Real world tests
Well, we now have a quite clear understand of the pros and cons of both memory controller configurations. But what about some real world, server oriented tests? I run some benchmarks for two common scenario: a web server and a database (MySQL) server.
Real wolrd tests: Apache simple test
Web server is one of the most common server role. How the ganged vs unganged question relates to this kind of workload? This test measures the performances of the system when simple, static pages are required:
With four concurrent threads, the unganged mode win by about 4%. Let's try to use 256 threads:
While the unganged mode remain the leader, its advantage now is only ~2%. This simple test probably does not load the main memory very much (here, Apache has a clear opportunity to run from the caches), and the unganged mode give us only a very small boost, if any (I consider a 2-4% difference within the margin of error).