Random Musings About the Value of CPUs vs CUDA

Message boards : Number crunching : Random Musings About the Value of CPUs vs CUDA
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · 4 . . . 11 · Next

AuthorMessage
Mithotar
Avatar

Send message
Joined: 11 Apr 01
Posts: 88
Credit: 66,037,385
RAC: 50
United States
Message 844044 - Posted: 23 Dec 2008, 5:52:38 UTC

Its good to see Seti able to do more workunits and get the results faster from less computers.....but I have to ask where does this leave those of us with older (non-CUDA capable) PCs.......

I for one dont need to nor want to upgrade my PCs...not worth the effort for just Seti....plus even if I wanted to the $$ arent there (for 4 PCs) to do so......

So does this leave those of us using only CPU's for crunching to the fate of a 486 PC ? Also-rans in the world of overclockers and new GPUs ??? At 1/10 the work done it hardly seems worth the price of the electricity....

I'll keep crunching away (for now) and see what comes of this situation. But from what I have been reading I'm not optimistic.
ID: 844044 · Report as offensive
Profile Vipin Palazhi
Avatar

Send message
Joined: 29 Feb 08
Posts: 286
Credit: 167,386,578
RAC: 0
India
Message 844054 - Posted: 23 Dec 2008, 6:29:37 UTC - in response to Message 844044.  

I agree to a cerain extent. The CUDA has taken the joy from crunching for seti. But it certainly doesnt mean that I am going to stop this. I volunteered for this cos I believe in the existence of extra terrestrial life. Whether we find it in our lifetime or not is a different story. Atleast we have to try.

CUDA has given the capability to compute more data in less time. As a result the servers are running dry and users, me included, are not getting any tasks to process. This has dampened the seti spirit, and I have noticed that a few of the top crunchers have either halted their efforts or left for good. I have been tempted on more than one occassions to join another project, but thats not the reason why I join distributed computing. I crunch only seti and nothing else. If this stops, then my crunching stops.

In light of recent events, I hope the administrators take some quick measures to resolve the issues. These are some of my musings:

1. The lack of work units is the biggest of them. As I suggested in another thread before, CUDA should be directed towards AP, as they are larger in size and take longer to crunch. This should keep the heat off the MB only crunchers.
2. They can increase the size of MB units and selectively send them out to CUDA machines only, so that the regular CPU based systems get the regular MB unit and dont run out of job. A separate category could be started just like AP shown in the server stats page for CUDA units.
3. I have read earlier that they have not yet started varifying the processed data being sent back. Something called NTPCKT is still being developed. Maybe that can be finished and the varification work can be based on CUDA systems. (I am not sure about this, others reading this post, please correct any mistakes)

Many have suggested that CUDA be stopped until it is fully tested and all the bugs rectfied. But then we would never be able to check the bugs in a real world environment.

Not everyone can afford a CUDA based GPU and even if they could, not everyone will be interested in the transition. I was lucky to get hold of a GTX 260 card which my friend was glad to lend me for the time he is away on holidays. Set it up on a unicore AMD system. But it hasnt been of much use since I cant get any work from the servers. Also used Raistmers' modified app_info file to process AP on CPU and MB on CUDA. Exept for the three work units that I received since yesterday morning, it has been lying idle. So switched it off to save power. Heat production is another drawback that I face. Living in a desert has its disadvantages.

Eric and Matt need to seriously and urgently weigh the pros and cons, decide the correct actions (which I hope is being done) and inform us about it.
______________


ID: 844054 · Report as offensive
Profile Dagar
Volunteer tester
Avatar

Send message
Joined: 3 Jul 02
Posts: 22
Credit: 704,294
RAC: 0
Sweden
Message 844087 - Posted: 23 Dec 2008, 8:13:08 UTC - in response to Message 844054.  

im all for that.. AP's r big cunks.. its nice to see quicker moving on ranking list, but more importand is to get maximum data calculated..

That will also somewhat solve storage problem of raw data ;)

Brgds Dagar
ID: 844087 · Report as offensive
Profile ML1
Volunteer moderator
Volunteer tester

Send message
Joined: 25 Nov 01
Posts: 10802
Credit: 7,508,002
RAC: 20
United Kingdom
Message 844123 - Posted: 23 Dec 2008, 11:27:21 UTC - in response to Message 844044.  

Its good to see Seti able to do more workunits and get the results faster from less computers.....but I have to ask where does this leave those of us with older (non-CUDA capable) PCs.......

I for one dont need to nor want to upgrade my PCs...not worth the effort for just Seti...

Time and technology move on. GPUs have been offering a vast resource just waiting to be used for a long time now. It is a fantastic development to now bring GPU crunching mainstream.

