Hi Zalster ;)

I got the GTX 750 Ti and put it in my store bought computer where it used to have a GT 640. It's crunching much faster now.

What kind of a computer is it you are using? Which CPU?

"...the longest mutlbeam went from about 1 hour 15 minutes down to 30 minutes, the quickest went from 40 minutes down to 20 minutes so in a sense it did cut the computation time in half.."

Can you explain this? The GPU time stamp is weird! And according to BilBg all work units are the same size. Then why is it, that the GPU takes 12 hours to crunch the package? The CPU only uses 1.5 hours!?!?

The GPU:
Run time ......CPU time
43,416.30 ..... 65.09 . pending ....... Period Search Application v101.11 (cuda55)

The CPU:
Run time ......CPU time
5,002.31 ...... 4,767.18 ...... pending ....... Period Search Application v102.10 (sse2)

I'll check back here periodically to see how you are doing. I wish you good luck.

Just got the new hardware! I'll be putting it together all night (day, where you are I guess:)

Thanks for the interest my friend ;)





Hi BilBg,

Don't be so shy ;)

OK, I see what you mean ;) Maybe I'll try it on anyway ;)
Thanks for that one ;)

Instead of re-posting in Number Crunching you may create a thread there with links to most important for you threads that you already have here in Questions and Answers

OK, I see... Its better that way, and pretty smart. Then you don't have 2 type everything again. Thanks ;)

---

Project Headless CLI Linux Multiple GPU Boinc Servers
Ubuntu Server 14.04.1 64bit
Kernel 3.13.0-32-generic
CPU's i5-4690K
GPU's GT640/GTX750TI
Nvidia v.340.29
BOINC v.7.2.42

Can you explain this? The GPU time stamp is weird! And according to BilBg all work units are the same size.

Actually, that's not entirely true, and I don't think that's what BilBg was trying to tell you.

More accurately, all workunits are the same physical size. That is, each SETI@home MultiBeam v7 workunit is 367KiBi (Kilo-binary bytes). However, the workunits can be recorded at various Angle Ranges on the satellite dish, which can affect how much terrestrial "noise" is in the workunit. For those with a lot of terrestrial noise, the compute times will be less. For those will less terrestrial noise, the compute times will be longer.

So not all workunits are made the same, even if they are the same size on your hard drive. If you want to properly compare CPU and GPU crunch times, you need to make sure you're comparing workunits of the same Angle Range.

Then, to complicate things, most of the GPU workunits with lots of terrestrial noise in them, fall back to the CPU for blanking as the CPU is better capable of handling this. Which means even if a workunit is supposed to be assigned to the GPU, it may be using the CPU for quite a bit of the processing, giving inaccurate results when comparing the two.


So it really comes down to, you have to pay attention to Angle Ranges before you compare run times.

Hi Volunteer ;)

All but the job done by CUDA has this funny timestamp to "cpu time". Runtime seems just right!

GPU: cpu time 65.09 sec (this actually took 11.8 hours.) [65.09 sec????]
GPU: Runtime 43,416.30 sec [about 723 minutes which is correct 12,06H]

CPU: cpu time 4,767.18 sec (this took 1.5 hours) [about 83 minutes which is correct]
CPU: Runtime 5,002.31 sec [about 83 minutes which is correct]

Thanks for a very good explanation!! Thank you!

---

Project Headless CLI Linux Multiple GPU Boinc Servers
Ubuntu Server 14.04.1 64bit
Kernel 3.13.0-32-generic
CPU's i5-4690K
GPU's GT640/GTX750TI
Nvidia v.340.29
BOINC v.7.2.42

GPU: cpu time 65.09 sec (this actually took 11.8 hours.) [65.09 sec????]
GPU: Runtime 43,416.30 sec [about 723 minutes which is correct 12,06H

I didn't mention AstroPulse workunits in my previous post, but the same holds true for AstroPulse type workunits. AstroPulse workunits are approximately 8MiB (mega-binary bytes) large on your physical hard drive, but processing time also greatly relies on the Angle Range.

