Returning to topic : > )
11) Message boards : Number crunching : 3rd.in - optimized apps (Message 10720)
Posted 14 Feb 2009 by Gecko
Post:
Not that many people are running the stock app :P


Sure there are....they just don't frequent these forums like the rest of us psychos ;>)
12) Message boards : Number crunching : No further Support for Milkyway at this time! (Message 6260)
Posted 18 Nov 2008 by Gecko
Post:

Well, to be honest I'm wondering now why the project admins made the project OpenSource, as far as I see there is only a big mess and it seems they are not capable of handling it right.
Don't understand me wrong, I'm happy that the science is better now because some "outsiders" could improve the code, but in my eyes it's useless to make a project OpenSource when you can't deal with the consequences.

I completely agree w/ you and eluded to same in my earlier post.
Since projects rely on the masses to crunch the data & in most cases, do not have massive budgets and large staff, it makes perfect sense to go Open Source and benefit from the skills of the volunteer community. Projects DO have overhead & a cost associated w/ data analysis.....therefore, any means of doing more work, quicker, has a direct financial impact & also impact's the viability of the project accomplishing it's goal. Having seen these types of contentious issues many times before, projects appear to underestimate the caveats associated & necessary to effectively operate as open source.

I don't think the issues are insurmountable, but at the same time I've yet to see a project have a clear "plan & process" to effectively manage it's open source aspect. The chaos here and @ other projects that have faced similar is to be expected if a policy is not clearly defined and followed.
13) Message boards : Number crunching : No further Support for Milkyway at this time! (Message 6248)
Posted 17 Nov 2008 by Gecko
Post:
... MW (or any DC project) had to PAY $$$ to the cruncher or issue IRS tax deduction slips for charitable contributions based on the current methodology,...

Doesn't apply to the just under 78% of us who do not live in The United States and still calculate for MilkyWay.

...the project doesn't pay the volunteers...

Rather defeats the definition of being a volunteer, doesn't it? If one is paid, then one is an employee.

I volunteer and have neither an expectation of being paid, nor any desire to be paid.

-ChinookFöhn




Of course : > )
My point is that the "valuation" of work & obligations would be different if they were required to compensate. Why should one's work be valued less just because there's no direct cost involved to a party? It might be "human nature", but it doesn't mean it's right ; > )
14) Message boards : Number crunching : No further Support for Milkyway at this time! (Message 6246)
Posted 17 Nov 2008 by Gecko
Post:
As an open source project, it's a level playing field & anyone can decide to improve the ap if they have the skills...w/ NO OBLIGATION to anyone else. If the project deems the open-source status to no longer be advantageous or counter-productive, they can always go "closed-source". This is a responsibility of the project, not the volunteer community.

The myriad of accusations, demands and generally unprofessional comments extended at developers and Admin the past few months would not ingratiate these folks to provide much of anything in this atmosphere of bitterness. Certainly, Admin is obliged to mission-critical needs, but why in the world after witnessing all of this, would any capable coder WANT to provide services after seeing the responses? They'd be better off building their uber-ap, staying anonymous & NOT sharing w/ the project, much to the loss of the project & not in the true spirt of open-source design.

If abused, project forums can present a significant liability to the project and an open source atmosphere. Moreover, if these discussion forums present too much consternation that detracts from project resources, Admin could always decide to do away w/ them and relegate communication to info sharing re: progress and issues reporting. All other discussion could easily be relegated to individual team forums. It's their project after all. Hopefully, this will not happen & cooler heads will prevail w/ sensible solutions to current issues.

Just some things to think about before drawing lines in the sand......
15) Message boards : Number crunching : No further Support for Milkyway at this time! (Message 6243)
Posted 17 Nov 2008 by Gecko
Post:
I'm curious as to the % of total MW crunchers that are active on the forums. Is the greatest % of MW crunchers and production among the "silent majority"? While this is the case w/ Seti & Folding, it may not be the case here.

It's interesting to look at the dichotomy that emerges if one introduces economic considerations. No question that the current credit award methodology is broken for multiple reasons that have already been discussed. Fact is, if MW (or any DC project) had to PAY $$$ to the cruncher or issue IRS tax deduction slips for charitable contributions based on the current methodology, they would not want to do so as they'd end up dramatically overpaying for the least of amount of production resulting from slower machines. This equation in reality, is reversed of course; the project doesn't pay the volunteers...the volunteers make an investment into the project via. electricity, wear & tear, depreciation, opportunity-cost (cost vs. devoting those cpu cycles elsewhere) & in some cases, software (volunteer developers/optimizers).

