As previously written, DigiCortex Engine since the version v0.99a comes with performance monitoring / benchmarking support and optimizations for the Intel(R) Ivy Bridge EP / (upcoming) EX Xeon(TM) CPUs (codename "IvyTown"). In addition, it implements performance fixes for systems with number of CPUs not equal to power of 2 (6/10/12/24, etc.).
With this change, DigiCortex engine is ready for the upcoming high-core-count Intel(R) Xeon(TM) processors which are going to significantly increase the number of cores available for entry-level and mid-range 2S and 4S server configurations as well as to significantly upgrade the expandable 8S server segment.
Since the Ivy Bridge EP platform is now public, we report the performance increase attained with the new dual Xeon 2697 v2, compared to the previous generation (Sandy Bridge EP) dual Xeon 2687W setup. Ivy Bridge EP setup brings 50% more cores (24 compared to 16) but due to TDP limit of 130W per CPU, they are clocked lower (2.7 GHz nominal frequency, 3 GHz maximum all-core turbo).
Nevertheless, DigiCortex engine scales almost perfectly and manages to attain ~1.45x speed-up compared to Intel Jaketown (SNB-EP) set-up.
This, in itself, is pretty remarkable as keeping 24 cores busy 100% is no small feat. Here are some quick benchmark results:
Test Description: 32768 Multi-Compartment Neurons, 1.8 million Synapses (2 receptors each)
|Platform Name:||Num. CPUs||Num. Cores||CPU Speed||RAM Speed||Avg. Simulation Performance|
|Dual Intel Xeon E5-2697 v2
("Ivy Bridge EP")
|2||24||2700 MHz||DDR3 2133 MHz||1.710x|
|Dual Intel Xeon E5-2687W
("Sandy Bridge EP")
|2||16||3100 MHz||DDR3 2133 MHz||1.170x|
Below you can find the snapshot of SpikeBenchmark running the simulation of 4 million neurons with 800 million synapses in 1/100 of real-time on dual Xeon 2697 v2 setup: