ATI AstroPulse beta test started on Lunatics

Message boards : Number crunching : ATI AstroPulse beta test started on Lunatics
Message board moderation

To post messages, you must log in.

Previous · 1 · 2

AuthorMessage
Luke
Volunteer developer
Avatar

Send message
Joined: 31 Dec 06
Posts: 2546
Credit: 817,560
RAC: 0
New Zealand
Message 995542 - Posted: 11 May 2010, 7:25:44 UTC - in response to Message 995540.  
Last modified: 11 May 2010, 7:27:37 UTC

Personally, I will be scared the day ATI support arrives mainstream at SETI@home. I appreciate all Lunatics does - and it does it beautifully! It's just it will be abused by credit whores.

That is the day SETI@home will turn from a science project into a credit hound project.

That is the day I leave BOINC, if it ever happens. Just my two cents. :(


Very strange position, I already pointed that at your post about person who owns many hosts. Try to understand, not everyone who attempt to do more work for SETI and granted big RAC due [s]his efforts do it for credits. Big RAC if no cheating involved really means big science done. What point to leave project then?....
GPU is just another device for doing work. If you strong in math, no one forbid you to do WUs manually or with calculator, your RAC would be in fempto-cobblestones probably, but it will be _very_ hard earned one. Do you really think that task computed in so inefficient way will be any better than task produced by GPU owned "credit whore"? Don't think so ;)


Admittedly, it is a strange position. I just care for the project... and with ATI available to the general public, the credit hounds from MW will arrive, with their huge appetite for work needing to be satisfied. It'll put more strain on the (already strained) servers, leading to more outages and less work available for the 90% of others. I'm talking RACs in the TENS of MILLIONS, not in the hundreds of thousands as it is currently. And the the tide will change from a science oriented project to that of a credit oriented project, for others to abuse.

It hasn't happened so much with CUDA, since it isn't (as of now, could be soon though) as powerful as some of the more top of the line Radeon GPU's. MW is relatively fine with the amount of WU's being sent out, SETI is a little more dependent on other things, such as whether the ALFA reciever is on, whether work can be sent back to Berkeley, and all assuming Arecibo is not closed!

LOL. I won't be seen doing SETI@home WU's with a calculator thanks... It'd take years.
- Luke.
ID: 995542 · Report as offensive
Profile Paul D. Buck
Volunteer tester

Send message
Joined: 19 Jul 00
Posts: 3898
Credit: 1,158,042
RAC: 0
United States
Message 995544 - Posted: 11 May 2010, 7:38:36 UTC - in response to Message 995542.  

Admittedly, it is a strange position. I just care for the project... and with ATI available to the general public, the credit hounds from MW will arrive, with their huge appetite for work needing to be satisfied. It'll put more strain on the (already strained) servers, leading to more outages and less work available for the 90% of others. I'm talking RACs in the TENS of MILLIONS, not in the hundreds of thousands as it is currently. And the the tide will change from a science oriented project to that of a credit oriented project, for others to abuse.

Their might be some shift... but it will not be by the credit hounds ... the pay per CPU second is far better at MW than it is here... and the efficiency of the SaH application using the GPU is far lower here than there, or Collatz or even DNETC.

If there is an influx, it will be by those that have an ATI card and are interested in SaH...

As an example, I am almost all over to ATI cards now and yet I have not bothered to try the AP application for the simple reason that it uses too much CPU and not enough GPU. MW and the rest are very low CPU usage and very high GPU usage (in the high 90s) so you get far more for your time in those other projects.

The other piece of good news from your perspective is that GPU Grid announced recently that they have made a breakthrough in their ATI application and may be close to a beta soon on it ... though they are still way inefficient with ATI cards being half the speed of CUDA cards instead of the "normal" other way around.

So, I would not get too concerned...
ID: 995544 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 995545 - Posted: 11 May 2010, 7:48:43 UTC - in response to Message 995542.  


Admittedly, it is a strange position. I just care for the project... and with ATI available to the general public, the credit hounds from MW will arrive, with their huge appetite for work needing to be satisfied. It'll put more strain on the (already strained) servers, leading to more outages and less work available for the 90% of others. I'm talking RACs in the TENS of MILLIONS, not in the hundreds of thousands as it is currently. And the the tide will change from a science oriented project to that of a credit oriented project, for others to abuse.

I will not go in thechnical details right now, need to go on work.
But to give you relieve from your fears: currently implemented AP algorithm is memory bound. ATI memory subsystem not very powerful one re NV's or CPU. So don't expect so high RAC as MW will pay. Actually, ATI on SETI main will be true holy grail for "science oriented" participants. Cause MW always will pay better, no sense to do SETI for "credit whores" then :D

ID: 995545 · Report as offensive
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19012
Credit: 40,757,560
RAC: 67
United Kingdom
Message 995549 - Posted: 11 May 2010, 8:14:30 UTC - in response to Message 995544.  
Last modified: 11 May 2010, 8:14:50 UTC

Personally from what I have read it is not that the gpu app is fantastically efficient on the other high paying projects. It is that their cpu apps are very inefficient, and credits are calculated from the cpu app, with old time * BM calculation.

