PGI SPEC® ACCEL results listed below were conducted by PGI on the dates listed.
Versions of the SPEC ACCEL 1.2 benchmarks are available for OpenACC, OpneMP, and OpenCL. Earlier SPEC ACCEL versions were available for OpenACC and OpenCL only. PGI has chosen to publish OpenACC results only. Please note that results for the OpenACC suite from version 1.2 cannot be compared to results using version 1.1 or 1.0.
SPEC ACCEL includes an optional Energy metric that measures the total energy consumed by the system during the run of the benchmark suite. On some tests, PGI has chosen to include this metric.
SPEC benchmarks include two fundamental metrics. "Base" performance requires the use of a uniform set of compiler options and runtime settings across all benchmarks in the suite. "Peak" performance allows use of arbitrary combinations of compiler options, runtime settings and profile-feedback optimization to maximize performance of individual benchmarks in the suite. PGI believes Base is most representative of "compile-and-go" performance, and it is for this reason that the comparisons on these pages list 64-bit Base performance only.
PGI uses SPEC benchmarks primarily to compare compiler performance over time. PGI does not perform direct comparative system peformance testing. To minimize performance fluctuations due to hardware variations, PGI retains a limited set of reference systems whose configurations are fixed and stable. As a general rule, reference systems are not modified or upgraded; instead new systems are added and old systems are retired from the reference system pool as needed. The results posted on this web site are organized by individual reference systems.
All SPEC results are a ratio compared to a reference system. Higher values are better.
SPEC ACCEL 1.2 OpenACC Results
PGI Reference System: "HSW8" (Intel Xeon E5-2698 v3)
|May-2017||PGI 17.5||8.02||--||NVIDIA Tesla P100 at 1328 MHz|
|May-2017||PGI 17.5||2.56||--||NVIDIA Tesla K40m at 745 MHz|
|May-2017||PGI 17.5||1.81||--||32 CPU cores only|
System Information:Supermicro SuperServer SYS-1028GR-TR with Intel Xeon E5-2698 v3 2.3GHz processor. The system has two 16-core chips and 256GB of memory (16 x 16 GB 2Rx4 PC4-2133P-R). Operating System: CentOS 7.2.1511 (64-bit). NVIDIA Tesla K40m GPU with 12MB ECC-enabled memory connected using PCIe 3.0 16x. NVIDIA Tesla P100 GPU with 16GB ECC-enabled memory connected using PCIe 3.0 16x. NVIDIA UNIX x86_64 Kernel Module 375.20.
SPEC ACCEL 1.1 OpenACC Results
PGI Reference System: "HSW2" (Intel Xeon E5-2698 v3 + NVIDIA Tesla K80c)
|Feb-2016||PGI 16.1||2.65||--||One GPU core at 823 MHz|
|Feb-2016||PGI 16.1||1.47||--||32 CPU cores only|
System Information:Supermicro SuperServer SYS-2028GR-TR with Intel Xeon E5-2698 v3 2.3GHz processor. The system has two 16-core chips and 256GB of memory (16 x 16 GB 2Rx4 PC4-2133P-R). NVIDIA Tesla K80C GPU with ECC enabled connected using PCIe 3.0 16x. Operating System: CentOS 7.1.1503 (64-bit). NVIDIA UNIX x86_64 Kernel Module 352.68.
PGI Reference System: "SBE02" (Intel Core i7-3930K + NVIDIA Tesla K40c)
|Feb-2016||PGI 16.1||2.46||--||Graphic clock frequency 745 MHz|
|Feb-2016||PGI 16.1||2.31||--||Graphic clock frequency 745 MHz; Using CUDA Unified Memory|
|Mar-2014||PGI 14.2||2.21||2.88||Default graphic clock frequency of 745 MHz|
|Mar-2014||PGI 14.2||2.47||3.03||Graphic clock frequency of 810 MHz|
|Mar-2014||PGI 14.2||2.59||3.01||Graphic clock frequency of 875 MHz|
System Information:ASUS motherboard with one Intel i7-3930 3.2GHz processor. The system has one six-core chip and 8GB of memory (2x4 GB PC3-14900R-9). NVIDIA Tesla K40C GPU with ECC enabled connected using PCIe 3.0 16x. Operating System: Red Hat Enterprise Linux 6.4 (64-bit). NVIDIA UNIX x86_64 Kernel Module 319.60.
SPEC® is registered trademark and SPEC Accel is a trademark of the Standard Performance Evaluation Corporation (SPEC).