Are you sure that the CUDA version is faster?

Questions and Answers : GPU applications : Are you sure that the CUDA version is faster?

Author	Message
luca1459 Send message Joined: 3 Apr 99 Posts: 10 Credit: 44,344 RAC: 0	Message 841156 - Posted: 17 Dec 2008, 23:05:53 UTC Last modified: 17 Dec 2008, 23:07:26 UTC
	Hi there, I got two questions: First: I got an Intel Core2 Quad CPU (Q9450) at 2.66ghz plus a GeForce 9800Gt 512mb. How much faster is the computing calculation using my GeForce graphic board rather that my 4-processor-CPUs? Second: Is it possibile to run both at the same time (gpu + cpu)? Thanks!
	ID: 841156 ·

Fred W Volunteer tester Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0	Message 841179 - Posted: 17 Dec 2008, 23:53:27 UTC - in response to Message 841156.
	Hi there, I got two questions: First: I got an Intel Core2 Quad CPU (Q9450) at 2.66ghz plus a GeForce 9800Gt 512mb. How much faster is the computing calculation using my GeForce graphic board rather that my 4-processor-CPUs? Second: Is it possibile to run both at the same time (gpu + cpu)? Thanks! First: I'm running my Q9450 at 3.6GHz with AK-V8 SSSE3x and do a 45 cred WU in about 2100 secs. I've seen a figure in another thread on this board which gives the time for a 45 cred WU on CUDA as about 1000 secs. So an emphatic YES. Second: Pass. I haven't seen a post from anyone claiming to be doing so or from anyone claiming to be running it in association with Optimised Apps (mind you, if it can't be done then the latter statement is meaningless :) F. ____________
	ID: 841179 ·

BlackLuke Volunteer tester Send message Joined: 19 Jul 99 Posts: 164 Credit: 54,828,301 RAC: 0	Message 858967 - Posted: 28 Jan 2009, 17:56:47 UTC - in response to Message 841179.
	> "I've seen a figure in another thread on this board which gives the time for > a 45 cred WU on CUDA as about 1000 secs. So an emphatic YES." Where did you see that? My results on an 8500 GT are slightly less than half as fast as the CPU. We need a list of GPUs that work and the Boinc/S@H configurations they worked on/with. ____________
	ID: 858967 ·

Fred W Volunteer tester Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0	Message 858998 - Posted: 28 Jan 2009, 20:18:23 UTC - in response to Message 858967.
	> "I've seen a figure in another thread on this board which gives the time for > a 45 cred WU on CUDA as about 1000 secs. So an emphatic YES." Where did you see that? My results on an 8500 GT are slightly less than half as fast as the CPU. We need a list of GPUs that work and the Boinc/S@H configurations they worked on/with. Between my last post and now, there has been a lot posted indicating that crunch-time is very dependent on which board you are using. So YMMV as they say. On the other hand, with the performance of the latest optimised build (r103) for Astropulse, on a Q9450 (my overclocked 9450 is knocking them out in 6.5 hours for ~750 credits each), however slow the GPU is you can increase your output by running AP's on the CPU and the MB's on the GPU are a bonus. F. ____________
	ID: 858998 ·

BlackLuke Volunteer tester Send message Joined: 19 Jul 99 Posts: 164 Credit: 54,828,301 RAC: 0	Message 859154 - Posted: 29 Jan 2009, 2:08:43 UTC - in response to Message 858998.
	>Between my last post and now, there has been a lot posted indicating that >crunch-time is very dependent on which board you are using. So YMMV as they >say. Where? Where? You keep writing about these miraculous GPU results. Where are they? A URL, if you please. ____________
	ID: 859154 ·

VanHalen Send message Joined: 1 Dec 99 Posts: 29 Credit: 27,001 RAC: 0	Message 859166 - Posted: 29 Jan 2009, 2:38:05 UTC - in response to Message 841156. Last modified: 29 Jan 2009, 2:40:31 UTC
	yes, you can run gpu and cpu at the same time. you need to create a CC_Config.XML with this... <cc_config> <options> <ncpus>3</ncpus> </options> </cc_config> I'm doing it on my core 2 duo dell laptop. Also I think you'll want to change the <ncpus>3</ncpus> switch to 5 to use all your cores (4 CPUs plus 1 GPU) And the CUDA client is DAMN fast ____________
	ID: 859166 ·

