General-purpose GPUs: which framework should you use?
25 May 2021
No
-  

 

 

Software Outlook have published a new technical report in which they summarise and compare a number of frameworks for harnessing the use of general-purpose GPUs.

Yes

​​

 

CoSeC_Logo_high-res.jpg

    Report_pic.PNG

    In the past decade, General-Purpose Graphical Processing Units (GPGPUs) have become widely available in both stand-alone and HPC compute resources. At the same time, a number of frameworks have evolved that can be used to utilise GPGPUs within codes but they range widely in terms of ease of code porting and how effectively the GPGPUs are used. Software Outlook​ have explored a number of the available frameworks to enable the scientific communities supported by CoSeC (and beyond) to choose the best framework for their situation. They have recently published a STFC Technical Report that containsa summary of the different frameworks with example codes
    • comparison of ease of code porting include resources available 
    • comparison of current compiler and architecture support
    • comparison of code performance using our benchmark code, which is available via the Software Outlook code repository​.  

    Thus, the scientific communities will be able use their resources (both compute and code development) more effectively.

    By performing this work as part of the Software Outlook Project, CoSeC is ensuring that the same work package is not being replicated multiple times within the different CoSeC communities.



    Contact: Geatches, Dawn (STFC,DL,SC)