extremely new, need advice and info on my setup please.

Hello,

First post here and I am really new to GPGPU computing. I have currently read 2 books on CUDA it doesn’t come close to getting my hands dirty with code, I have just been trying to get the ideas concepts down, while trying to download the CUDA toolkit & SDK (I am out on a job site and our internet connection is far from supreme we have to use an internet sat dish, and the wind gets to mach 5).

Anyway my main development workstation will be a HP Elitebook 8770W 3rd Generation i7 quad core with either the NVIDIA Quadro K5000M or the K4000M Running Windows 7 Pro 64 (waiting upon approval through IT and Corporate billing though my direct manager has approved already). I have requested in my mind a rather powerful beast, compared to the target workstations that will be running the software. My Idea is to have something that will support future generations of our field workstaions, plus being able to test the software in multiple VM’s.

Now for the target platform workstations we have in the field. They are Dell Precision R5400 (rack mount, on most jobs we will run 3 at a time) Intel Xeon E5450 @ 3.0 GHz 4 GB RAM with Nvidia Quadro FX 3700 GPU running Windows XP SP3. Now I know running a 32 bit OS has its disadvantages but the company will not change choice of OS soon. I know that Windows XP SP3 will only handle up to 4 gb ram and only 2 CPU’s (if that’s correct). Now I really need to utilize the GPU to see if I can speed these up.

My main question is, what pit falls will await me just by looking at my setup? Anyone know more limitations the Windows XP SP3 has (I have googled and read the wiki and nothing beyond what I have mentioned). Also the current GPU what issues? I understand it only supports CUDA 1.1

What issues would anyone see that I would have developing on such a different platform.

I know you all do not have a crystal ball, and don’t know the software situation and what is to be investigated and developed. I can’t say much about that. But I can say that I am in the Oil and Gas Industry and we rely on Real Time data being displayed in charts / graphs, and the meat and potatoes of our system is calculations, simulations, flow modeling and handling OPC and Instrumentation data.

I apologize if I am asking very specific combination of hardware, OS and software setup. I assume each has a very different system they are working on, and the chances of anyone matching the exact setup that I will have is slim.

I also apologize if I haven’t provided enough info. If there is anything else that anyone would need info wise please let me know. I would very much like to know any restrictions, draw backs bottlenecks hurdles that my current setup may have.

As I progress in this field I will be more than happy to help and answer questions and share code… that is if I can ever download the tools.

Thank you all!

landroid / Lanny1

Oh I didn’t mention that the Xeon is quad-core

That I can think offhand, running CUDA kernels on Windows XP that take longer than 5 seconds is not possible to my knowledge: https://devtalk.nvidia.com/default/topic/405795/need-solution-of-quot-kernel-launch-timeout-quot-from-nvidia/ With Windows Vista/7 it is possible to let kernels run for larger amounts of time by disabling Timeout Detection & Recovery (TDR) feature. This might not be an issue if your kernel(s) are rather simple.

The other thing I can think of is that the Quadro FX 3700 is a 128 core, Compute Capability (CC) 1.1 card. This means it does not support native double-precision (DP) arithmetic if that support is required for your use… you’d need at card capable of at least CC 1.3 support. See: http://en.wikipedia.org/wiki/CUDA#Version_features_and_specifications for support of specific features of cards with different CC.

If the algorithms you plan to execute are relatively simple/don’t require DP support or any additional CC features above 1.1, you should be fine. Otherwise, you’d be looking to upgrade those video cards on your target hardware platform. CPU/hard drives on your target platforms is mostly irrelevant unless you’re attempting to do very intensive processing (i.e. process HD video streams in real-time, etc)

I can only talk from my experience. CUDA programming on linux is really easy. I found the windows cuda programming a pain. If you are in for performance, use linux.