Borgholio Send message Joined: 2 Aug 99 Posts: 625 Credit: 10,806,022 RAC: 4,407	Message 859168 - Posted: 29 Jan 2009, 2:40:43 UTC - in response to Message 859166.
	And the CUDA client is DAMN fast My 9600GT 512mb crunches a 6.08 WU in around half an hour. Even my Core 2 Duo 2.6ghz takes 3 - 5 hours. I can't wait for the team to release a Cuda-astropulse client. ____________ You will be assimilated...bunghole!
	ID: 859168 ·

VanHalen Send message Joined: 1 Dec 99 Posts: 29 Credit: 27,001 RAC: 0	Message 859196 - Posted: 29 Jan 2009, 4:00:22 UTC - in response to Message 859168.
	And you know what else, My dell laptop has the nvidia quadro nvs140 videocard, which I think is the "smallest" one. And it busts out work units like crazy. I'd love to see this running on some quad core dual board SLI monster. ____________
	ID: 859196 ·

Fred W Volunteer tester Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0	Message 859249 - Posted: 29 Jan 2009, 6:54:12 UTC - in response to Message 859154.
	>Between my last post and now, there has been a lot posted indicating that >crunch-time is very dependent on which board you are using. So YMMV as they >say. Where? Where? You keep writing about these miraculous GPU results. Where are they? A URL, if you please. There are so many CUDA related threads now that I can't put my finger on the post that gave this figure. Just give me a day or so to finish off the backlog of VLAR tasks that got dumped on me when I tried out 6.2.2 so I can release all the tasks that got orphaned in my messing about with ap_info / no ap_info. I will then take off NNT from my machine and hopefully get some more "normal" tasks and get some "wall-time" readings for those. But that will be for a stock-clocked GTX295 on an overclocked Q9450. F. ____________
	ID: 859249 ·

The Vampire Lestat Send message Joined: 12 Aug 03 Posts: 16 Credit: 1,066,875 RAC: 0	Message 859298 - Posted: 29 Jan 2009, 12:40:17 UTC - in response to Message 858998.
	I don't know how you are doing AP workunits in 7.5 hours? My Core i7 920 @ 4.0Ghz is doing then in around 19 hours (x4) Did you work out the average and mean 30 hours for 4 x AP units? in total?
	ID: 859298 ·

Byron S Goodgame Volunteer tester Send message Joined: 16 Jan 06 Posts: 1151 Credit: 3,936,993 RAC: 0	Message 859304 - Posted: 29 Jan 2009, 13:09:28 UTC - in response to Message 859298.
	I don't know how you are doing AP workunits in 7.5 hours? My Core i7 920 @ 4.0Ghz is doing then in around 19 hours (x4) Did you work out the average and mean 30 hours for 4 x AP units? in total? I don't know how you are doing AP workunits in 7.5 hours? My Core i7 920 @ 4.0Ghz is doing then in around 19 hours (x4) Did you work out the average and mean 30 hours for 4 x AP units? in total? You seem to be only using stock apps on the MB and AP tasks you are doing that I browsed through. After checking CPU-Z to see which apps your pc is able to use, you could check out the optimized apps in the thread Modified SETI MB CUDA + opt AP package for full GPU utilization or maybe even the AK V8 + CUDA MB team work mod and these could offer you much faster process times than you'll get with the stock versions. ____________
	ID: 859304 ·

Fred W Volunteer tester Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0	Message 859310 - Posted: 29 Jan 2009, 13:21:26 UTC - in response to Message 859298.
	I don't know how you are doing AP workunits in 7.5 hours? My Core i7 920 @ 4.0Ghz is doing then in around 19 hours (x4) Did you work out the average and mean 30 hours for 4 x AP units? in total? Nope. I mean that, using the r103 Optimised App for Astropulse, each AP WU is crunched in 6.5 hours; i.e. 4 AP results uploaded every 6.5 hours. The Op Apps are wonderful that way :) F. ____________
	ID: 859310 ·

The Vampire Lestat Send message Joined: 12 Aug 03 Posts: 16 Credit: 1,066,875 RAC: 0	Message 859312 - Posted: 29 Jan 2009, 13:34:45 UTC - in response to Message 859310. Last modified: 29 Jan 2009, 13:38:29 UTC
	I am still quite new to all this and the instructions and subsequent posts make this look quite complicated I am more than game to give it a go, but are there any idiot proof guides to setting this up? Thanks very much Barry
	ID: 859312 ·

