X

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

SPARC T7-1 Delivers 1-Chip World Records for SPEC CPU2006 Rate Benchmarks

Brian Whitney
Principal Software Engineer

This page has been updated on November 19, 2015. The SPARC T7-1 server results have been published at www.spec.org.

Oracle's SPARC T7-1 server delivered world record SPEC CPU2006 rate benchmark results for systems with one chip. This was accomplished with Oracle Solaris 11.3 and Oracle Solaris Studio 12.4 software.

  • The SPARC T7-1 server achieved world record scores of 1200 SPECint_rate2006, 1120 SPECint_rate_base2006, 832 SPECfp_rate2006, and 801 SPECfp_rate_base2006.

  • The SPARC T7-1 server beat the 1 chip Fujitsu CELSIUS C740 with an Intel Xeon Processor E5-2699 v3 by 1.7x on the SPECint_rate2006 benchmark.  The SPARC T7-1 server beat the 1 chip NEC Express5800/R120f-1M with an Intel Xeon Processor E5-2699 v3 by 1.8x on the SPECfp_rate2006 benchmark.

  • The SPARC T7-1 server beat the 1 chip IBM Power S812LC server with a POWER8 processor by 1.9 times on the SPECint_rate2006 benchmark and by 1.8 times on the SPECfp_rate2006 benchmark.

  • The SPARC T7-1 server beat the 1 chip Fujitsu SPARC M10-4S with a SPARC64 X+ processor by 2.2x on the SPECint_rate2006 benchmark and by 1.6x on the SPECfp_rate2006 benchmark.

  • The SPARC T7-1 server improved upon the previous generation SPARC platform which used the SPARC T5 processor by 2.5 on the SPECint_rate2006 benchmark and by 2.3 on the SPECfp_rate2006 benchmark.

The SPEC CPU2006 benchmarks are derived from the compute-intensive portions of real applications, stressing chip, memory hierarchy, and compilers.  The benchmarks are not intended to stress other computer components such as networking, the operating system, or the I/O system.  Note that there are many other SPEC benchmarks, including benchmarks that  specifically focus on Java computing, enterprise computing, and network file systems.

Performance Landscape

Complete benchmark results are at the SPEC website. The tables below provide the new Oracle results, as well as select results from other vendors.

Presented are single chip SPEC CPU2006 rate results. Only the best results published at www.spec.org per chip type are presented (best Intel, IBM, Fujitsu, Oracle chips).

SPEC CPU2006 Rate Results – One Chip
System Chip Peak Base
  SPECint_rate2006
SPARC T7-1 SPARC M7 (4.13 GHz, 32 cores) 1200 1120
Fujitsu CELSIUS C740 Intel E5-2699 v3 (2.3 GHz, 18 cores) 715 693
IBM Power S812LC POWER8 (2.92 GHz, 10 cores) 642 482
Fujitsu SPARC M10-4S SPARC64 X+ (3.7 GHz, 16 cores) 546 479
SPARC T5-1B SPARC T5 (3.6 GHz, 16 cores) 489 441
IBM Power 710 Express POWER7 (3.55 GHz, 8 cores) 289 255
  SPECfp_rate2006
SPARC T7-1 SPARC M7 (4.13 GHz, 32 cores) 832 801
NEC Express5800/R120f-1M Intel E5-2699 v3 (2.3 GHz, 18 cores) 474 460
IBM Power S812LC POWER8 (2.92 GHz, 10 cores) 468 394
Fujitsu SPARC M10-4S SPARC64 X+ (3.7 GHz, 16 cores) 462 418
SPARC T5-1B SPARC T5 (3.6 GHz, 16 cores) 369 350
IBM Power 710 Express POWER7 (3.55 GHz, 8 cores) 248 229

The following table highlights the performance of the single-chip SPARC M7 processor based server to the best published two-chip POWER8 processor based server.

SPEC CPU2006 Rate Results
Comparing One SPARC M7 Chip to Two POWER8 Chips
System Chip Peak Base
  SPECint_rate2006
SPARC T7-1 1 x SPARC M7 (4.13 GHz, 32core) 1200 1120
IBM Power S822LC 2 x POWER8 (2.92 GHz, 2x 10core) 1100 853
  SPECfp_rate2006
SPARC T7-1 1 x SPARC M7 (4.13 GHz, 32 cores) 832 801
IBM Power S822LC 2 x POWER8 (2.92 GHz, 2x 10core) 888 745

Configuration Summary

System Under Test:

SPARC T7-1
1 x SPARC M7 processor (4.13 GHz)
512 GB memory (16 x 32 GB dimms)
800 GB on 4 x 400 GB SAS SSD (mirrored)
Oracle Solaris Studio 12.4 with 4/15 Patch Set

Benchmark Description

SPEC CPU2006 is SPEC's most popular benchmark. It measures:

  • Speed — single copy performance of chip, memory, compiler
  • Rate — multiple copy (throughput)

The benchmark is also divided into integer intensive applications and floating point intensive applications:

  • integer: 12 benchmarks derived from applications such as artificial intelligence chess playing, artificial intelligence go playing, quantum computer simulation, perl, gcc, XML processing, and pathfinding
  • floating point: 17 benchmarks derived from applications, including chemistry, physics, genetics, and weather.

It is also divided depending upon the amount of optimization allowed:

  • base: optimization is consistent per compiled language, all benchmarks must be compiled with the same flags per language.
  • peak: specific compiler optimization is allowed per application.

The overall metrics for the benchmark which are commonly used are:

  • SPECint_rate2006, SPECint_rate_base2006: integer, rate
  • SPECfp_rate2006, SPECfp_rate_base2006: floating point, rate
  • SPECint2006, SPECint_base2006: integer, speed
  • SPECfp2006, SPECfp_base2006: floating point, speed

Key Points and Best Practices

  • Jobs were bound using the "pbind" command.

See Also

Disclosure Statement

SPEC and the benchmark names SPECfp and SPECint are registered trademarks of the Standard Performance Evaluation Corporation.  Results as of November 19, 2015 from www.spec.org.

SPARC T7-1: 1200 SPECint_rate2006, 1120 SPECint_rate_base2006, 832 SPECfp_rate2006, 801 SPECfp_rate_base2006; SPARC T5-1B: 489 SPECint_rate2006, 440 SPECint_rate_base2006, 369 SPECfp_rate2006, 350 SPECfp_rate_base2006; Fujitsu SPARC M10-4S: 546 SPECint_rate2006, 479 SPECint_rate_base2006, 462 SPECfp_rate2006, 418 SPECfp_rate_base2006. IBM Power 710 Express: 289 SPECint_rate2006, 255 SPECint_rate_base2006, 248 SPECfp_rate2006, 229 SPECfp_rate_base2006; Fujitsu CELSIUS C740: 715 SPECint_rate2006, 693 SPECint_rate_base2006; NEC Express5800/R120f-1M: 474 SPECfp_rate2006, 460 SPECfp_rate_base2006; IBM Power S822LC: 1100 SPECint_rate2006, 853 SPECint_rate_base2006, 888 SPECfp_rate2006, 745 SPECfp_rate_base2006; IBM Power S812LC: 642 SPECint_rate2006, 482 SPECint_rate_base2006, 468 SPECfp_rate2006, 394 SPECfp_rate_base2006.

Join the discussion

Comments ( 1 )
Please enter your name.Please provide a valid email address.Please enter a comment.CAPTCHA challenge response provided was incorrect. Please try again.