07.12.2015 Views

2H 2015

intel-xeon-phi-sw-ecosystem-guide-2h-2015-public3

intel-xeon-phi-sw-ecosystem-guide-2h-2015-public3

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Comparative Performance<br />

Monte Carlo *<br />

European Option Pricing<br />

30<br />

25<br />

20<br />

15<br />

10<br />

5<br />

0<br />

Intel® Xeon® processor E5-2697 v2 + Java* managed code<br />

Intel® Xeon® processor E5-2697 v2 + C/C++ native<br />

Intel® Xeon Phi coprocessor 7120A + C/C++ native<br />

Intel® Xeon Phi coprocessor 7120A + Hadoop*<br />

Java: version 8 update 30+ C/C++: Intel® Parallel Composer XE 15.0<br />

For configuration details, go here.<br />

Monte Carlo European Option Pricing Speed Up<br />

0.69X<br />

1<br />

8.91X<br />

Hadoop: Cloudera* Distribution of Hadoop 5.3+<br />

31.25X<br />

SOURCE: INTEL MEASURED RESULTS AS OF MARCH, <strong>2015</strong><br />

1 NODE<br />

Application: Monte Carlo European option<br />

Description: Implements European Option Pricing using Monte Carlo. It<br />

compares the performance of<br />

1) Java* code, 2) C/C++ native code, 3) C/C++ offload accelerated code,<br />

4) C/C++ accelerated code using Hadoop*.<br />

Availability:<br />

• Code and Recipe: Available here.<br />

APPROVED FOR PUBLIC PRESENTATION<br />

NEW<br />

Usage Model: Java (Managed Offload), Native on the host, Native and<br />

Accelerators, Hadoop on native and Accelerated.<br />

Highlights:<br />

• Java code can use parallelism but cannot vectorize any control flows<br />

• Native C/C++ code can take advantage of both vectorization &<br />

parallelization<br />

• Native Accelerated code offloads the whole workload to Intel® Xeon<br />

Phi Coprocessor.<br />

• Hadoop distribute the workload to 4 remote modes using mapreduce<br />

• Remote nodes accelerate the workload and send the run result back<br />

to the head note.<br />

Results:<br />

• Intel Java Stream can parallelize the workload but not vectorize<br />

• Native interface bring vectorization and parallelization to the<br />

workload<br />

• Acceleration extends the parallelism from Multicore to manycore<br />

• Hadoop map reduce can further distribute the application into 4<br />

nodes and achieve up to 31X improvement.<br />

Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems,<br />

components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated<br />

purchases, including the performance of that product when combined with other products. For more information go to http://www.intel.com/performance *Other names and brands may be claimed as the property of others<br />

61

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!