Byron S Goodgame Volunteer tester Send message Joined: 16 Jan 06 Posts: 1151 Credit: 3,936,993 RAC: 0	Message 859316 - Posted: 29 Jan 2009, 13:56:54 UTC - in response to Message 859312. Last modified: 29 Jan 2009, 14:36:13 UTC
	I am still quite new to all this and the instructions and subsequent posts make this look quite complicated I am more than game to give it a go, but are there any idiot proof guides to setting this up? Thanks very much Barry There is an Illustrated Guide to Installing an optimised application that shows how to install the optimized apps. The CUDA+opti apps I mentioned are installed the same way. The only other thing you would have to do is add a cc_config.xml file to the Boinc data folder so that you'd be able to use all the cores if you decide to use the CUDA apps. Edit: Since it's your first time with the opti apps, if you decide to try, you might want to set Boinc Manager to No New Tasks and fisish the tasks you have before installing any opti apps to avoid trashing your cache. Also feel free to ask any questions you may have during the process. Edit:2 Also if you decide not to use the modified CUDA apps but the opti app instead with a current cache of 6.08 tasks, you'll have to add a 6.08 tag to the app info file. This can be explained to you before you install the app if that's what you decide to do. ____________
	ID: 859316 ·

The Vampire Lestat Send message Joined: 12 Aug 03 Posts: 16 Credit: 1,066,875 RAC: 0	Message 859334 - Posted: 29 Jan 2009, 15:04:23 UTC - in response to Message 859316.
	Thanks very much for your help Byron, i am trying to get to number one position on the Isle of Man, should be there very soon !
	ID: 859334 ·

Fred W Volunteer tester Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0	Message 859375 - Posted: 29 Jan 2009, 17:25:38 UTC - in response to Message 859334.
	Thanks very much for your help Byron, i am trying to get to number one position on the Isle of Man, should be there very soon ! Ahh, a place where it is very difficult to swing a cat - nothing to hold it by P). F. ____________
	ID: 859375 ·

Fred W Volunteer tester Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0	Message 859434 - Posted: 29 Jan 2009, 21:39:38 UTC - in response to Message 859154.
	>Between my last post and now, there has been a lot posted indicating that >crunch-time is very dependent on which board you are using. So YMMV as they >say. Where? Where? You keep writing about these miraculous GPU results. Where are they? A URL, if you please. OK! Real figures from my Q9450 (overclocked) + GTX295 (stock speed) running Raistmer's offering that allows a version of AK-V8 to run MB on the CPU and 6.08 to run on the GPU at the same time. The first 3 WU's to complete and report were AR = .388 (so 62 credit WU's). The CPU took 3268 s (CPU time) The GPU took 723 s and 760 s (wall time from Start to Finish in the log). So at this AR, the GPU can do 4 WU's while the 4 overclocked CPU cores are doing one each - equivalent to another complete quaddie cruncher! Now, if I can only find a way to get the other side of the GTX to fire up under my Vista x64... F. ____________
	ID: 859434 ·

The Vampire Lestat Send message Joined: 12 Aug 03 Posts: 16 Credit: 1,066,875 RAC: 0	Message 859655 - Posted: 30 Jan 2009, 7:26:37 UTC - in response to Message 859434.
	Guys you are amazing ! I am now hammering out AP WUs in 6 hours!
	ID: 859655 ·

The Vampire Lestat Send message Joined: 12 Aug 03 Posts: 16 Credit: 1,066,875 RAC: 0	Message 859666 - Posted: 30 Jan 2009, 8:56:34 UTC - in response to Message 859655.
	Just out of interest, what has changed in the program for an AP to go from 21 hours to 6? thanks
	ID: 859666 ·

Byron S Goodgame Volunteer tester Send message Joined: 16 Jan 06 Posts: 1151 Credit: 3,936,993 RAC: 0	Message 859697 - Posted: 30 Jan 2009, 12:10:19 UTC - in response to Message 859666.
	Just out of interest, what has changed in the program for an AP to go from 21 hours to 6? thanks The AP app you installed has SSE3 instructions, the stock app doesn't. ____________
	ID: 859697 ·

1 · 2 · Next

Questions and Answers : GPU applications : Are you sure that the CUDA version is faster?