| GPULib on my 64-bit WinXP machine [message #63052] |
Thu, 23 October 2008 09:11  |
Vince Hradil
Messages: 574
Registered: December 1999
|
Senior Member |
|
|
I just wanted to share with the group that I got the GPULib [http://
www.txcorp.com/technologies/GPULib/] to run on my 64-bit Windows XP
machine with the new NVidia Quadro FX 5600 graphics card (1.5Gb frame
buffer [http://www.nvidia.com/object/quadro_fx_5600_4600.html]). Many
thanks to Peter Messmer at Tech-X for his help getting the correct
executable.
The results are impressive. I ran all the "demos" and the difference
is about 21-24 X! I can't wait to try to do some "real" work using
this.
The CPU on this machine is a Dual Quad-core Intel Xeon (X5482 @
3.2GHz), and it has 64Gb RAM, so it's not too shabby itself ;^)
Here's the log from running "bench.pro":
IDL> dlm_load, 'gpulib'
% Loaded DLM: GPULIB.
0.756607 0.756607 1.19352 0.206724 0.0188206
0.756607
0.756607 0.756607 1.19352 0.206724 0.0188206
0.756607
N iter = 50
CPU Time = 0.64100003
GPU Time = 0.031000137
Speedup = 20.677329
IDL> print, !version
{ x86_64 Win32 Windows Microsoft Windows 7.0 Oct 25 2007 64
64}
Vince
http://vincehradil.wordpress.com/
|
|
|
|
|
|
| Re: GPULib on my 64-bit WinXP machine [message #63155 is a reply to message #63052] |
Tue, 28 October 2008 19:23  |
Vince Hradil
Messages: 574
Registered: December 1999
|
Senior Member |
|
|
On Oct 28, 8:58 am, "Kenneth P. Bowman" <k-bow...@null.edu> wrote:
> In article <MPG.23702f206ddb6fc998a...@news.giganews.com>,
> David Fanning <n...@dfanning.com> wrote:
>
>> b_...@hotmail.com writes:
>
>>> I also see that the GTX200 series supports limited double precision
>>> operations, which might be another trump card.
>
>> The point was made at the User's Group meeting that almost
>> all of the double precision stuff, on any processor, is much
>> slower than floating point operations. The suggestion was
>> made to keep everything in floating values if at all possible.
>
>> Cheers,
>
>> David
>
> Do GPUs do IEEE arithmetic (single precision)?
>
> Ken Bowman
I believe that is what Peter Messmer said at the user meeting. Of
course, we're talking specific the CUDA tools on the NVIDIA cards.
More details at: http://en.wikipedia.org/wiki/CUDA
including this "limitation"
Various deviations from the IEEE 754 standard. Denormals and
signalling NaNs are not supported; only two IEEE rounding modes are
supported (chop and round-to-nearest even), and those are specified on
a per-instruction basis rather than in a control word (whether this is
a limitation is arguable); and the precision of division/square root
is slightly lower than single precision.
|
|
|
|
|
|
|
|
| Re: GPULib on my 64-bit WinXP machine [message #63249 is a reply to message #63052] |
Wed, 29 October 2008 21:57 |
David Fanning
Messages: 11724
Registered: August 2001
|
Senior Member |
|
|
Kenneth P. Bowman writes:
> This is like a flashback to the paleocomputing erra of attached array
> processors. (Anyone want to admit that they remember FPS, Inc.?)
Tony Kehoe, who was working for FPS in England, attended
the very first IDL class I (or anyone else, probably)
ever taught. FPS was to become the first European
distributors of IDL. Tony hadn't been in the class 10 minutes
before we started laughing about something, and it was pretty
much non-stop laughter for the next 10 years, before
he died unexpectedly at a very young age.
He would invite me over twice a year to teach an IDL
class, but that was completely secondary. Our primary
mission was to find somewhere interesting in the UK
to walk. We would spend days tramping around, laughing,
and talking about IDL. I miss him a lot, and thinking
about FPS just makes me miss him more.