At 723 minutes, it almost sounds like an AstroPulse workunit. You definitely can't compare AstroPulse GPU to SEI@home CPU. Not a fair comparison.

[Edit]At first even 723 minutes sounded like a lot to me, so I checked your GPU and saw that you're using a GT 610. The performance of these, while better than CPU crunching, is not going to be that great. My Radeon HD 7970, which isn't nearly as good as it's nVidia counterparts (at least here on SETI@home, while on other projects AMD GPUs are better), crunches an AstroPulse workunit in about 30 minutes or so.

Hi Zalster,


I've jumped back into the thread where it began... Just to keep it to a limit.
http://setiathome.berkeley.edu/forum_thread.php?id=73032&postid=1484254#1484254

---

Project Headless CLI Linux Multiple GPU Boinc Servers
Ubuntu Server 14.04.1 64bit
Kernel 3.13.0-32-generic
CPU's i5-4690K
GPU's GT640/GTX750TI
Nvidia v.340.29
BOINC v.7.2.42

Hi Dan,

It's a computer I bought from a Dell. It's an optiplex. Here are the specifics..

GenuineIntel
Intel(R) Core(TM) i5-3470 CPU @ 3.20GHz [Family 6 Model 58 Stepping 9] (4 processors)
NVIDIA GeForce GTX 750 Ti (2048MB) driver: 334.89 OpenCL: 1.01
Microsoft Windows 7 Professional x64 Edition, Service Pack 1, (06.01.7601.00)



Ok, as to my statement of the times to complete my seti work units. I was averaging the time to complete.
Here is a rough comparison of our graphic cards.

Your current GT 610 has 48 cuda cores
My previous GT 640 had 384 cuda cores.
My current GTX 750 Ti has 640 cuda cores


As the number of cuda increase, there are more to do the repetitive task of analyzing the data, stopping only get get the CPU to check.

So GPU Runtime is time GPU is looking at Data
GPU cpu runtime is the time that CPU went over and looked at the data that the GPU flagged
for it to check.

The GPU will always use a portion of the CPU to check the data. Volunteer is right, your GPU isn't much fast than your CPU. Hope this helped explain, if not I can post a LONGER explanation.

Keep Crunching

Hi Zalster ;)

The GPU will always use a portion of the CPU to check the data. Volunteer is right, your GPU isn't much fast than your CPU. Hope this helped explain, if not I can post a LONGER explanation.

OK, it's just that the CUDA/GPU job was 12 hours to finish, and the CPU did 4 jobs in 1.5 hours!?!?

Ok, as to my statement of the times to complete my seti work units. I was averaging the time to complete.
Here is a rough comparison of our graphic cards.

Your current GT 610 has 48 cuda cores
My previous GT 640 had 384 cuda cores.
My current GTX 750 Ti has 640 cuda cores

OK, on this server 2/test system 2 I chose GT640. I think it was Guy who found this card. If I can find another larger card, which is Low Profile, I'll try that one too. It's my plan to build a headless cruncher which is not so pricy and pretty easy to build. Then I'll make a todo. The hardware for this machine costed around 1200,00US$ It's the industrial PSU and the 2U Rack Case which is a little expensive. I don't know how much parts cost in the US, but I'll guess it's about the same as us minus 30% or so ;)

It's a computer I bought from a Dell. It's an optiplex. Here are the specifics..
GenuineIntel
Intel(R) Core(TM) i5-3470 CPU @ 3.20GHz [Family 6 Model 58 Stepping 9] (4 processors)
NVIDIA GeForce GTX 750 Ti (2048MB) driver: 334.89 OpenCL: 1.01
Microsoft Windows 7 Professional x64 Edition, Service Pack 1, (06.01.7601.00)

