OpenCL: Device information [Updated Sep 04 2011]

[Petr Schreiber]

The following script does tell you the basic information about your OpenCL device:

Name
Vendor
Driver version
Profile (full or embeded)
Device version
Extensions


... and as the bonus also:

Number of compute units
Top clock frequency
... and max parameter size


I attach the screen I get on my PC.

Important note: The code requires the OpenCL headers (http://www.thinbasic.com/community/showthread.php?10159-OpenCL-Headers-Updated-Sep-04-2011).

[kryton9]

Petr you are loving this OpenCL I see. THanks for all your efforts.

Here is my info:

Create OpenCL device & context...
Query all devices available to the context...
Query available text information...
======================================================================

GeForce 8800 GT
NVIDIA Corporation
195.62
FULL_PROFILE
OpenCL 1.0 CUDA

cl_khr_byte_addressable_store
cl_khr_gl_sharing
cl_nv_compiler_options
cl_nv_device_attribute_query
cl_khr_global_int32_base_atomics
cl_khr_global_int32_extended_atomics

======================================================================
Query available numeric information...
======================================================================

Max compute units : 14 => 112 cores
Device clock (MHz): 1715
Max parameter size: 4352

======================================================================

Press any key to quit...

[Charles Pegge]

Mine is a mini version :)




Create OpenCL device & context...
Query all devices available to the context...
Query available text information...
======================================================================

GeForce GT 120
NVIDIA Corporation
190.89
FULL_PROFILE
OpenCL 1.0

cl_khr_byte_addressable_store
cl_nv_compiler_options
cl_nv_device_attribute_query
cl_khr_global_int32_base_atomics
cl_khr_global_int32_extended_atomics

======================================================================
Query available numeric information...
======================================================================

Max compute units : 4 => 32 cores
Device clock (MHz): 1350
Max parameter size: 4352

======================================================================

Press any key to quit...

[Petr Schreiber]

Thank you guys,

very interesting comparison.
I beat Charles by 25MHz (all power of my 486SX25 based laptop :D).

Charles, if I may, I recommend you to install 195.62 WHQL (http://www.nvidia.com/Download/index.aspx?lang=en-us) drivers, they are very good and bring neat boost in OpenCL perfomance.


Petr

[Michael Clease]

Petr you are tempting me to buy a new gfx card just see what it can do but I dont really play games so I dont want to spend mega bucks. I did see a GTX250 896B DDR3 for £129 yesterday at pcworld which didnt seem to bad for off the shelf.

Stop tempting me :grrrr: :diablo:

[Petr Schreiber]

GT250 looks very good, but I have something to save you :P,

NVIDIA Fermi based cards should appear soon, so maybe worth a wait.
They should offer good double precision performance (my 9500GT does only floats).

Or maybe something new from ATi?

Oh wait ... you mean ... to choose a life without OpenCL? Nooooo :lol:

Ok, to settle down the excitement - I am writing gauss blur kernel currently, and it is a bit of pain to code.
But I must admit this new way of coding is very refreshing - you have to think differentely, and you can get speedup usually achieved using assembler with relatively high level syntax code.


Petr

[Charles Pegge]

You could get something cheap and cheerful first Michael - The high-end cards for this year will come down in price and become the norm for next year.

@Petr: 158 meg download for the latest drivers (for Vista64). I thought 1 meg was extravagant!

[Petr Schreiber]

I am sorry Charles, but it is worth it :D

The thing is the drivers contain also PhysX runtimes and other stuff ... no longer just "graphic drivers".
But yes, it is extreme.

[Charles Pegge]

My new 8Meg broadband only took about 4 minutes to download it. Seems to be the new standard here - even in the outer reaches of West Wales.


Fermi: Double precision would be perfect for general purpose parallel computing - You have to be quite careful to avoid cumulative errors with single precision.

[Petr Schreiber]

Fermi: Double precision would be perfect for general purpose parallel computing - You have to be quite careful to avoid cumulative errors with single precision.


Yes, exactly.

But I think GT 2xx series support double precision already, could you please try to run the test again, to see if support appears in form of new extension?

Fermi then just should add brutal boost in dp calculations.

[kryton9]

I'm itching to get a new video card already. When you play the new games you see the difference!
I saw that the 6Core Intels are leaked on Amazon, so hopefully the i7's will drop in price soon!!

[Charles Pegge]

Petr: The GT120 I assume belongs to the 100 series - no mention of double precision after installing the latest drivers. But I don't think this lower spec will hinder development unduly.



Create OpenCL device & context...
Query all devices available to the context...
Query available text information...
======================================================================

GeForce GT 120
NVIDIA Corporation
195.62
FULL_PROFILE
OpenCL 1.0 CUDA

cl_khr_byte_addressable_store
cl_khr_gl_sharing
cl_nv_compiler_options
cl_nv_device_attribute_query
cl_khr_global_int32_base_atomics
cl_khr_global_int32_extended_atomics

======================================================================
Query available numeric information...
======================================================================

Max compute units : 4 => 32 cores
Device clock (MHz): 1350
Max parameter size: 4352

======================================================================

Press any key to quit...

[Petr Schreiber]

Updated the code to more specification compliant form.

[Charles Pegge]

Yes Petr, that is now working fine with latest NVIDIA drivers / GT100 / Vista
Charles

[Petr Schreiber]

Thanks Charles!