Seti@Home optimized science apps and information
Optimized Seti@Home apps => Windows => GPU crunching => Topic started by: Skywalker66_Bln on 25 Feb 2011, 10:31:57 am
-
When comes a never better version fpr fermi ???
last: Lunatics_x32f_win32_cuda30_preview.exe
from installer 0.,37
i see on Vyper`s blog it gives any more beta`s
Can i help ????
-
Can i help ????
Nope, Sorry. The community was so utterly helpful in accelerating development that I managed to refine the entire first 60% of Multibeam Code 'out of branch', within the Powerspectrum Unit tests. (in the timeframe of a couple of weeks)
As you can imagine, incorporating an entire rewrite of 60% of the Cuda application from those developments is still going to take me some time to accommodate, and I'm fairly thinly spread at the moment. It will take some time & dedication to incorporate properly.
Plan A is to let the Ati stuff catch up a bit first, so we can have some pressure on us before optimising the pulsefinding code, that is the real challenge (remaining ~40%). Plan B, should the Ati stuff catch us earlier than expected, is under my hat :-X
Jason
[PS:] Plan C is that nVidia ships me a truckload of money, much of which would be donated to SaH, so we can stay in front of these AMD clowns. This Plan Is GOOOD IMO, and would preempt plans A & B
-
You mean it takes more than 5 minutes to throw it together and send it out the door? As to the ATI guys, they are already nipping at our heels. Hmmm, wonder where I can find a truckload of three dollar bills I can pick up cheap? Oh wait, you want real maney!! Oh well! ;D
-
You mean it takes more than 5 minutes to throw it together and send it out the door? As to the ATI guys, they are already nipping at our heels. Hmmm, wonder where I can find a truckload of three dollar bills I can pick up cheap? Oh wait, you want real maney!! Oh well! ;D
LoL, yep, that pretty much explains it ;) Watch the top hosts pages while we struggle to get an installer update with the OpenCL apps going (that happens to be highest on my own priority list at this moment)... You could see some Radeon only hosts get intp the top 100, then it'd be time to Panic ;D
[Mod:] editing thread title, by request, due to page stretching.
-
You mean it takes more than 5 minutes to throw it together and send it out the door? As to the ATI guys, they are already nipping at our heels. Hmmm, wonder where I can find a truckload of three dollar bills I can pick up cheap? Oh wait, you want real maney!! Oh well! ;D
LoL, yep, that pretty much explains it ;) Watch the top hosts pages while we struggle to get an installer update with the OpenCL apps going (that happens to be highest on my own priority list at this moment)... You could see some Radeon only hosts get intp the top 100, then it'd be time to Panic ;D
[Mod:] editing thread title, by request, due to page stretching.
Thanks, i was about to do the same,
Edit: Cthulhu is already at 77 with his Radeon only host!
Claggy
-
Edit: Cthulhu is already at 77 with his Radeon only host!
Oh No! PANIC! can't have that for too long :P
-
With the server problems and working on Aqua and Collatz, I slipped down to 483.
I was as high as 380 or so at one time.
-
More seriously I'd like to see a lot more Ati Cards getting up there soon. Not only would it place more development challenge on the table (always a bonus), but the bang for buck, in terms of both initial purchase & running costs, represented by the newer AMD cards looks pretty good IMO. It should begin to put at least some small amount of downward price pressure on nVidia products, as the OpenCL tools mature, and pressure for higher performance too.
I happen to prefer both the nVidia hardware & Cuda development tools etc, over OpenCL, mostly due to the level of maturity overall, and nVidia having had the wherewithall to front up the initial application development as a head start. As the OpenCL apps get more polished, the Cuda app will be entering its second generation, and We'll be expecting to push development into SaH V7. A mature OpenCL platform IMO can only help everyone involved.
Jason
-
My only problem with the OpenCL app right now is the skewing of the DCF I was seeing on my ATI machine. It might have been a combination of factors of using the rescheduler on all those vlars that were going -177.
Right now my machine has 435 units on board and it things it will take almost 135 days to complete them. DCF was at 0.7 last I checked. Currently on NNT.
-
Can i help ????
Nope, Sorry. The community was so utterly helpful in accelerating development that I managed to refine the entire first 60% of Multibeam Code 'out of branch', within the Powerspectrum Unit tests. (in the timeframe of a couple of weeks)
As you can imagine, incorporating an entire rewrite of 60% of the Cuda application from those developments is still going to take me some time to accommodate, and I'm fairly thinly spread at the moment. It will take some time & dedication to incorporate properly.
