Device Query via driver API on any video card - how?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi people! External Image

I have one question: How to get info about GPU directly via driver api or even low level? I mean when there is no information about the system my app is gonna run on, the first step - which kind of gpu? if it’s nVIDIA then is it PTX-supported? if yes, how many thread processors? core frequency? Texture/surface memory?.. full info. Can someone tell me how to do this?

2
  1. Try to load nvcuda.dll. If failed => no device or incompatible (old) driver.
  2. cuGetDeviceCount() to get number of compatible devices
  3. Programming Manual for the rest.

Search the forum, it’s been answered many times already…