cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

akasheee
Journeyman III

Two Radeon 5870 GPUs for OpenCL computing

Hi

I am new in GPU computing. I have following two issues.

1.  I have two GPU Radeon 5870 GPUs and want to use OpenCL to accelerate some of my programs. The motherboard that I am using  is ASUS P6X58D-E LGA 1366 Intel X58 SATA 6Gb/s USB 3.0 ATX Intel Motherboard. Though the motherboard has 3 PCIe slots, I can use only two to plug the GPUs, as there are not enough space in the third PCIe slot. Now the usable two PCIe slots are very close to each other and thus if I plug these two GPUs in those PCIe slots, both the GPUs become hot. Actually then the space between two GPUs is about 0.5 inch and probably the GPUs cannot dissipate heat properly. Is it ok, to use these GPUs in this condition? I am scared, they might be damaged if I keep them in this condition.

2. Right now I removed one of the GPUs from the PCIe slot and using only one GPU for my computing. The monitor is also connected to one of the DVI port of this GPU. I am using Ubuntu 9.10 and I installed the ati driver 10.4, sdk 2.1 properly. They are working fine. But somehow the GPU is not detected and only cpu is detected for usable OpenCL device. I mean when I type ./CLInfo I got the following:

ashraf@apollo1:~/Desktop/ati-stream-sdk-v2.1-lnx32/samples/opencl/bin/x86$ ./CLInfo
No protocol specified
Number of platforms:                            1
  Platform Profile:                             FULL_PROFILE
  Platform Version:                             OpenCL 1.0 ATI-Stream-v2.1 (145)
  Platform Name:                                 ATI Stream
  Platform Vendor:                              Advanced Micro Devices, Inc.
  Platform Extensions:                   cl_khr_icd


  Platform Name:                                 ATI Stream
Number of devices:                              1
  Device Type:                                  CL_DEVICE_TYPE_CPU
............. etc.

Now my question is: can I not use the GPU (Radeon 5870) for computing if I connect it to the monitor for video output? If I want to use this GPU for computing, do I need to buy another ATI graphic card for video output? I have already an NVIDIA graphic card (9500GT, 512mb, Dual DVI) for video. Can I use this NVIDIA card (9500GT, 512mb, Dual DVI) for video output and use the ATI Radeon 5870 for computing?

 

I will really appreciate if someone can help me about these issues.

Thanks in advance

Ashraf

 

Tags (2)
0 Likes
10 Replies
nou
Exemplar

Two Radeon 5870 GPUs for OpenCL computing

no you do not need another GPU.

did 3D acceleration work properly? try run FindNumDevices from SDK CAL examples.

0 Likes
akasheee
Journeyman III

Two Radeon 5870 GPUs for OpenCL computing

Actually I did not fing the 3D acceleration. Following is the ./FindNumDevices

 

ashraf@apollo1:~/Desktop/ati-stream-sdk-v2.1-lnx32/samples/cal/bin/x86$ ./FindNumDevices

Supported CAL Runtime Version: 1.3.185
Found CAL Runtime Version: 1.4.696
Use -? for help
No protocol specified
CAL initialized.
Finding out number of devices :-
Device Count = 0
CAL shutdown successful.

Press enter to exit...

 

It seems it sees no devices. The driver is not installed properly? How do I know that the driver is properly installed or not?

0 Likes
omkaranathan
Journeyman III

Two Radeon 5870 GPUs for OpenCL computing

The driver is not installed properly.

Try doing a clean and reinstall.

Device count will show the number of GPUs present in case the driver is installed properly.

Also afte the driver installation, run the command line

'glxinfo | grep direct' and post the output.

0 Likes
nou
Exemplar

Two Radeon 5870 GPUs for OpenCL computing

i recomend build packages with running installer with

--buildpkg Ubuntu/lucid

and then install them wit

dpkg -i *.deb

after that run aticonfig --initial

0 Likes
akasheee
Journeyman III

Two Radeon 5870 GPUs for OpenCL computing

I am new to Linux. Thus I need some clarification.

I have Ubuntu 9.10 installed, 32 bit in a Intel i7 system. The X Protocol Version 11, Revision 0.

So where do I get running installer as you recommend. Is it in the link http://forums.amd.com/devforum/messageview.cfm?catid=390&threadid=125792&enterthread=y ?

I have downloaded the ati-opencl-runtime_2.1_i386.deb. Then shall  I do the following?

./ati-opencl-runtime_2.1_i386.deb --buildpkg Ubuntu

Then

dpkg -i ati-opencl-runtime_2.1_i386.deb

Then

aticonfig --initial

Do I need clean the driver that I tried to installe before. If so, how shall I clean it?

Thanks for help

0 Likes
psath
Journeyman III

Two Radeon 5870 GPUs for OpenCL computing

It seems to go without saying, but since you didn't specify its worth noting that, as of right now, the AMD implementation of OpenCL requires that you to be logged into an X windowing system (Gnome or KDE if using Kubuntu). AKA in 'runlevel 5'.

If you're interfacing from a virtual terminal inside a GUI then this is probably not your problem. But if you're trying from a pure command line then it probably is.

Try running the command 'runlevel' and making sure it displays 'x 5' (where 'x' is your previous runlevel and 5 is your current.)

0 Likes
nou
Exemplar

Two Radeon 5870 GPUs for OpenCL computing

no that is packages with OpenCL runtime.

and as you have 9.10 then you run with --buildpkg Ubuntu/jaunty

here are how do you install it. http://wiki.cchtml.com/index.php/Ubuntu_Jaunty_Installation_Guide

0 Likes
sniper_sung
Journeyman III

Two Radeon 5870 GPUs for OpenCL computing

You should use a motherboard with better PCIE spacing, e.g. ASUS P6TD Deluxe, EVGA Classified E760, Gigabyte UD9 etc. If the temperature of a graphics card exceeds 80C, the probability of getting an error is increased seriously.

0 Likes
akasheee
Journeyman III

Two Radeon 5870 GPUs for OpenCL computing

Thanks for the useful suggestions.

I have another question. If a the ATI 5870 is connected with a monitor and provide the video  for normal computer operation, how much of its resources are occupied for the video. How much resource of this GPU are left for scientific computing.

0 Likes