OK! It's the same CPU as in test system 1 !! Actually, I'm working on a old HP Workstation xw4300 a factory computer. I've got 2 notebooks, 1 Asus with a large monitor 21 inches with a Intel duo CPU and a little Compac Netbook with a Intel Atom CPU. Both "up'ed" a little with al the RAM they can take and with SanDisk SSD'. Windows 7 on the large one due to my graphical jobs and Ubuntu Desktop on the little netbook because this is the one I use at the University.
But, when I work at my desk, here, writing you and the other guys, it's an old HP Workstation with an old Pentium CPU ;) Isn't funny? All that nize hardware, all goes into the Rack for Boinc, and then I use this "museum piece" for work ;)

As you can see below, the Rack Case from Germany is great! I converted it to Low Profile, the whole length of the backside. So that more than one graphic card can be installed. Im' waiting for the second parcel, with 2 more Asus cards, 1 for PCIe2.0 and one for PCIe3.0. The 3.0 for this system 2 and the 2.0 for server 1 which still has the sleepy GT610 ;)
You can see the difference between server 2 on top, and server 1 in the bottom.:




Asus doesn't deliver there cards with the Low Profile bracket, this is why I still haven't mounted/installed the vga port (whats the right word for this?):




So, I'm installing Ubuntu Server 12.04.4 64bit right now, and then later, I will try the different things regarding the GPU. First installation of the driver to see if it works right away with this card, and the go on with the suggestions I've got through the last 14 days. It's early morning here. 7:35 am.
I'll have to be in Copenhagen at the russian embassy at 3 pm. this afternoon for a major protest! So the GPU will first be tested this evening. I'm looking forward to it ;)
.

---

Project Headless CLI Linux Multiple GPU Boinc Servers
Ubuntu Server 14.04.1 64bit
Kernel 3.13.0-32-generic
CPU's i5-4690K
GPU's GT640/GTX750TI
Nvidia v.340.29
BOINC v.7.2.42

Hi,


All right, here we go... The first error after installing Ubuntu Server 12.04.4 on the test server 2 (Headless Linux Boinc Server - No GUI - All CLI)

I'm controlling it via SSH/Putty, but this was what where on the monitor connected during installation ;)



AFTER INSTALLATION - UBUNTU SERVER 12.04.4 - UPDATE/UPGRADE - NETWORK STATIC

Ubuntu 12.04.4 LTS halifax.domain.tld tty1

halifax login: [ 3.881440] nouveau E[ DEVICE][0000:02:00.0] unknown chipset, 0x108020a1
3.881444] nouveau E[ DRM] failed to create 0x80000080, -22



Anybody who know this error message? It's the nouveau driver which conflicts. Yes, but I haven't installed driver for the Asus card yet ;) Just a little cookie for later ;)

---

Project Headless CLI Linux Multiple GPU Boinc Servers
Ubuntu Server 14.04.1 64bit
Kernel 3.13.0-32-generic
CPU's i5-4690K
GPU's GT640/GTX750TI
Nvidia v.340.29
BOINC v.7.2.42

it's just that the CUDA/GPU job was 12 hours to finish, and the CPU did 4 jobs in 1.5 hours!?!?

Hi Dan,

Lets try something different. Usually a GPU is faster than a CPU if it is a larger GPU (meaning more cuda cores) With entry level GPU, there are fewer cores to share the work load. The best way I can think to explain how it works is the following.

I'm going to borrow from another webpage to try and explain CPU and GPU. Think of the CPU like a foreman of a factory. His job is to check the data that flows on a conveyer belt. He can stop and look at it if he needs to, change things as needed. But this is very repetitive task. So now he decides to get a core from a GPU and tell the GPU core what to look for normally. The GPU core checks the data as it comes down the conveyer belt and when he sees something wrong, he calls the CPU over to check it. If it's ok, they restart the line again till another abnormal occurs, which the CPU comes over again and checks. This is a slow process since there is only 1 line. But what if we had 100 conveyer belts , each with a GPU core to check the data flowing, then the CPU could move from conveyer line to conveyer line checking each time one of them stops to check the abnormal. By doing this, much more data is flowing through the factory as not all the lines are stopping, only those that need the attention of the CPU. Each file of data from seti is divide up between the conveyor belts and processed by the different cores of the GPU.

