4 Replies Latest reply on Oct 13, 2015 2:34 AM by dipak

    Need Radeon R9 270 Specs


      I have looked everywhere, but haven't found the specs that I need for ocl.

      In particular I am looking for:


      1) # hardware queues (software?)

      2) VRAM banks and channels


      These are needed for kernel development. Any info or link would be much appreciated



        • Re: Need Radeon R9 270 Specs

          It would be nice if clinfo could detect these settings and report them;-)

            • Re: Need Radeon R9 270 Specs

              Thanks Nikos for your suggestion.

              Actually clinfo reports platform and device specific many useful information queried using clGetPlatformInfo  and clGetDeviceInfo  APIs. These information are retrieved by standard OpenCL parameters. The same can be done by any other OpenCL program.

              The information you are looking for are not standard ones (at least for now, may become in future), rather than specific to h/w architecture. So, I've doubt whether they will be part of clinfo or not.

              However, I do understand  your point. I know that such information are useful to achieve higher performance, but portability of code may get affected as well. At this point, I'm not sure whether these information can be programmatically accessible or not. And if so, at which level.




                • Re: Need Radeon R9 270 Specs

                  Thanks for your fast reply,


                  At this point, I'm not interested in portability, just in performance. I am not asking for programmatic access to these parameters, just personal for architectural design.

                  Most of the hardware specs I have found about my card focus on # of transistors, ROPs and TMUs. The Optimization guide is a bit vague on them "volcanic island devices have 4-8 hardware queues", and I'm wondering whether its memory table is common to all GCN devices.

                  Clinfo is not necessarily constrained to a pure, exclusive ocl application. Could employ some proprietary calls for some of the more shady components. It is understandable that direct GPU access is much more difficult than through ocl.


                  If anyone knows where I can find this info, even in a hardware picture, it would be great help.