For what applications are you using LabVIEW and GPUs?

[jschwartz]

Hi Everyone,

I am working with a team here at NI to build awareness and knowledge of GPU computing, specifically as it relates to using GPUs tied to real-world I/O.  While I am currently working with GPUs tied to I/O, I would like to ganther any information possible on GPU computing with LabVIEW. If you could answer any or all of the following, it would give me much more insight into the possibilities and will help us to steer our development of these tools:

1) What applications do you see LabVIEW and GPUs being a fit?

2) What experience do you have with LabVIEW and GPU programming?

3) What applications have you put together that use LabVIEW and GPUs?

4) What applications do you see GPUs being integral to solving?

Thanks for your time and your answers!

Jim Schwartz

National Instruments

[jschwartz]

Because we are building the business case for increasing investment in GPU support for LabVIEW, it would be great to hear any applications that actual users  are implementing with the current toolkit.  Thanks!

[SDR Kid]

Jim,

Here's my input from a Navy lab perspective:

1)  We are looking into the GPU performing many of the functions we typically program on to FPGAs.  We are doing various software defined radio functions such as polyphase filtering for channelizing a wideband spectrum, modulation/demodulation, error-correction, pattern recognition, beam-forming, etc.  One of my projects looks into video enhancement through cyclostationary covariance pattern extraction, which inherently has plenty of parallelism process threads.  Once we have the algorithm established, we may move it to a GPU if it's too taxing for a typical core.

2)  I have a pretty indepth knowledge of LabVIEW and a couple years of programming DSPs with C/C++ tools like TI Code Composer and ADs Visual DSP++.  Last year I downloaded the LabVIEW Black-Scholes Model GPU example along with Visual C++ project.  I used a copy of Visual C++ 2008 to compile and build the DLL, which looks like it works.  I plan to build off that project to experiment with some of the CUDA calls such as CUFFT.

3)  I have not put any applications together yet.

4)  I'm currently streaming 100MSamples per second from an AlazarTech 9462 digitizer PCIe card to my 8 core Dell using LabVIEW, so my next step is sending it through the GPU to perform the channel filtration and then the CUFFT so I can use the I&Q information for demodulation.  It would be great to have a GPU version of the NI filters and modulation tookits instead of me coding all these functions.  From what I've been reading about the don'ts of GPUs is that you have to avoid a lot of IO between the cores and the GPU, so a GPU toolkit may need to pass local GPU pointers between functions or the IO flow may lose the advantages.

Hope this information helps in your business case study.

David Neumann

Advanced Technology Branch

SPAWAR Atlantic - Charleston

[jschwartz]

Hi David,

Thank you for your reply!  These kind of ideas are exactly what we are looking for.

Jim

[momolxg]

Dear Jim,

I have no experience in GPU programming but wish to incorporate future multicore GPU's into a tokamak control project.

I would like to be the abe to harness the parallel capability of multicore coprocessor cards without having to write software for any particular hardware.

With an Intel MIC copprocessor card (Knights Corner) I should be able to carry out parallel calculations of the shape and position of the plasma as well as ray-tracing, tomography etc simultaneously. Such a coprocessor card is especially well suited as a minimal number of inputs and outputs need to be transported to and from the coprocessor card. (I am visiting Lothar Wenzel in Austin next week)

Greetings, Louis, IPP Garching

[rpursley8]

We are currently using a GPU card to perform 3D cross correlations of volume data acquired from a microscope system for the purpose of motion tracking.  We use LabVIEW as the interface, but had to develop code into dlls to call from LabVIEW because we are using a 64-bit OS (Windows Vista/7).

We have several possible projects, mostly involving image processing, that we plan to incorporate a GPU card into.

[inversion]

Hi Jim,

We have just been given the go ahead for an automated 24 channel audio analysis tool for amplifier and head unit development where 24 channels of audio will be analysed using CUFFT where applicable e.g. THD+N measurements. The concept for this project utilises a GTX 460 for this task.

1) What applications do you see LabVIEW and GPUs being a fit?

Multichannel audio analysis using FFTs also in real-time

2) What experience do you have with LabVIEW and GPU programming?

I can bind DLLs into LabVIEW and program elements in visual studio using C

3) What applications have you put together that use LabVIEW and GPUs?

About to start my first

4) What applications do you see GPUs being integral to solving?

massively parallel tasks... slow user interfaces (graph updates, onscreen real-time analysis)... offloading tasks from the CPU (silently/auto if possible)...

audio applications... heavy dsp algorithms... updating graph data directly from GPU memory to the screen without flowing over the CPU!? (is that even possible currently!?)

Would be great to see a complete toolkit that fully supports 32-bit and 64-bit systems. I would like to use 64-bit in this new project, but im not too sure if all the hardware drivers are mature enough yet to even consider giving it a go down the 64-bit river.

hope the info helps

[jschwartz]

Thanks for the post - as this is in the early stages of development, suggestions are great to hear!  We have begun compiling a list of possible applications could take advantage of the power of GPUs.  If you would like to read up on these, check out the linked documents at:

http://zone.ni.com/devzone/cda/tut/p/id/12434

Again, thanks for your post and keep us updated as things come up!

[carlosmt]

Hello everyone.

In my case I am making a system of adaptive optics and multi optic conjugate, for astronomy

I decide to try using a GPU, for the large amount of CPU, s in parallel.

Thanks.

[mjochinsen]

1) What applications do you see LabVIEW and GPUs being a fit?

OCT (Optical Coherence Tomography), Medical Imaging Processing.

2) What experience do you have with LabVIEW and GPU programming?

Advanced LabView and LabView FPGAs. No experience on GPU

3) What applications have you put together that use LabVIEW and GPUs?

None yet

4) What applications do you see GPUs being integral to solving?

Medical Imaging. Mostly FFT and complex FFTs