Non-CUDA (GPU) people can still add their contribution to s@h. You can also choose to contribute to other Boinc projects that are not yet CUDA enabled.

Regardless, if your PC is on, it may as well do something useful with the idle time while it's waiting for your next keystroke!

Keep searchin',
Martin

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
ID: 844123 · Report as offensive
Profile popandbob
Volunteer tester

Send message
Joined: 19 Mar 05
Posts: 551
Credit: 4,673,015
RAC: 0
Canada
Message 844334 - Posted: 23 Dec 2008, 23:46:20 UTC - in response to Message 844054.  


1. The lack of work units is the biggest of them. As I suggested in another thread before, CUDA should be directed towards AP, as they are larger in size and take longer to crunch. This should keep the heat off the MB only crunchers.
2. They can increase the size of MB units and selectively send them out to CUDA machines only, so that the regular CPU based systems get the regular MB unit and dont run out of job. A separate category could be started just like AP shown in the server stats page for CUDA units.
3. I have read earlier that they have not yet started varifying the processed data being sent back. Something called NTPCKT is still being developed. Maybe that can be finished and the varification work can be based on CUDA systems. (I am not sure about this, others reading this post, please correct any mistakes)

Many have suggested that CUDA be stopped until it is fully tested and all the bugs rectfied. But then we would never be able to check the bugs in a real world environment.


The reason for the lack of work units was because there was not enough storage. Not because of cuda.

NTPKR is not distributed work. It can only be done by SETI servers (due to amount of data to be compared)

ALL of the bugs reported could have been caught in BETA testing. They could of easily been found. The project simply did not leave them in beta testing long enough nor allowed tasks to be validated by the time cuda was here at main.

~Bob


Do you Good Search for Seti@Home? http://www.goodsearch.com/?charityid=888957
Or Good Shop? http://www.goodshop.com/?charityid=888957
ID: 844334 · Report as offensive
Josef W. Segur
Volunteer developer
Volunteer tester

Send message
Joined: 30 Oct 99
Posts: 4504
Credit: 1,414,761
RAC: 0
United States
Message 844341 - Posted: 24 Dec 2008, 0:01:22 UTC - in response to Message 844044.  

Its good to see Seti able to do more workunits and get the results faster from less computers.....but I have to ask where does this leave those of us with older (non-CUDA capable) PCs.......

I for one dont need to nor want to upgrade my PCs...not worth the effort for just Seti....plus even if I wanted to the $$ arent there (for 4 PCs) to do so......

So does this leave those of us using only CPU's for crunching to the fate of a 486 PC ? Also-rans in the world of overclockers and new GPUs ??? At 1/10 the work done it hardly seems worth the price of the electricity....

I'll keep crunching away (for now) and see what comes of this situation. But from what I have been reading I'm not optimistic.

I see no reason to change my attitude. I've never had the hardware to compete in the top host arena, but there is work to do and I'll continue having my hosts contribute.

I don't know how much of the data received from Arecibo has been put in storage at HPSS without being split, but I suspect it is about half. If the hosts running CUDA increase our crunching capability by 50% and the project can deliver the work, perhaps they'll only have to store a quarter of the incoming data unsplit.

The 10X estimate for CUDA GPU over CPU is largely pure guesswork. There's a large range of CUDA capable GPUs and an even larger range of CPUs. The BOINC core client has a rough estimate of CPU capability on a host from its benchmarks, and apparently there's a rough idea of GPU capability from published data. If they can mesh that data well enough to achieve something approximating heterogeneous computing (most efficient use of available resources), it will be good for both this project and distributed computing in general.
                                                                 Joe
ID: 844341 · Report as offensive
Profile Francois Piednoel
Avatar

Send message
Joined: 14 Jun 00
Posts: 898
Credit: 5,969,361
RAC: 0
United States
Message 846534 - Posted: 29 Dec 2008, 22:22:56 UTC

I am still waiting to see anything impressive from CUDA ... and then, it is my turn ...

It is going to be fun ...
ID: 846534 · Report as offensive
Profile Bambi

Send message
Joined: 15 May 99
Posts: 26
Credit: 5,704,701
RAC: 1
United Kingdom
Message 846590 - Posted: 30 Dec 2008, 0:39:41 UTC

I have just discovered the joys of optimized clients since CUDA. The whole thing made me thing am I getting the most out of my computers eating up electricity.
I have since swapped to the latest 'custom' apps for seti and AP on vista and 64b Linux.
I now have 4 cpu's and 1 gpu working on my main box and a further 3 cores chugging away on other computers, waiting to see the increase in points.
The SSSE3(?) optimized AP client makes a huge difference on my Quad core AMD
Bambi

ID: 846590 · Report as offensive
Profile Voyager
Volunteer tester
Avatar

Send message
Joined: 2 Nov 99
Posts: 602
Credit: 3,264,813
RAC: 0
United States
Message 846613 - Posted: 30 Dec 2008, 1:10:26 UTC

Is it dumb not to use cuda? I see gpus for less money than a memory upgrade ,$59 with what sounds like more improvment. I know nothing about graphic cards ,but am I getting this right?This was a lowend gpu 512mb 455mz core.
ID: 846613 · Report as offensive
tfp
Volunteer tester

Send message
Joined: 20 Feb 01
Posts: 104
Credit: 3,137,259
RAC: 0
United States
Message 846620 - Posted: 30 Dec 2008, 1:20:55 UTC - in response to Message 846534.  

I am still waiting to see anything impressive from CUDA ... and then, it is my turn ...

It is going to be fun ...


I just hope your turn is going to end up being more then the last two(+?) turns you took. I'll be happy to see something but I expect little.
ID: 846620 · Report as offensive
Profile Francois Piednoel
Avatar

Send message
Joined: 14 Jun 00
Posts: 898
Credit: 5,969,361
RAC: 0
United States
Message 846677 - Posted: 30 Dec 2008, 2:21:05 UTC - in response to Message 846557.  

I am still waiting to see anything impressive from CUDA ... and then, it is my turn ...

It is going to be fun ...

Glad you are lurking and keeping tabs on us here, Who?....
The new Nehi is now Frozen and crunching 'em up....

Keep an eye on it for the next few weeks.....if the kitty gods smile on me and nothing goes terribly wrong....
It is an impressive chippy!!


after trying the CUDA stuff, it pretty well sucks if you run it on the celeron with a GT280 ... I thought the processor did not matter ...
one more marketing hype ... The CPU still does most of the work on the CUDA stuff ...

pffffff...

Who?
ID: 846677 · Report as offensive
Profile popandbob
Volunteer tester

Send message
Joined: 19 Mar 05
Posts: 551
Credit: 4,673,015
RAC: 0
Canada
Message 846682 - Posted: 30 Dec 2008, 2:28:16 UTC - in response to Message 846677.  

I am still waiting to see anything impressive from CUDA ... and then, it is my turn ...

It is going to be fun ...

Glad you are lurking and keeping tabs on us here, Who?....
The new Nehi is now Frozen and crunching 'em up....

Keep an eye on it for the next few weeks.....if the kitty gods smile on me and nothing goes terribly wrong....
It is an impressive chippy!!


after trying the CUDA stuff, it pretty well sucks if you run it on the celeron with a GT280 ... I thought the processor did not matter ...
one more marketing hype ... The CPU still does most of the work on the CUDA stuff ...

pffffff...

Who?


The CPU has to tell the GPU what to doU what to do... Also the memory goes through the cpu to the GPU...
But then again this cuda app is really buggy so its not working quite right...


Do you Good Search for Seti@Home? http://www.goodsearch.com/?charityid=888957
Or Good Shop? http://www.goodshop.com/?charityid=888957
ID: 846682 · Report as offensive
tfp
Volunteer tester

Send message
Joined: 20 Feb 01
Posts: 104
Credit: 3,137,259
RAC: 0
United States
Message 846687 - Posted: 30 Dec 2008, 2:34:11 UTC - in response to Message 846658.  
Last modified: 30 Dec 2008, 2:35:30 UTC

Quit the sniping..........
He merely said that he was waiting to see if the app worked correctly, he would have fun with it....


Yes I can't wait to see how Intel makes out in all of this too.
ID: 846687 · Report as offensive
tfp
Volunteer tester

Send message
Joined: 20 Feb 01
Posts: 104
Credit: 3,137,259
RAC: 0
United States
Message 846692 - Posted: 30 Dec 2008, 2:37:17 UTC - in response to Message 846689.  
Last modified: 30 Dec 2008, 2:38:49 UTC

I said...quit the sniping.

He is as welcome here as any of you.....

Stop it right now!!!


Well I didn't know you ran the place, I actually updated my post before I expected it was read but now to late.
ID: 846692 · Report as offensive
Profile Crunch3r
Volunteer tester
Avatar

Send message
Joined: 15 Apr 99
Posts: 1546
Credit: 3,438,823
RAC: 0
Germany
Message 846695 - Posted: 30 Dec 2008, 2:38:37 UTC - in response to Message 846682.  


But then again this cuda app is really buggy so its not working quite right...


Yepp, and someone made a fool of himself releasing that crap into the open.

The question is whose gonna take the blame... the berkley boyz or Nvidia ...




Join BOINC United now!
ID: 846695 · Report as offensive
1 · 2 · 3 · 4 . . . 11 · Next

Message boards : Number crunching : Random Musings About the Value of CPUs vs CUDA


 
©2020 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.