As to why your GPU is slower than the CPU. Again, the more cores the GPU has the faster it can process all the data. Right now, you are dealing with a GPU with small amount of cuda cores to analyze the data. The analogy I would give is that your current GPU is like 1 line of conveyor belts. The CPU is faster as it knows what to look for and can go through the data faster. The GPU is slower since it has to stop and wait for the CPU to double check it's result. Once you get those GT 640 installed and running, you will have many more conveyer belts running and you should see the times to analyze each data drop significantly.

I hope this analogy helps. If you already knew all this then I apologize. I mean no disrespect. But it was the best thing I could think of to try and explain the process of data.

Good luck with the protest. Be careful out there. Unfortunately I don't believe there will be much help coming from our side of the Atlantic. I think this is only the beginning of many bad things to come..

Keep us informed.

Keep Crunching.......

Can you explain this? The GPU time stamp is weird! And according to BilBg all work units are the same size.

Actually, that's not entirely true, and I don't think that's what BilBg was trying to tell you.

More accurately, all workunits are the same physical size. That is, each SETI@home MultiBeam v7 workunit is 367KiBi (Kilo-binary bytes).

No to both (= I didn't imply either meaning)

In fact my sentence was:
"Wrong assumption for both Asteroids@home and SETI@home as tasks send to CPU or GPU are the same."

I.e. exactly the same WU is sent to somebody's CPU and your GPU (or vice versa) (or to 2 different CPUs, or to 2 different GPUs)

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

The GPU will always use a portion of the CPU to check the data. Volunteer is right, your GPU isn't much fast than your CPU. Hope this helped explain, if not I can post a LONGER explanation.

OK, it's just that the CUDA/GPU job was 12 hours to finish, and the CPU did 4 jobs in 1.5 hours!?!?

OK, somebody told you that it is faster to go to town by car and not by bicycle.

Then you decide to try a car from 1920 driven by your grandma (slow GPU, slow/new Asteroids@home CUDA app)
to compare to your professional/sports bicycle (carbon fibers, etc.) ridden by a racer (fast CPU, fast/optimized Asteroids@home CPU apps - for a year they become 4-5 times faster)

If you try this comparison on SETI@home CUDA apps you will see much better speed on the same GPU
(I already gave you a link to Asteroids@home forum thread with table that shows how slow Asteroids@home CUDA app is)


Your CPU core do a task in 1.5 hours, my ~500 MHz CPU from 2000 do it in 3-5 days - there is nothing strange that different Hardware/Software will do the same work in different time.

For CUDA tasks the 'Run time' (clock time) is what matters, 'CPU time' is only used to feed the GPU with data and some more management functions (e.g. to update 'Progress %')

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

Hi,


OK.. Just installed the Desktop edition on test system 2. Using the Asus GeForce GT640 on this machine, there's a little improvement. Instead of 12 hours a job now takes about 7 hours. And the GPU does it in about 1 hour and 21 minutes per core.
So is this because of the card being a bicycle compared to others? It is I guess :(

Test system 1 now only does work using the CPU. I guess the 3 wheeled bicycle couldn't take any more, RIP!!! No one showed up at the funereal by the way!

I got the results for the last month and a half, and it seems that making a system all command line based using only the CPU is faster than using a GUI based system with a sleepy GPU. Not the result I was looking for.

Another problem which I just noticed, is that the GPU only runs asteroid jobs and not SETI jobs. This I don't get.

[/img]

---

Project Headless CLI Linux Multiple GPU Boinc Servers
Ubuntu Server 14.04.1 64bit
Kernel 3.13.0-32-generic
CPU's i5-4690K
GPU's GT640/GTX750TI
Nvidia v.340.29
BOINC v.7.2.42

Another problem which I just noticed, is that the GPU only runs asteroid jobs and not SETI jobs. This I don't get.

