Parallella Community

Supercomputing for Everyone
https://parallella.org/forums/

Results of simple BRAM to user space throughput testing.

https://parallella.org/forums/viewtopic.php?f=51&t=2231

Page 1 of 1

Results of simple BRAM to user space throughput testing.

PostPosted: Tue Mar 24, 2015 4:36 pm
by boeserbaer
Hi All,

I thought I would share some early results, using the Parallella with a simple AXI BRAM controller.

44MBytes/s per thread for 4k page size transfers. I Only tested with 2 threads, but was able to get an aggregate of 88MBytes/s using mmap and memcpy to user space. Right now I have a single 8k BRAM attached.

I hope this helps someone else.

Best Regards, Mike

Re: Results of simple BRAM to user space throughput testing.

PostPosted: Wed Apr 01, 2015 3:18 am
by aolofsson
Mike,
A good data point. Thanks for sharing! The two vs one thread bw difference is something I hadn't measured. Makes sense. We also found that the memcpy in the new rootfs/kernel seems faster than the old one...
Andreas