So if task credits = 'very long time' * 'bench mark' and then you add app where the task get the same credits and time is very, very short. Then it starts to look as though it is a very, very efficient high paying app. Which is probably not true, in is probably only a run of the mill fairly efficient app. That most programmers could do using the correct compiler.
ID: 995549 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 995743 - Posted: 12 May 2010, 6:53:34 UTC - in response to Message 995544.  



As an example, I am almost all over to ATI cards now and yet I have not bothered to try the AP application for the simple reason that it uses too much CPU and not enough GPU. MW and the rest are very low CPU usage and very high GPU usage (in the high 90s) so you get far more for your time in those other projects.

Then it's time to try ;) GPU load near 97% most of time.
ID: 995743 · Report as offensive
Profile Paul D. Buck
Volunteer tester

Send message
Joined: 19 Jul 00
Posts: 3898
Credit: 1,158,042
RAC: 0
United States
Message 995811 - Posted: 12 May 2010, 15:50:18 UTC - in response to Message 995743.  



As an example, I am almost all over to ATI cards now and yet I have not bothered to try the AP application for the simple reason that it uses too much CPU and not enough GPU. MW and the rest are very low CPU usage and very high GPU usage (in the high 90s) so you get far more for your time in those other projects.

Then it's time to try ;) GPU load near 97% most of time.

What is the CPU load? And which version?
ID: 995811 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 995838 - Posted: 12 May 2010, 17:20:06 UTC - in response to Message 995811.  

On 5770 for 22k seconds of elapsed time were ~500 CPU seconds used.
On HD 4870 result slightly worse. For ~same elapsed time ~1000 secs of CPU used.

CPU is Core2-based in both cases. Duo and Quad I suppose, small difference in freqs not very important for this estimate.
ID: 995838 · Report as offensive
Profile Paul D. Buck
Volunteer tester

Send message
Joined: 19 Jul 00
Posts: 3898
Credit: 1,158,042
RAC: 0
United States
Message 995867 - Posted: 12 May 2010, 20:45:21 UTC - in response to Message 995838.  

On 5770 for 22k seconds of elapsed time were ~500 CPU seconds used.
On HD 4870 result slightly worse. For ~same elapsed time ~1000 secs of CPU used.

CPU is Core2-based in both cases. Duo and Quad I suppose, small difference in freqs not very important for this estimate.

That is not that bad ...

Though my recollection is that the beta is semi-closed ...

At any rate I have a huge amount of SaH D/L for the one system I am running SaH on though the CUDA card is working pretty hard to catch up on the work that BOINC went wild and obtained... another bug that UCB keeps telling me that it is all in my imagination ...

Oh, my, I just got to seem my computer crash!
ID: 995867 · Report as offensive
Profile Fred J. Verster
Volunteer tester
Avatar

Send message
Joined: 21 Apr 04
Posts: 3252
Credit: 31,903,643
RAC: 0
Netherlands
Message 996136 - Posted: 14 May 2010, 10:36:08 UTC - in response to Message 995867.  

Opened an Account in SETI Bêta, tried 2 app.'s, r280 &r393, but made an error in app_info.xml and gone where my 4 AP WU's.
Now I have them running at MW, both 5770 and 4850, or there are 2 running at the EAH4850, since that is a Double_Prrecision, HD5770 is not.
Also DL'ed the ATI SDK 2.1, but don't know which app. should be tested?


ID: 996136 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 996139 - Posted: 14 May 2010, 10:53:37 UTC - in response to Message 996136.  

I replied to you on SETI beta forum.
In short, with your GPUs you could try rev420 build.
Can be downloaded here:
http://lunatics.kwsn.net/18-astropulse-testing/ati-sse3-astropulse-app-openclbrook-beta-testing.0.html
ID: 996139 · Report as offensive
Profile Fred J. Verster
Volunteer tester
Avatar

Send message
Joined: 21 Apr 04
Posts: 3252
Credit: 31,903,643
RAC: 0
Netherlands
Message 996146 - Posted: 14 May 2010, 11:35:59 UTC - in response to Message 996139.  
Last modified: 14 May 2010, 11:40:23 UTC

My fault, I didn't notice it, in time.
Had a look at your results with rev420, which looks promising.
I'll (try) to set it up and report results at Lunatics Website.
Thanks for your reply.
ID: 996146 · Report as offensive
Previous · 1 · 2

Message boards : Number crunching : ATI AstroPulse beta test started on Lunatics


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.