Re-read your threads - maybe 2-3 times I gave you links on how to do SETI@home CUDA on Linux ...
http://setiathome.berkeley.edu/forum_thread.php?id=73359&postid=1479147#1479147

For the 'speed' of Asteroids@home CUDA app ask on their thread (this link for this table was already posted by me, no need to ask here for the speed of another project app - they will know better):
http://asteroidsathome.net/boinc/forum_thread.php?id=233&postid=2228#2228

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

Re-read your threads - maybe 2-3 times I gave you links on how to do SETI@home CUDA on Linux ...

So, talking about 64bit and Linux, when not crunching in v.6.03 or 6.07 it's not done by Nvidia/Cuda!? There's one "AstroPulse v6 v6.03 " though. And according to "the list" http://setiathome.berkeley.edu/apps.php this means done by GPU, right?

If not, please enlighten me ;)

---

Project Headless CLI Linux Multiple GPU Boinc Servers
Ubuntu Server 14.04.1 64bit
Kernel 3.13.0-32-generic
CPU's i5-4690K
GPU's GT640/GTX750TI
Nvidia v.340.29
BOINC v.7.2.42

To be done by (sent to) GPU the label of the task have to contain word CUDA or OpenCL, like:
Linux/x86_64 7.04 (opencl_ati5_linux) (this is for ATI AMD OpenCL SETI@home v7)
Linux/x86_64 6.07 (cuda_opencl_100) (this is for NVIDIA OpenCL AstroPulse v6)

Tasks marked:
Linux/x86_64 7.01 (for SETI@home v7)
Linux/x86_64 6.03 (for AstroPulse v6)

... are of course done by CPU only.

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

Also see this post for more links:
http://setiathome.berkeley.edu/forum_thread.php?id=74253&postid=1485135#1485135

And you'll need to edit ('merge') the app_info.xml file accordingly.

The info about app_info.xml is in the packages and in this general description: Anonymous platform


Of course there will be no point in 'installing' (manually) any CUDA app if BOINC do not detect CUDA
or 'installing' any OpenCL app if BOINC do not detect OpenCL

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

Sorry, I'm not quite sure about this. I get some, but not all. I'll have to do some more reading. But thanks for pointing it out...

Noticed "Jon Rock's" message:
Wed 05 Mar 2014 12:49:59 PM EST | SETI@home | Message from server: Your app_info.xml file doesn't have a usable version of SETI@home v7.

From wiki:
"...Each BOINC-based project has application versions for one or more platforms. When the BOINC client requests work from the project's server, the client tells the server its platform, and the server gives it the appropriate version.."

"...Create a file app_info.xml in the project directory. This file describes the applications you have compiled or downloaded. Its format is described below.
Run the BOINC client again. When it requests work from the scheduling server, it will report its platform as 'anonymous' and send a list of the applications you have supplied. The server will sends tasks for those applications.."


So, the jobs has to be specifies in this file. It's not enough to tell boinc in the cc_config.xml file to "use_all_gpus" ?? You'll have to tell it what to do with it as well!?!?

Sorry if I don't get it.. This is news to me.. If I'm being to g.. d... thick headed, just ignore the questions..


Then what exactly does this do:

08-Linux 64bit OpenCL AstroPulse for NV GPUs (r1844), May 2013

OpenCL based Seti@Home AstroPulse v6 application for 64bit Linux platform. Intended for all NVIDIA CUDA capable GPUs using a proprietary NVIDIA driver (258.19, better newer). Please install the NV driver with OpenCL runtime before using this application! MD5 : 9c255304571e3cd0602795cc0166e031 APv6.07r1844_clNV_linux64.7z


I saw these on your profile/berkeley site:

AstroPulse v6
Anonymous platform (ATI GPU)

SETI@home v7
Anonymous platform (CPU)

Again this "Anonymous platform"!?!?
To get Boinc SETI units crunching, do I have to make this "app_info.xml" file and specify which jobs I want to be running!? And what are the file from above doing "08-Linux 64bit OpenCL AstroPulse for NV GPUs (r1844), May 2013" ???
.

