## Friday, April 17, 2015

### Get the max CPU performance for scientific computing : Intel Xeon height cores and more ("Haswell-EP")

Haswell is the codename for the Intel processor microarchitecture that is the successor to the Ivy Bridge microarchitecture. Most Haswell products are branded as 4th Generation Intel® Core™ Processors for client systems, and Intel® Xeon® v3 Processors for server systems.
Intel officially announced processors with this microarchitecture in 2013.

List:
http://ark.intel.com/products/codename/42174/Haswell#@All

Example: 8cores
http://ark.intel.com/products/82766
http://ark.intel.com/products/82767

Datasheet
http://www.intel.com/content/www/us/en/processors/xeon/xeon-e5-v3-datasheet-vol-1.html

One of the best choice Price/power:
Intel® Xeon® Processor E5-1660 v3  (3.00 GHz)
L2 cache=8 × 256 KB
L3 cache=20 MB

Electrical Specifications
C-states (W= Watt):
Model Number TDP                  C1E (W) 2 C3 (W) 2 C6 (W)
E5-1660 v3       140W 8-Core   34               25            12

Idle States (C-states) are used to save power when the processor is idle. C0 is the operational state, meaning that the CPU is doing useful work. C1 is the first idle state, C2 the second, and so on, where more power saving actions are taken for numerically higher C-states.

 # of Cores 8 # of Threads 16 Processor Base Frequency 3 GHz Max Turbo Frequency 3.5 GHz TDP 140 W
 TCASE 65.9°C
Case Temperature is the maximum temperature allowed at the processor Integrated Heat Spreader (IHS).

First to get the max CPU performance, it is necessary to

1. manage the Power Management States. https://software.intel.com/en-us/articles/power-management-states-p-states-c-states-and-package-c-states,. control C-state power. When the Temperature increases, above 70°C the performance must decrease!
2. always work at turbo Frequency (no switch) then it is very important to control the Temperature of CPU (and also of DDR RAM). For the E5-1660 v3, it's 3.5GHz. Typically the CPU temperature is around 55-60°C (benchmark with LAMMPS dynamics of 800 000atoms/5000timesteps).  T must be below 65°C.

 Intel® Turbo Boost Technology ‡ 2.0 Intel® vPro Technology ‡ Yes Intel® Hyper-Threading Technology ‡ Yes Intel® Virtualization Technology (VT-x) ‡ Yes Intel® Virtualization Technology for Directed I/O (VT-d) ‡ Yes Intel® VT-x with Extended Page Tables (EPT) ‡ Yes Intel® TSX-NI No Intel® 64 ‡ Yes Idle States Yes Enhanced Intel SpeedStep® Technology Yes Intel® Demand Based Switching Yes Thermal Monitoring Technologies Yes Intel® Flex Memory Access No Intel® Identity Protection Technology ‡ Yes

Intel® Turbo Boost Technology dynamically increases the processor's frequency as needed by taking advantage of thermal and power headroom to give you a burst of speed when you need it, and increased energy efficiency when you don’t. But the problem is the "switch" between electrical/thermal systems and performance.

Intel® Hyper-Threading Technology (Intel® HT Technology) delivers two processing threads per physical core. Highly threaded applications can get more work done in parallel, completing tasks sooner.
THE HYPER-THREADING must be disabled in the domain of scientific computation (like LAMMPS): only one thread/core.

Intel® Demand Based Switching is a power-management technology in which the applied voltage and clock speed of a microprocessor are kept at the minimum necessary levels until more processing power is required. This technology was introduced as Intel SpeedStep® Technology in the server marketplace.

Thermal Monitoring Technologies protect the processor package and the system from thermal failure through several thermal management features. An on-die Digital Thermal Sensor (DTS) detects the core's temperature, and the thermal management features reduce package power consumption and thereby temperature when required in order to remain within normal operating limits