Cheers,
David
--
David Fanning, Ph.D.
Fanning Software Consulting, Inc.
Coyote's Guide to IDL Programming: http://www.dfanning.com/
Sepore ma de ni thui. ("Perhaps thou speakest truth.")
|
|
|
|
|
|
|
|
| Re: GPULib on my MacBook Pro [message #63474 is a reply to message #63303] |
Sat, 08 November 2008 16:39 |
Marshall Perrin
Messages: 44
Registered: December 2005
|
Member |
|
|
Brian Larsen <balarsen@gmail.com> wrote:
> I wanted to add my macbook pro (GeForce 8600M GT) results to this
> thread. It took a while to get this running, there were 64/32 bit
> issues and patience (thanks to Nathaniel Sizemore at Tech-X).
Hi Brian,
Is there any chance that you or Nathaniel could post some tips or tricks for
replicating your success? I just downloaded GPUlib a little while ago to
test it out a bit, so I'd be most grateful if you could share any guidance
on the pitfalls to avoid while taking it for a spin. :-) Thanks much,
- Marshall
|
|
|
|
| Re: GPULib on my 64-bit WinXP machine [message #63482 is a reply to message #63052] |
Fri, 07 November 2008 15:30 |
b_gom
Messages: 105
Registered: April 2003
|
Senior Member |
|
|
OK, I've run some tests on a Core2 Extreme 3 GHz system (Q6850
processor) with 8 GB ram running WinXP 64-bit, against a GeForce GTX
280 video card (~240 pipelines, 1GB RAM). Running the 'bench.pro'
routine included with the GPUlib package, I get the following:
% Compiled module: BENCH.
0.756607 0.756607 1.19352 0.206724 0.0188206
0.756607
0.756607 0.756607 1.19352 0.206724 0.0188206
0.756607
N iter = 50
CPU Time = 2.4840000
GPU Time = 0.016000032
Speedup = 155.24968
This seems impressive, but the bench routine is highly artificial; it
just runs the gpuLGamma function repeatedly on a static input array.
If I include the time to load the array to GPU memory and read it out
after each iteration, the speedup drops to a factor of ~12.
For the other test programs, which have more realistic problems, the
speedup ranges from 10 to 100 times.
At any rate, assuming the PC has a fast bus and memory, it appears
that for appropriate types of calculations, you are much better off
buying a $500 video card then spending the premium for multiple or
faster CPUs. Once the bugs get worked out of the library (memory
management is a serious limitation), this should be an extremely
useful package.
Brad
On Oct 23, 9:11 am, Vince Hradil <vincehra...@gmail.com> wrote:
> I just wanted to share with the group that I got theGPULib[http://www.txcorp.com/technologies/GPULib/] to run on my 64-bit Windows XP
> machine with the new NVidia Quadro FX 5600 graphics card (1.5Gb frame
> buffer [http://www.nvidia.com/object/quadro_fx_5600_4600.html]). Many
> thanks to Peter Messmer at Tech-X for his help getting the correct
> executable.
>
> The results are impressive. I ran all the "demos" and the difference
> is about 21-24 X! I can't wait to try to do some "real" work using
> this.
>
> The CPU on this machine is a Dual Quad-core Intel Xeon (X5482 @
> 3.2GHz), and it has 64Gb RAM, so it's not too shabby itself ;^)
>
> Here's the log from running "bench.pro":
>
> IDL> dlm_load, 'gpulib'
> % Loaded DLM:GPULIB.
> 0.756607 0.756607 1.19352 0.206724 0.0188206
> 0.756607
> 0.756607 0.756607 1.19352 0.206724 0.0188206
> 0.756607
> N iter = 50
> CPU Time = 0.64100003
> GPU Time = 0.031000137
> Speedup = 20.677329
> IDL> print, !version
> { x86_64 Win32 Windows Microsoft Windows 7.0 Oct 25 2007 64
> 64}
>
> Vincehttp://vincehradil.wordpress.com/
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
