CUDA development on different maps?

I am just starting to learn how to do CUDA development (using version 4), and I wondered if it is possible to develop another card that I plan to use? As I found out, it would be nice to know this, so I can keep track of whether the differences will affect me.

I have a macbook pro in mid-2010 with Nvidia GeForce 320M graphics cards (this is a pretty simple integrated card for laptops), but I plan to run my code on the NVIDIA Tesla "Fermi" M2050 EC2 GPUs. I am wondering if it can be localized on my laptop and then run it on EC2 with minimal changes (I do this for a personal project and do not want to spend $ 2.4 on development).

Specific question: I heard that recursions are supported on new cards (or maybe not on my laptops), what if I run recursion on my gpu laptop? will there be an error or will it be launched, but will not use the hardware functions? (I do not need a specific answer to this question, but this is what I get).

If this is a problem, are there emulators for features not available in my current map? or does the SDK imitate him for me?

Sorry if this question is too simple.

0
source share
3 answers

, . nVidia GPU , , ( , 320M (CC1.3)), , , M2070 (CC2.0)).

, , , , , , - 1.x 2.0. , , GPU - .

: 1.3 :

nvcc rec.cu -arch=sm_13
./rec.cu(5): Error: Recursive function call is not supported yet: factorial(int)

( , ), , , : 1.x , , .

: GPGPU , , . , , .

+3

Cuda . .

cc 1.0, , .

cc, .

ptx jit:

​​ , .

, , , 100% . , , cc .

, cc.

(cudaGetDeviceProperties) ptx jit, , - . cc 1.1 cuda - !

64- 3D-. ( , , )

, , cc 1.0, 100% . , cc 1,2, maxregcount , - 100% - .

!

​​ .

+2

NVidia Kepler K20 GPU, 2012 CUDA 5, .

0
source

All Articles