Warning for Cuda

Message boards : Number crunching : Warning for Cuda
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · 4 . . . 6 · Next

AuthorMessage
Darth Beaver Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 20 Aug 99
Posts: 6728
Credit: 21,443,075
RAC: 3
Australia
Message 951893 - Posted: 3 Dec 2009, 6:45:21 UTC

To all users Cuda was and is the most stupid idea that has come out . So if you are going to use Cuda or ATI then understand you are asking a chip that runs at 650mhz to do units then remember this a piii650mhz cpu chip did seti in aprox 18hrs your cuda chip runs at same speed and will take 20 hrs to do units you are asking the eqvilet of a piii chip to do work so don't be surprised if your mouse stops working or the sound of music you are listening to stops or you lose the screen or weird things start happening GPU's are designed to display things not to do seti do not be stupid like the people that came up with this idea if you have more than i cuda then only have 1 cuda running you will need the other for displaying things and if like me you have a hi def tv card you will defintaly need the second cuda chip and i would also keep in mind your graphic crad will probbly burn out within 8mths unless you clean the heat sink as the fan will run flat out and clog up quickly .....proof of what i am saying ....cuda units ran for 58mins only did .567 % ... about 40% of units take this long and i can give other examples if anybody wish to argue the point i have been around scince 98 and the average computer was a pii , piii maxium speed of about 450mhz Cuda ,STUPID,STUPID,STUPID,STUPID,STUPID,STUPID
ID: 951893 · Report as offensive
Profile Wingless Wonder

Send message
Joined: 14 May 99
Posts: 14
Credit: 12,157,146
RAC: 0
United States
Message 951916 - Posted: 3 Dec 2009, 8:59:20 UTC - in response to Message 951893.  

Duly noted.
ID: 951916 · Report as offensive
Darth Beaver Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 20 Aug 99
Posts: 6728
Credit: 21,443,075
RAC: 3
Australia
Message 951918 - Posted: 3 Dec 2009, 9:34:50 UTC

PS Also remember that Nvida do not support the use of Cuda for Seti which says it all and they are the world leaders in graphic cards
ID: 951918 · Report as offensive
OzzFan Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Apr 02
Posts: 15691
Credit: 84,761,841
RAC: 28
United States
Message 951955 - Posted: 3 Dec 2009, 14:43:32 UTC

Well then, I guess there's no use explaining why you've had problems since you already have all the answers. So I won't tell you that CUDA cards take longer to crunch Very High Angle Range Workunits, but the real speed comes in on normal workunits or Very Low Angle Range Workunits, because you've been around since 1998 and you know that CUDA sucks because of a singular experience.

I also won't bother to tell you that nVidia wrote the CUDA application for SETI@Home for free, so if that isn't unofficially supporting SETI on CUDA, I don't know what is. More than likely you went to nVidia for tech support and they turned you away saying that they don't support it because tech support costs money in manpower, and the application isn't theirs to support since it was given freely to SETI@Home - leaving SETI to support the application for which they did not write.

Yep, CUDA sucks alright. That's why most of the top crunchers and most of the hardcore crunchers in the Number Crunching forum use it - because they like things that suck so bad!
ID: 951955 · Report as offensive
Profile skildude
Avatar

Send message
Joined: 4 Oct 00
Posts: 9541
Credit: 50,759,529
RAC: 60
Yemen
Message 951964 - Posted: 3 Dec 2009, 15:27:13 UTC - in response to Message 951955.  

Switch that. CUDA has trouble with Very low angle range(VLAR) WU's the High Angle Range WU's are very fast on a cuda card. yes the speed of the card may be 650Mhz. What isnt taken are the actual cores onboard the chip. the card runs the WU's in parallel on the chips which makes the process very efficient. Lets not forget that the onboard RAM is much more advanced than the pc100 or pc 133 that ran on the old P3 boards. ATI HD series of cards have a double precision which in essence makes them 2X faster than a comparable CUDA card. So no these cards aren't running old P3 technology. Please check into the Gflops they can process. Compare that with a new CPU and you'll see that the Video cards are surpassing the CPU's in processing speed


In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope
ID: 951964 · Report as offensive
OzzFan Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Apr 02
Posts: 15691
Credit: 84,761,841
RAC: 28
United States
Message 951966 - Posted: 3 Dec 2009, 15:28:39 UTC - in response to Message 951964.  

