» Current | 2017 | 2016 | 2015 | Subscribe

Linley Newsletter

SPEC CPU2017 Updates Benchmarks

August 15, 2017

Author: David Kanter

CPU2017 is the newest iteration of SPEC’s industry-standard suite of compute-performance benchmarks. It updates the rules, subtests, and scoring of CPU2006 to reduce run times and more accurately measure the performance of modern processors. Over the last 11 years, computer hardware and software has evolved substantially—for example, multiple cores have become the norm and applications use far more memory. Refreshing a benchmark ensures consistency with current practices while also removing broken tests that no longer reflect real performance.

SPEC CPU2017 follows in the footsteps of its predecessors: it continues to stress the processor, memory hierarchy, and compilers for different platforms, and it comprises a suite of tests that derive from real-world applications. The SPECspeed latency test and the SPECrate throughput test now operate on different data. The former requires 16GB of physical memory (compared with just 1GB in the prior generation); the latter requires 1–2GB of memory per instance, and most submissions have one instance per hardware thread (roughly double the number in CPU2006). These requirements are ideal for comparing the compute performance of notebooks, desktops, workstations, servers, and high-end embedded systems, including ARM-based processors. Smartphones, tablets, and other small devices must use different benchmarks and will continue to unofficially employ CPU2006.

The new SPECspeed tests can benefit from both OpenMP and compiler-driven autoparallelization, which allow a single thread to run on multiple cores; SPECrate tests prohibit these techniques because they typically assign at least one instance per core. Therefore, many customers may prefer to look at SPECrate per core or SPECrate per thread in addition to the official speed and rate metrics.

Subscribers can view the full article in the Microprocessor Report.

Subscribe to the Microprocessor Report and always get the full story!

Purchase the full article

Events

Linley Processor Conference 2017
Covers processors and IP cores used in deep learning, embedded, communications, automotive, IoT, and server designs.
October 4 - 5, 2017
Hyatt Regency, Santa Clara, CA
More Events »

Newsletter

Linley Newsletter
Analysis of new developments in microprocessors and other semiconductor products
Subscribe to our Newsletter »