Chih-Che 2017-08-16 4:53 GMT-04:00 Chih-Che Chueh <[email protected]>: > It is nice to meet you again. I still remember we met and chatted a little > bit in 2013 deal.II workshop, which was held in Texas A&M University. It has been a while! Nice to meet you again.
> Is there any reason why you still used the previous way? Is this because > your GPU accelerator was launched some years ago so that you are unable to > use the new simple way? The problem with UVM is that it makes it easy to write bad code. I am working on a code where we use UVM and it happens quite often that we run some code that was intended for the GPU on the CPU. Because of UVM it works but at the cost of performance and so it is harder to detect that you made a mistake. Another advantage of copying the data manually is that you don't hide the costly copy operation. This is also the philosophy used in Kokkos. So yes UVM makes writing the code easier but you need to be a lot more careful when you write the code if you don't want to perform a bunch of useless copy. Of course, if you have NVLink it doesn't matter but that only works for POWER chips which most people don't have access to. Best, Bruno -- The deal.II project is located at http://www.dealii.org/ For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en --- You received this message because you are subscribed to the Google Groups "deal.II User Group" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