Economically speaking, there should be an equilibrium of mutual benefit in the relationship between project & volunteer. The project would be inclined to balance the greatest return from it's "volunteer" resources w/ the best rewards (& least expense to volunteer) available to guarantee continuance. Pretty simple...greater production = greater value to project & compensated by greater reward to the worker. Unfortunately, this is not the case at the moment. Interestingly, in the attempt to appease the largest number of mid to upper-mid producers, the current system not only penalizes the most efficient worker using the opt aps, but also penalizes everyone using standard aps & therefore hits the lowest producers even harder.

What a great micro-example of labor economics that mirrors a similar debate in the recent US election. Fascinating!!! A person could actually use a DC computing project as an economic model for a thesis or dissertation research. Hummmm.......might just DO that myself ; > )
16) Message boards : Number crunching : Increased WU Credit (Message 2801)
Posted 25 Mar 2008 by Gecko
Post:
Reasonable, balanced & fair IMO.
Not the highest, not the lowest. Falls into a good range.

More importantly, I like the collaboration and communication between Project staff and members.
This project has GREAT potential!

Keep up the wicked-cool work!

Cheers!
17) Message boards : Number crunching : updated granted credit (Message 2653)
Posted 24 Mar 2008 by Gecko
Post:

I like your argument. I'm going to show it to Travis. He's the one who controls the credit and the main aspects of the project. I'd like to say though that I really don't care about credit. I'd set it to 1,000 per work unit if it'd keep the user base together. This is a great community of people and i'm going to try hard to keep it that way.


Thanks Dave! I understand & appreciate your comments above. Group unity and the credit subject is a challenging tightrope to walk.
Keep up the great work!

18) Message boards : Number crunching : updated granted credit (Message 2650)
Posted 24 Mar 2008 by Gecko
Post:
I yelped a bit a couple of weeks ago when it was mooted that credits should be reduced. That is because I am running a PPC Mac and it didn't get any optimisation in the last round of app updates. However, I subsequently looked at the figures again and I concede that 6.5 per result was too high. Trouble is, 2.17 is way too low. Somewhere around 4 would be reasonable I would think.



I think that 4.2 is a reasonable number.


I agree that 4.2 is reasonable and acceptable if the more likely alternative is the current 2.17. Admittingly, as most agree in all fairness, 6.5 was on the high side to be a stock ap. Interestingly, I run 32 bit XP on a T7200 laptop & MW was good for 100 credits per hour. Also run current Einstein "power users Ap" which grants 78.8 per hour. It's from the project, not 3rd party, so it's what "stock" is awarding.

I agree w/ Crunch3r and Philly that Seti is not a good reference & their methodology is flawed & broken. One has to have a strong motivation to crunch it, given the numerous deficiencies. It's in pretty sad shape actually and a shadow of its former self. But I digress.....

4.2 will drop credit to @ 65 per hour on my rig, a fair amount. Not the highest, not the lowest...but "acceptable". 2.17 means 33 per hour on this rig...not very motivating. That said, I crunch MW, Einstein etc. because I like the projects and they have awesome scientific relevance. To me, good credits are enough appreciation for time, electricity and the shortened lifespan of components running 100% load 24/7. Since 99', the sum of my personal investment in distributed computing considering these things & donations is surely a few thousand $$$ by now.

Also,there should be some consideration (for any project that is open-source or welcomes development from volunteers) that dedicated & capable volunteer programmers/optimizers are rather rare. The difference that they can make in productivity can be quite substantial and the potential financial impact to the project from time & hardware savings etc....HUGE. I'd think many projects would like to keep such an individual(s) on "their team" as a franchise player ; > ) & not encourage free-agency. For the extra work and time they dedicate, there's got to be some sort of incentive to improve the ap performance if personal interest is not the prime motivator. The project might not ask this of a person, but still, are they any less valuable?
A little higher credit base is a rather inexpensive, but effective reward (in some cases) for their improvements.

Just offering my opinion.

Gecko
19) Message boards : Number crunching : More Work !!! Please :) (Message 2145)
Posted 10 Mar 2008 by Gecko
Post:
Would like to dedicate a box to this project, but even w/ c2d, drains cache quickly then loads-up w/ another project to 100%. A little frustrating trying to keep MW as the higher allocated project when cache runs dry and stuffs full w/ another project. Granted, I've got cache limited to 1 day to minimize this inconvenience somewhat (which is a gamble in itself that the other project doesn't go off line longer than 24 hours as well).

I'll echo "more work, or longer WUs please".

I understand these things take time and everyone is doing their best.

Hat's-off to the fantastic application work to date by project developers and Crunch3r!

I'm excited to to be a part of Milky Way!

Cheers,
Gecko




©2019 Astroinformatics Group