X

Everything you want and need to know about Oracle SPARC systems performance

Neural Network Models Using Oracle R Enterprise: SPARC T7-4 Beats 4-Chip x86 E7 v3

Brian Whitney
Principal Software Engineer

Oracle's SPARC T7-4 server executing neural network algorithms using Oracle R Enterprise (ORE) is up to two times faster than a four-chip x86 E7 v3 server.

  • For a neural network with two hidden layers, 10-neuron with 5-neuron hyperbolic tangent, the SPARC T7-4 server is 1.5 times faster than a four-chip x86 T7 v3 server on calculation time.

  • For a neural network with two hidden layers, 20-neuron with 10-neuron hyperbolic tangent, the SPARC T7-4 server is 2.0 times faster than than a four-chip x86 T7 v3 server on calculation time.

Performance Landscape

Oracle Enterprise R Statistics in Oracle Database
(250 million rows)
Neural Network
with Two Hidden Layers
Elapsed Calculation Time SPARC Advantage
4-chip x86 E7 v3 SPARC T7-4
10-neuron + 5-neuron
hyperbolic tangent
520.1 (sec) 337.3 (sec) 1.5x
20-neuron + 10-neuron
hyperbolic tangent
1128.4 (sec) 578.1 (sec) 2.0x

Configuration Summary

SPARC Configuration:

SPARC T7-4
4 x SPARC M7 processors (4.13 GHz)
2 TB memory (64 x 32 GB dimms)
Oracle Solaris 11.3
Oracle Database 12c Enterprise Edition
Oracle R Enterprise 1.5
Oracle Solaris Studio 12.4 with 4/15 patch set

x86 Configuration:

Oracle Server X5-4
4 x Intel Xeon Processor E7-8895 v3 (2.6 GHz)
512 GB memory
Oracle Linux 6.4
Oracle Database 12c Enterprise Edition
Oracle R Enterprise 1.5

Storage Configuration:

Oracle Server X5-2L
2 x Intel Xeon Processor E5-2699 v3
512 GB memory
4 x 1.6 TB 2.5-inch NVMe PCIe 3.0 SSD
2 x Sun Storage Dual 16Gb FC PCIe HBA
Oracle Solaris 11.3

Benchmark Description

The benchmark is designed to run various statistical analyses using Oracle R Enterprise (ORE) with historical aviation data.  The size of the benchmark data is about 35 GB, a single table holding 250 million rows. One of the most popular algorithms, neural network, has been used against the dataset to generate comparable results.

The neural network algorithms support various features. In this workload, the following two neural network features have been used: neural net with two hidden layers 10-neuron with 5-neuron hyperbolic tangent and neural net with two hidden layers 20-neuron with 10-neuron hyperbolic tangent.


See Also

Disclosure Statement

Copyright 2015, Oracle and/or its affiliates. All rights reserved.  Oracle and Java are registered trademarks of Oracle and/or its affiliates. Other names may be trademarks of their respective owners. Results as of 25 October 2015.

Be the first to comment

Comments ( 0 )
Please enter your name.Please provide a valid email address.Please enter a comment.CAPTCHA challenge response provided was incorrect. Please try again.Captcha
Oracle

Integrated Cloud Applications & Platform Services