---

Project Headless CLI Linux Multiple GPU Boinc Servers
Ubuntu Server 14.04.1 64bit
Kernel 3.13.0-32-generic
CPU's i5-4690K
GPU's GT640/GTX750TI
Nvidia v.340.29
BOINC v.7.2.42

Get the 3 packages (.7z files), see what's inside them, then ask
(the 'problem' part is: which of the .xml files to 'merge' to get proper app_info.xml)


In short:
Copy from 'Files_to_install' all files except app_info.xml
to SETI@home directory (<BOINC_Data>\projects\setiathome.berkeley.edu\)
I heard on Linux you may need also to:
chmod +x

Compose proper app_info.xml by 'merging' some of the example files (what to 'merge' depend on which combination of packages you will use, more info only if you are interested)

WARNING: Bad app_info.xml may make BOINC delete all SETI@home tasks you have now, you may prefer to finish/report them all first, then:
Copy that combined app_info.xml to SETI@home directory and restart BOINC


* 'merge' in this case means: concatenate 2 files, then remove these 2 lines in the middle:
</app_info>
<app_info>


* What all of the above (if you use all the 3 packages) will give you:
Optimized SETI@home v7 CPU SSE3 app (MBv7_7.05r1848_sse3_linux64_CPU0.7z) - faster than the stock app
SETI@home v7 CUDA app so your NVIDIA can do SETI@home tasks - the only way for now (as already noted there is no stock SETI@home CUDA app for Linux)
Newer AstroPulse OpenCL app for NVIDIA (I think it have less CPU usage than the stock (which is of the same author))

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

I saw these on your profile/berkeley site:

AstroPulse v6
Anonymous platform (ATI GPU)

SETI@home v7
Anonymous platform (CPU)

Again this "Anonymous platform"!?!?

I use app_info.xml - everyone that do this is shown on the website as using Anonymous platform

My app_info.xml (I don't use all possible applications for my computer (Windows XP + AMD CPU + ATI AMD GPU), only what I prefer - SETI@home v7 on CPU + AstroPulse OpenCL for ATI GPU)
http://pastebin.com/hnjci4UX


There are some standard/predefined platforms:
<platform>windows_intelx86</platform>
<platform>windows_x86_64</platform>
<platform>x86_64-pc-linux-gnu</platform>
.....
https://boinc.berkeley.edu/trac/wiki/BoincPlatforms

Your BOINC tells BOINC server on what 'platform' it runs, BOINC server looks what apps are available on the server that can run on your 'platform' + CPU + GPU type and sends them (if any).


'Anonymous platform' means platform-with-no-name or 'platform of unknown name':
http://dictionary.reference.com/browse/anonymous

This is the way to tell BOINC server:
"don't mind what my 'platform' is, just send me tasks of 'this kind' for my CPU and 'that kind' for my GPU, I already have apps that can run on my computer and I described them in an app_info.xml"

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

Hi BilBg,


BOINC and SETI using app_info.xml. For what and for what reason??
OK! So we are using the app_info.xml to tell SETI servers (and Asteroid servers in my case, but this is another case for another thread) which applications we would like to receive and run.

Before app_info.xml
Looking at the applications appointed my cruncher "halifax" without the use of app_info.xml, I can see this:
http://setiathome.berkeley.edu/results.php?hostid=7238932
http://setiathome.berkeley.edu/host_app_versions.php?hostid=7238932

Under jobs, I can see the following being crunched:
SETI@home v7 v7.01
AstroPulse v6 v6.03
AstroPulse v6 v6.01

Under Applications I can see:
AstroPulse v6 6.01 i686-pc-linux-gnu
AstroPulse v6 6.03 86_64-pc-linux-gnu
SETI@home v7 7.01 i686-pc-linux-gnu
SETI@home v7 7.01 x86_64-pc-linux-gnu

So right now this cruncher "halifax" is doing both 32bit jobs and 64bit jobs of both AstroPulse and SETI@home!?!?