Switch that. CUDA has trouble with Very low angle range(VLAR) WU's the High Angle Range WU's are very fast on a cuda card.


Thanks. I always get the two mixed up.
ID: 951966 · Report as offensive
Profile skildude
Avatar

Send message
Joined: 4 Oct 00
Posts: 9541
Credit: 50,759,529
RAC: 60
Yemen
Message 951968 - Posted: 3 Dec 2009, 15:33:16 UTC - in response to Message 951966.  
Last modified: 3 Dec 2009, 15:34:39 UTC

No problem. ; ) I thought had remembered the information wrong since you're never wrong. Then I remembered that the Lunatic app has a VLAR killer.


In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope
ID: 951968 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 951970 - Posted: 3 Dec 2009, 15:40:13 UTC - in response to Message 951966.  

Charlie, memory aid: Very (S)Low Angle Rate, Very High(speed) Angle Rate.
ID: 951970 · Report as offensive
Profile BigDaddyDave
Avatar

Send message
Joined: 8 Oct 00
Posts: 67
Credit: 20,337,995
RAC: 15
United States
Message 952173 - Posted: 4 Dec 2009, 3:12:54 UTC - in response to Message 951893.  

Glenn,

Per Ageless - Courtesy starts with your first post of the thread and I agree but I disagree with you statement. I have a 2004 P4-3.2GHz with an old ATI 800X video card- 128MBs, so I added a Nvidia GTX260 back in late April. Check out the difference in ave WU. Can you tell what month it was. It's like I put my computer on Viagra. Woot!




It use to take me 17 months to gain 100,000 credit now I can crunch it in 3 months. That is a MASSIVE difference. That's around 570% more WUs then prior. The ave WU went from a CPU of 18hrs to a GPU of 22mins.


Regarding your time watching TV and crunching, man you aren't kidding. I'll get the blue screen of death if I run both. But I always leave the computer running when I'm not at home and I only watch TV around 2-3 hours a night. During that time I turn off BOINC.


So get a new GPU everyone, just don't play COD4 and TV while SETI running ;)

BDDave
ID: 952173 · Report as offensive
Darth Beaver Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 20 Aug 99
Posts: 6728
Credit: 21,443,075
RAC: 3
Australia
Message 952174 - Posted: 4 Dec 2009, 3:21:04 UTC - in response to Message 951955.  

FOR OZZIEFAN .My system is a e6550 cpu at 2.3 gig,2 gig dd2 ram at 666mhz dual channel giveing f/b speed of 1.2 gig, gigabyte s p31 motherboard, 600 watt power supply, gigabyte nvida 9600gt i gig video card ,1 gigabit eathernet onboard ,8 usb ports , netgear 802.1g wifi card ,winfast hi def tv tuner card ,1800 ver., i terrabit of h/d space , 2 lg dvd burners , 2 web cams 2 megpixels each, 1 hp scanner printer (elchepo version), 8 speakers as the board has 7.1 channel thd sound .So now you know i have a pretty good system that is upgradeable to a quaddy core if i wish to . And yet i still have troubble. At the moment i am not doing Cuda units and guess what has happend ? . You know all so you can tell me what is happening now with the seti units the cpu's are doing . If you can't reply with what is happening then i know you are a drrrrrrrr brain and not as smart as you think .looking forward to your explination on the affects of not doing cuda units and how fast i do units with just the cpu's doing units
.I did say i was prepared to debait cuda and it's affects .(and to tink i had a very simular converstion with a scintist at nasa 10 years ago and i was able to prove to him what i was saying ). Anybody can use a particular unit as a arguement but i am talking about all the units and averages and how cuda can actully overlaod peoples system under normal conditions not special server set up by smart alex's which are in the minorty of users .
ID: 952174 · Report as offensive
OzzFan Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Apr 02
Posts: 15691
Credit: 84,761,841
RAC: 28
United States
Message 952197 - Posted: 4 Dec 2009, 6:41:05 UTC - in response to Message 952174.  
Last modified: 4 Dec 2009, 7:07:23 UTC

