Benchmark Stresses Big Chips: Page 2 of 3

February 28, 2015 //By Rick Merritt
Benchmark Stresses Big Chips
The Embedded Microprocessor Benchmark Consortium (EEMBC), a trade group of 47 chip and system designers, released CoreMark-Pro. The suite of benchmarks for 32- and 64-bit processors expands the original CoreMark, a single performance test released in 2009 for microcontrollers and processors.
The CoreMark software performs self-verification of tests and a process to help users submit scores to the EEMBC Web site. A database of published results of CoreMark-Pro and AndEBench-Pro results are already available online. The Web site lets users sort and compare results across a wide variety of factors and products.

EEMBC provided a sample of CoreMark-Pro results to date from various vendors. For better resolution click here.

Among CoreMark-Pro results currently posted, “clearly Nvidia is at the top by a fairly significant margin,” with its Tegra K1 in the Google Nexus 9 and Xiaomi handsets, said Markus Levy, president of EEMBC. Samsung’s latest Exynos and current versions of Qualcomm’s Snapdragon came in second and third, respectively.

Levy was quick to note many factors influence the tests, such as high memory bandwidth on the Nvidia chip. And the same chip can get different scores when running on different systems with different memory configurations and options.

CoreMark-Pro “probably won’t have as much use as the original CoreMark, which was simple to download, build and run on anything,” said Levy. The complexity of the Pro suite, which can take up to 3 Mbytes of memory, means “probably only qualified engineers will use it,” on higher-end processors…“some MCUs can’t even run it,” he said.

The original CoreMark was downloaded 13,000 times and 500 scores using it have been posted. They include tests of chips from Atmel, Freescale, Imagination, Microchip, NXP, Renesas, and STMicroelectronics.


Vous êtes certain ?

Si vous désactivez les cookies, vous ne pouvez plus naviguer sur le site.

Vous allez être rediriger vers Google.