luxas / benchmark

Dockerized C benchmarks for both ARM and amd64 hardware

luxas/benchmark

Extends [Roy Longbottoms Raspberry Pi benchmarks](http://www.roylongbottom.org.uk/Raspberry Pi Benchmarks.htm). This Docker image compiles and executes the most of them, and collects hand-picked data from the benchmarks. Then it runs a C++ script, which summarizes and calculates the benchmark data.

The reference point is 100 points and a Raspberry Pi 1 Model B should perform 100 in every test.

So, this way you may roughly measure how fast other computers or ARM boards are compared to a RPi 1. For the moment it benchmarks only single-core performance.

And seriously, you should take these benchmarks with a big grain of salt. But they're useful anyway.

Running the benchmarks

The images are pushed to Docker Hub, so it's easy to run directly:

$ docker run -it luxas/benchmark-arm
$ docker run -it luxas/benchmark-amd64

Building the images

For amd64:

$ make ARCH=amd64

For arm, from an amd64 host:

$ make ARCH=arm

Results

A results file will be in /bench/results.json inside the container.

Example output for Raspberry Pi 2:

{
	"overall": "223",
	"cpu": "150",
	"ram": "298",
	"real": "222",
	"subPerf": {
		"floatingPoint_whetstone": "126",
		"integerPerf_dhrystone": "116",
		"linearAlgebraSinglePoint_linpack": "182",
		"linearAlgebraDoublePoint_linpack": "185",
		"kernelDoublePrecision_liverloops": "142",
		"L1CacheMFLOPS_l1flops": "130",
		"L2CacheMFLOPS_l2flops": "318",
		"RAMCacheSpeed_ramflops": "388",
		"RAMMIPS_rammips": "511",
		"wordReadingBusSpeed_busspeed": "144",
		"compileTimeAllBenchmarks": "222"
	},
	"rawData": {
		"floatingPoint_whetstone": "296",
		"integerPerf_dhrystone": "1010",
		"linearAlgebraSinglePoint_linpack": "104",
		"linearAlgebraDoublePoint_linpack": "78",
		"kernelDoublePrecision_liverloops": "87",
		"L1CacheMFLOPS_l1flops": "121",
		"L2CacheMFLOPS_l2flops": "121",
		"RAMCacheSpeed_ramflops": "70",
		"RAMMIPS_rammips": "686",
		"wordReadingBusSpeed_busspeed": "989",
		"compileTimeAllBenchmarks": "57"
	}
}

Some notes:

• overall is the average value of cpu, ram and real
• cpu is the average value of floatingPoint_whetstone, integerPerf_dhrystone, linearAlgebraSinglePoint_linpack, linearAlgebraDoublePoint_linpack and kernelDoublePrecision_liverloops
• ram is the average value of L1CacheMFLOPS_l1flops, L2CacheMFLOPS_l2flops, RAMCacheSpeed_ramflops, RAMMIPS_rammips and wordReadingBusSpeed_busspeed
• real is the value of compileTimeAllBenchmarks
• subPerf is calculated values of how much better the machine did score compared to a RPi 1 Model B.
• rawData is the (sometimes calculated) raw output from Roy Longbottom benchmarks. Except for compileTimeAllBenchmarks, which represents how many seconds it took for the machine to compile the source code into binary files.

Contribute

If you like more precise benchmarks and like the idea of running the same benchmarks on ARM, amd64 and other architectures, feel free to create issues and pull requests.