FOR OZZIEFAN .


My handle is OzzFan, as in Ozzy Osbourne. Not Ozzie.

And yet i still have troubble. At the moment i am not doing Cuda units and guess what has happend ? . You know all so you can tell me what is happening now with the seti units the cpu's are doing . looking forward to your explination on the affects of not doing cuda units and how fast i do units with just the cpu's doing units


This machine is getting a RAC of 19,000 using a GeForce GTX 260 and an Intel Core 2 Quad Q9450 overclocked. Is all that RAC coming from the quad core CPU alone?

This machine is getting a RAC of 11,000 using an old Athlon 64 3000+ CPU, not overclocked. A normal Athlon 64 3000+ without CUDA gets a RAC of about 300 on it's own.

This machine is getting a RAC of 49,500+ using a current AMD Phenom II X4 BE and a dual GeForce GTX 260 with each video card having two GPUs (for a total of 4 CUDA processors). A normal Phenom II X4 get about a 4,000 to 6,000 RAC on it's own.

These are all non-server setups, but most SETI users are anything but normal. I can go on, but I think I've made my point.

...but i am talking about all the units and averages and how cuda can actully overlaod peoples system under normal conditions not special server set up by smart alex's which are in the minorty of users .


Since I don't actually own a CUDA card because I'm an ATi guy, are you saying that the owners of all those machines I just pointed out to you are "smart alex's" who run special server setups? They might be in the minority of users, but only because they are hardcore about their hardware and getting the most out of it, and I think they might take offense to some of your comments.

[Edit] I'll tell you what, since you're so adamant about debating your views, and since I don't actually own a CUDA card, why don't I move this over to the Number Crunching forum so that the actual owners of CUDA systems can argue with you instead? That way you can convince all of them that CUDA is bad and they are all a bunch of "smart alex's".
ID: 952197 · Report as offensive
Darth Beaver Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 20 Aug 99
Posts: 6728
Credit: 21,443,075
RAC: 3
Australia
Message 952202 - Posted: 4 Dec 2009, 7:09:20 UTC - in response to Message 952197.  

dear ozzie what ever at the moment i am doing a unit in 1 hrs 16 mins on each cpu when i do cuda units it blows out to 7 hrs i rest my case i averaged 20,000 points + for the month of november if i'm to listen to you i should do less right ! well lets see shall we .(i have allready tried this hahahahahahahahah).
ID: 952202 · Report as offensive
OzzFan Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Apr 02
Posts: 15691
Credit: 84,761,841
RAC: 28
United States
Message 952203 - Posted: 4 Dec 2009, 7:11:26 UTC - in response to Message 952202.  
Last modified: 4 Dec 2009, 7:12:04 UTC

dear ozzie what ever at the moment i am doing a unit in 1 hrs 16 mins on each cpu when i do cuda units it blows out to 7 hrs i rest my case i averaged 20,000 points + for the month of november if i'm to listen to you i should do less right ! well lets see shall we .(i have allready tried this hahahahahahahahah).


So it's not possible that the workunits you are doing on your GPU are the ones that CUDA isn't very good at, i.e. Very Low Angle Range? Wouldn't that have an effect on your workunit performance?
ID: 952203 · Report as offensive
Darth Beaver Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 20 Aug 99
Posts: 6728
Credit: 21,443,075
RAC: 3
Australia
Message 952211 - Posted: 4 Dec 2009, 7:37:04 UTC - in response to Message 952203.  