Plan A is to let the Ati stuff catch up a bit first, so we can have some pressure on us before optimising the pulsefinding code, that is the real challenge (remaining ~40%). Plan B, should the Ati stuff catch us earlier than expected, is under my hat :-X
Jason
[PS:] Plan C is that nVidia ships me a truckload of money, much of which would be donated to SaH, so we can stay in front of these AMD clowns. This Plan Is GOOOD IMO, and would preempt plans A & B
Uhm, sounds not promising and like a neverending Story to me. Well, i will keep my GTX465/GTS450 and maybe
GTX590 until - 2013 to test this Cuda Client. ;)
Helli
-
Uhm, sounds not promising and like a neverending Story to me. Well, i will keep my GTX465/GTS450 and maybe
GTX590 until - 2013 to test this Cuda Client. ;)
It's smart arsed comments like that that make we wonder why I bother, and feel less motivated to get the application rewritten. Those users that kindly contributed to the PowerSpectrum Unit tests thread have some inkling of what's coming, and how much work is involved. [Self moderated comment]. I don't recall your participation in that development.
-
Well, i have made my part of supporting the development of a optimized Fermi Client. Remember? June 2010? GTX480?
Maybe it's the lack of information we get of the current status of the development that made me sarcastic.
Not everybody reads every day here at Lunatics. I for myself been mostly in Number Crunching at Seti.
Helli
-
Well, i have made my part of supporting the development of a optimized Fermi Client. Remember? June 2010? GTX480?
Maybe it's the lack of information we get of the current status of the development that made me sarcastic.
Not everybody reads every day here at Lunatics. I for myself been mostly in Number Crunching at Seti.
Helli
Right, so you contributed toward the Fermi Card ? By all means contact SciManSteve and I will arrange for return of your funds, and any others that wish it should do the same.
Putting up with people coming here behaving like arseholes was not part of any promise I made.
-
Well, i have made my part of supporting the development of a optimized Fermi Client. Remember? June 2010? GTX480?
Maybe it's the lack of information we get of the current status of the development that made me sarcastic.
Not everybody reads every day here at Lunatics. I for myself been mostly in Number Crunching at Seti.
Helli
Right, so you contributed toward the Fermi Card ? By all means contact SciManSteve and I will arrange for return of your funds, and any others that wish it should do the same.
Putting up with people coming here behaving like arseholes was not part of any promise I made.
Uhm...i forgot. You are a Moderator here. Better i say nothing more - here.
Helli
-
Uhm, sounds not promising and like a neverending Story to me. Well, i will keep my GTX465/GTS450 and maybe
GTX590 until - 2013 to test this Cuda Client. ;)
It's smart arsed comments like that that make we wonder why I bother, and feel less motivated to get the application rewritten. Those users that kindly contributed to the PowerSpectrum Unit tests thread have some inkling of what's coming, and how much work is involved. [Self moderated comment]. I don't recall your participation in that development.
Ouch. It must be one of those days...
Guys, can we at least get 'lost in translation' out of the way? Do I have to set up a translation service?!
Helli - ich bin mir ziemlich sicher, Du hast das nicht so verletzend gemeint, wie das auf Englisch ruebergekommen ist.
-
ARGH
would you two please let me get a word inbetween before you start a flame war ?!
-
Was meinst du mit verletzend? Das bissel Sarkasmus?
Ich sollte wohl besser bei technischen Dingen bleiben wenn ich in englisch schreibe....
Helli
-
Thank you.
Right. First off Jason, Helli couldn't have contributed to the powerpectra - we did that just before Christmas iirc and he joined just a month ago.
Second I noticed that sarcasm works very badly across the german - english divide.
Seems like it always gets takes offensively in both directions. It requires both a high level of intelligence (tick ;) ) and very good language capabilities. Both to phrase properly and to get the sarcasm.
So, have you both calmed down now? Your babelfish surrogate is at your service ;)
Carola
-
Ok, the last Thing i will say here.
If i hurt someone, especially Jason, it was not my intention. I have thought
that the Smilie says: "I don't mean it seriously".
Sorry again.
Zu deutsch, falls mein englisch nicht das wiedergibt was ich meine:
Falls ich jemanden verletzt habe mit meinem Beitrag, speziell Jason, tut es mir Leid, war nicht meine Absicht.
Ich dachte das zwinkernde Smilie zeigt das es nicht Ernst gemeint war.
Helli
-
Thank you.
Right. First off Jason, Helli couldn't have contributed to the powerpectra - we did that just before Christmas iirc and he joined just a month ago.
Second I noticed that sarcasm works very badly across the german - english divide.
Seems like it always gets takes offensively in both directions. It requires both a high level of intelligence (tick ;) ) and very good language capabilities. Both to phrase properly and to get the sarcasm.
So, have you both calmed down now? Your babelfish surrogate is at your service ;)
Carola
Right, Thanks Carola.
@Helli: I apologise for taking your comments as barbed attacks. I should know by now that some things don't translate well.
I suggest taking the time to run the PowerSpectrum Unit test #10, and maybe a read of some of the 21 pages of development contained in that thread, and hopefully you might feel a bit better having some numbers to indicate what is coming.
The public test is at: http://lunatics.kwsn.net/12-gpu-crunching/split-powerspectrum-unit-test.msg33244.html#msg33244
That test is closed now, A necessary step toward getting something into the app, but feel free to see how it performs on your hardware. The test piece accurately represents 40%<->60% of multibeam processing depending on angle range, along with a similar proportion of the codebase that needs rewriting.
