CPU: L5518 No Of Processors: 2 ie No of Cores is : 4 + 4 ie No of threads with hyperthreading is: 8 + 8 = 16. RAM: 100GB
The specification page( http://ark.intel.com/Product.aspx?id=40727 ) says that the maximum memory throughput is 25.6GB/s for one CPU of 4 cores. I wanted to measure the maximum memory throughput of my system which is having 8 cores because I've 2 L5518s.
The test I done is,
Launched 16 threads and set each thread's affinity to unique CPU number. Allocated 100MBX2 memory in each of these threads. Done write to one 100MB. And copy to the other 100MB. Thus, write-read-write.
But the maximum throughput I'm getting is only around 25.5GB/s(Measured with PTU). I expected 50GB/s because I'm having two L5518 processors.
25.6Gb/s is the max throughput. real throughput depends on channels count, bus and memory frequency, latency. large amount of core couldmake max workload, there are no linear increasing of the memorythroughput
try to measure usingbanchmarking tools like sisoft sandra. there is no any idea until you show your code
For example, we can't see from here whether you set your BIOS to NUMA, if it has such a setting, and whether you set affinity appropriately. You can't expect to beat everyone else by30% without even thinking about your method.