Synthetic benchmarks: MEMBENCH sparse memory speed

What about single process sparse (semi-random) memory access?

MEMBENCH random memory speed - one process

Well, we see a common picture: the read test execute faster on ganged configuration, while the other test benefit from the unganged mode.

Now, let's see what happen if two processes use sparse (semi-random) memory access:

MEMBENCH random memory speed - two processes

The read advantage of ganged mode is almost vanished, while the advantage of unganged configuration in write and mixed (copy) test grow.

In the end, let's look at the performance of the concurrent (a reading and a writing processes) sparse memory access

MEMBENCH random memory speed - concurrent read/write operations

Quite surprisingly, the score are practically tied.

These tests show that, based on the access pattern to main memory, the ganged mode can be either faster or slower than unganged mode, and vice versa. However, it seems that unganged mode win more tests that it loses, so it seems the preferred memory controller configuration.