I should add that the pressing issue mentioned at the end, related to detecting numerical disagreement between CPU<->GPU results with all current appplications. As a result of those developments, refinements have already gone into the SaH_v7 codebase to take care of that, it turned out CPU application Results were sometimes 'noisy', and GPU apps in the clear.
Jason
-
Was meinst du mit verletzend? Das bissel Sarkasmus?
Ich sollte wohl besser bei technischen Dingen bleiben wenn ich in englisch schreibe....
Helli
Jep - das 'bisserl Sarkasmus' ist grad mal bei mir als vermutlich solcher angekommen - wie Du an Jasons Reaktion gemerkt haben solltest, war das im Englischen recht verletzend.
Nichts gegen Dein Englisch, da muss ich wenigstens nicht staendig ins Deutsche rueckuebersetzen, um draus schlau zu werden ::)
Nur von Sarkasmus wuerd ich in Zukunft vielleicht Abstand nehmen ;) Und mir ist das auch schon mehrfach so gegangen, dass ich englischen Sarkasmus als verletzend empfunden habe - obwohl ein Smile dran war und es von einem engen Freund kam, wo man eigentlich erstmal perse davon ausgehen sollte, das es nicht verletzend gemeint sein kann.
Right, let's get that into some nice english..
Helli: hurtful? that touch of sarcasm? I should stick to technical when writing in english.
yes, I barely got that 'touch of sarcasm' as in all likelyhood being such - as Jason's reaction will have shown you, it was quite hurtful in englsih.
Nothing against your english (my that babelfish is really getting tired) at least I'm not forced to backtranslate to german to make head or tail of it.
I would suggest not to use sarcasm in the future. Speaking for myself, there have been various occasions, where the sarcasm of an english native speaker hurt me - even with a smily and coming from a close friend where the basic assumption should always be, that it just can not be meant hurtful.
You're welcome Jason. Sorry I didn't catch it sooner.
-
@ helli...........................
Better you should say nothing more here or anywhere else. At least not till you remove your foot from your mouth.
-
@ helli...........................
Better you should say nothing more here or anywhere else. At least not till you remove your foot from your mouth.
It's OK Geek, I think the misunderstanding has been sorted out now.
But what Helli, and anyone else who hasn't tried writing truly optimised code for themselves, could usefully do is read Jason's post http://setiathome.berkeley.edu/forum_thread.php?id=63033&nowrap=true#1080372 at the main project. You might be surprised at some of the points, like the "plenty of code around here that could do with a few thousand man hours of refinement as homework :D"
And that, I would judge, is several thousand hours of experienced, postgraduate refinement.
-
And that, I would judge, is several thousand hours of experienced, postgraduate refinement.
ROFL, such level will require more hours cause more options will be available ;D ;D ;D
[
And in general, I would like to cite cat Leopold from one of our cartoons: "Ребята, давайте жить дружно" :)
http://www.youtube.com/watch?v=K2KfNSmBt60
]
-
"Children, give to live harmoniously"
-
) _ _ _)_ o _ _
(__ (_( ) ) (_( (_ ( (_ (
not bad for a human... _)
Multibeam x36c Preview, Cuda 3.2
Is there any way to get this build? Faster or not compared to build from 0.37 instaler?
Thanks for answering..
-
Hi Pepi,
It is a bit faster, though that's a prototype build tuned fairly specifically for certain hardware and has been in alpha testing on selected hosts for data gathering purposes. It doesn't work well on general cards, but upper 2xx series & higher end Fermis it's very good, especially on hosts with lots of high end cards, due to reduced CPU & Bus contention.
There is an even newer x36e build about to go to beta that is more suited for general cards, is also faster than x32f, and is 'V7-ready'.
At the moment I'm deciding how best to put x36e (the one suitable for all cuda cards) to wider beta, without causing too much disruption, as the V7 code is unrefined and has a heavy CPU usage penalty (as with Raistmer's test V7 builds). That doesn't appear to adversely effect legacy V6 processing on main though, and early V7 readiness would be a bonus.
Hang tight & x36e will be in the wild soon enough. We've had to spend more time on V7 migration instead of polishing, which just means having to be careful with a few things.
Jason
[Later:] I've dropped PMs to pepi & Ghost, so that it gets a slightly wider online test before I get time to set up a proper beta.
-
Let me know if my little GTS-450 could be of help. I'm not a beta tester on SETI * but I would love to see what it could do with the more refined build.
* I'm afraid I would be more of a hindrance than a help over there. I need too much hand holding and barely understand what my tests mean.
-
Jason, if any help, I am also interested in this new x36e beta - currently using and GTX560Ti and i7-2600K...
-
Jason, if any help, I am also interested in this new x36e beta - currently using and GTX560Ti and i7-2600K...
Let me know if my little GTS-450 could be of help. I'm not a beta tester on SETI * but I would love to see what it could do with the more refined build.
* I'm afraid I would be more of a hindrance than a help over there. I need too much hand holding and barely understand what my tests mean.
Thanks both, will be putting to wider beta as soon as I get the time. A bit tied up for now ;)