GPU computing with 'Quadro 600'

Hi,

I am new to GPU computing. I will be doing CFD calculation in 3D which will requires working with really big 3D matrices like (500,550,300). I used a workstation which has 'Quadro 600' GPU which has 1GB ram but when I start a calculation more than 200,250,150 size matrices it say out-memory. You requested 85mb memory only available 37mb. I did not understand this error. I thought Matlab will be using 1gb ram as its memory. Am I wrong?

Thank you

[Walter Roberson]

On the line just before you do that, please show the output of

memory

(Note: only works in MS Windows)

[James Lebak]

Even better, since the matrices are on the GPU, issue the gpuDevice() command and look at the value of the 'FreeMemory' field. This works on all platforms. It's possible that other matrices have used all the free memory, or that memory fragmentation caused by repeated allocations and de-allocations has reduced the size of the largest available piece of memory. Either way, clearing GPU variables can help.

Answer 1

[Jason Ross]

The Quadro 600 has 1 GB of RAM, but it's also a display card, so if you have a monitor hooked up to it, some of the RAM is being used for the display.

Clearing variables (as James suggested) can help, but if you are using the same card for both compute and display, the resources available for display are going to compete with the resources for compute, and you aren't going to be able to use all the memory on the card for compute alone. You might want to look at the GeForce cards if you are going to be using more RAM and cost is a concern, as your eventual problem size is 11x larger (1 GB). They also have more CUDA cores than the Quadro 600.