Boy you realy are like ozzie ozborne . So i am not doing cuda units i have dumped all cuda and stoped downloading .If you do not do cuda units it looks like the gpu and cpu are working togeather and so hence the units are being done quickly however if i then ask cuda to do units as well it slows down and the units average 5hrs or more normally about 6 hrs + i have just done 10 units in less than 10 hrs .If your right i shouldn't have been able to do that many. Seeing as i do not have AMD chips i can't comment on them and if your system can only do 300 points then i would say AMD are not that good. Also cuda units points can be as low as 1 point or as much as 50 so you need to do heaps of units to reach just 300 and again i rest my case what good is doing cuda if most of the unit only give you 38 points when normal seti give you 80-120 you don't need to do as many to achive a high Rac in fact if i do Enstein my RAC jumps to over 1000 , and i don't member reciveing 1 unit just for the cuda chip at Eistein so next time quote hrs and number of units as the RAC can be fickkle .I can get 2000 points awarded in 24 hrs and then the next day i can only get 280 so RAC is fickle.
ID: 952211 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 952214 - Posted: 4 Dec 2009, 7:47:48 UTC - in response to Message 952211.  
Last modified: 4 Dec 2009, 8:08:15 UTC

A GF9400 has 16 shader processors, which will all attack one task at the same time. This as opposed to your CPU, which will only use one processor-core per task.

What Ozzfan is trying to explain to you is that there are certain tasks here at Seti, the so-called VLARs, which will take a long time on the GPU, as it has problems with tasks with that low an angle rate. Those are better crunched on CPUs.

So yes, if you're running a bunch of VLARs at the moment, they're better off on the CPU.

But would you run into a bunch of non-VLARs later on, which would take an equal amount of time or longer on your CPU, but which can be done by your GPU in less than the lowest time on your CPU, then you would notice the difference.

Before you ask, no I don't crunch with CUDA either. Yet I did write the CUDA/CAL FAQ. Why don't you go read a little through it?
ID: 952214 · Report as offensive
kittyman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Jul 00
Posts: 51468
Credit: 1,018,363,574
RAC: 1,004
United States
Message 952216 - Posted: 4 Dec 2009, 8:01:33 UTC

Stupid, stupid, stupid?

My 3 Cuda cards have given me my highest RAC ever.......

I really miss the point he is trying to make.

???
"Freedom is just Chaos, with better lighting." Alan Dean Foster

ID: 952216 · Report as offensive
OzzFan Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Apr 02
Posts: 15691
Credit: 84,761,841
RAC: 28
United States
Message 952218 - Posted: 4 Dec 2009, 8:06:12 UTC - in response to Message 952211.  

So i am not doing cuda units i have dumped all cuda and stoped downloading .If you do not do cuda units it looks like the gpu and cpu are working togeather and so hence the units are being done quickly however if i then ask cuda to do units as well it slows down and the units average 5hrs or more normally about 6 hrs + i have just done 10 units in less than 10 hrs .If your right i shouldn't have been able to do that many.


What I am telling you is that only Very Low Angle Range workunits take a long time (longer than the CPU) to complete. Any other type of workunit finishes in a fraction of the time it takes the CPU, but it depends on what GPU we're talking about as to how much of a fraction. Some GPUs have more shader processors than others, and this affects their overall performance.

Also cuda units points can be as low as 1 point or as much as 50 so you need to do heaps of units to reach just 300 and again i rest my case what good is doing cuda if most of the unit only give you 38 points when normal seti give you 80-120 you don't need to do as many to achive a high Rac in fact if i do Enstein my RAC jumps to over 1000 , and i don't member reciveing 1 unit just for the cuda chip at Eistein so next time quote hrs and number of units as the RAC can be fickkle .I can get 2000 points awarded in 24 hrs and then the next day i can only get 280 so RAC is fickle.