Use app_info.xml or not!?!?
OK, if we use app_info.xml, we can define which applications we want to run. But what's to gain?
1. Apparently my system "halifax" is currently crunching both 32bit jobs and 64bit jobs. This we could change by specifying this in app_info.xml!?!?
2. Currently "halifax" is NOT using the GPU to crunch. This we can fix by specifying it in the app_info.xml!?!?
How do we specify that we want to crunch using the GPU's as well?? At "Asteroid@home" GPU crunching application name is "Period Search Application 101.12 x86_64-pc-linux-gnu (cuda55)" We need both the application name and a file I guess.

I found this, and I guess its something like that which needs to be added to my app_info.xml file. But I cant find the files for Linux. Here's the content as I guess it needs to be:

<app>
    <name>setiathome_enhanced</name>
</app>
<file_info>
    <name>setiathome_6.6_windows_intelx86.exe</name>  --> SOMETHING _x86_64-pc-linux-gnu ????
    <executable/>
</file_info>
<file_info>
    <name>setiathome_graphics_6.6_windows_intelx86.exe</name>  --> SOMETHING _x86_64-pc-linux-gnu ????
    <executable/>
</file_info>
<app_version>
    <app_name>setiathome_enhanced</app_name>
    <version_num>660</version_num>
    <api_version>6.1.0</api_version>   --> OK, IM LOST!!
    [
      <coproc>
          <type>CUDA</type>
          <count>1</count>
      </coproc>
    ]
    [ <plan_class>cuda</plan_class> ]
    [ <flops>X</flops> ]
    [ <avg_ncpus>X</avg_ncpus> ]
    [ <max_ncpus>X</max_ncpus> ]
    [ <cmdline>args</cmdline> ]
    <file_ref>
        <file_name>setiathome_6.6_windows_intelx86.exe</file_name>  --> SOMETHING _4.35_x86_64-pc-linux-gnu ????
        <main_program/>
    </file_ref>
    <file_ref>
        <file_name>setiathome_graphics_6.6_windows_intelx86.exe</file_name>  --> SOMETHING _4.35_x86_64-pc-linux-gnu ????
        <open_name>graphics_app</open_name>
    </file_ref>
</app_version>

And then we need to add the other applications too. So could this be the right for this 64bit system with 2 GPU's?:

seti@home_enhanced v.something
SETI@home v7 v7.01
AstroPulse v6 v6.03

Using these files:

SETI@home vSOMETHING 86_64-pc-linux-gnu
AstroPulse v6 6.03 86_64-pc-linux-gnu
SETI@home v7 7.01 x86_64-pc-linux-gnu

So that we run AstroPulse 64bit, SETI@home 64bit and SETI Enhanced?? for GPU's!?!?


Not sure! SOS!!
I'm not sure which files to use to crunch with GPU's.

I'm not sure why the system runs both 32bit and 64bit now! If we are using app_info.xml will there be a reason for running 32bit on a 64bit system?

I'm not sure where to put the files. The app_info.xml, yes, but not the others. I have read that they go into the projects directory. But where is this when only using the boinc-client. I need to find that too.


Conclusion!?
Using the app_info.xml we can crunch data using our GPU's! But where else is the advantage in using app_info.xml? Is there other advantages?
.

---

Project Headless CLI Linux Multiple GPU Boinc Servers
Ubuntu Server 14.04.1 64bit
Kernel 3.13.0-32-generic
CPU's i5-4690K
GPU's GT640/GTX750TI
Nvidia v.340.29
BOINC v.7.2.42