Even if RAC is fickle, if CUDA is so much slower than the CPU, then how do you explain the RACs of those systems I posted? How is it that a system with a AMD Athlon 64 3000+, which normally gets about 300 RAC has an 11,000+ RAC with a CUDA card? How is it that a Core 2 Quad Q9450, which normally gets about 4,000 RAC is getting 19,000 RAC with a CUDA card? How do you explain those increases in numbers?
ID: 952218 · Report as offensive
OzzFan Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Apr 02
Posts: 15691
Credit: 84,761,841
RAC: 28
United States
Message 952220 - Posted: 4 Dec 2009, 8:08:26 UTC - in response to Message 952216.  

I really miss the point he is trying to make.


He's misinformed. I'm assuming he got a VLAR on his GPU which was taking longer than his CPU would normally do a workunit, so he arrived at the presumptuous conclusion that CUDA is slower.
ID: 952220 · Report as offensive
Luke
Volunteer developer
Avatar

Send message
Joined: 31 Dec 06
Posts: 2546
Credit: 817,560
RAC: 0
New Zealand
Message 952221 - Posted: 4 Dec 2009, 8:13:29 UTC - in response to Message 952211.  
Last modified: 4 Dec 2009, 9:41:48 UTC

Boy, you really are like Ozzie Osborne.
So I am not doing CUDA units, I have dumped all CUDA and stopped downloading (them). If you do not do CUDA units it looks like the gpu and cpu are working together and so hence the units are being done quickly, however if I then ask CUDA to do units as well it slows down and the units average 5hrs or more normally about 6 hrs + I have just done 10 units in less than 10 hrs.
If you're right, I shouldn't have been able to do that many. Seeing as I do not have AMD chips I can't comment on them and if your system can only do 300 points then I would say AMD are not that good. Also CUDA units points can be as low as 1 point or as much as 50 so you need to do heaps of units to reach just 300 and again i rest my case what good is doing CUDA if most of the unit only give you 38 points when normal SETI@home give you 80-120 you don't need to do as many to achieve a high RAC.
In fact if I do Einstein my RAC jumps to over 1000, and I don't remember receiving 1 unit just for the CUDA chip at Einstein so next time quote hrs and number of units as the RAC can be fickle. I can get 2000 points awarded in 24 hrs and then the next day I can only get 280 so RAC is fickle.


We need to go back to the basics here.
1. The credits awarded between CUDA, SETI@home and Einstein are all roughly the same. Credits are a BOINC-wide points system, so they vary very little between projects.
2. The amount of workunits you receive does not in any way affect how long it takes to crunch and send back a result for a workunit.
3. Just because a single-core AMD system gives out a RAC of 300, do you expect a quad-core extreme AMD CPU to also output a RAC of 300? No. You can clearly see a AMD is the top crunching CPU in SETI@home (with a little help from a GPU farm).
4. CUDA workunits can give out 0.01 points to 150 points, exactly the same as S@H Multibeam.
5. Again, there is no (major) difference in the amount of credits given out per project. It is based on a BOINC-wide system.

Thank you,
- Luke.
- Luke.
ID: 952221 · Report as offensive
kittyman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Jul 00
Posts: 51468
Credit: 1,018,363,574
RAC: 1,004
United States
Message 952222 - Posted: 4 Dec 2009, 8:13:33 UTC - in response to Message 952220.  

I really miss the point he is trying to make.


He's misinformed. I'm assuming he got a VLAR on his GPU which was taking longer than his CPU would normally do a workunit, so he arrived at the presumptuous conclusion that CUDA is slower.

OK.....
He needs lessons on Lunatics auto installer, probably correct drivers and 2.3dlls, rescheduler, and patience....LOL.
"Freedom is just Chaos, with better lighting." Alan Dean Foster

ID: 952222 · Report as offensive
1 · 2 · 3 · 4 . . . 6 · Next

Message boards : Number crunching : Warning for Cuda


 
©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.