Nano PoW benchmark

This is a tool used for benchmarking Nano's proof of work generation on OpenCL devices.

Unlike other POW benchmarking tools, the hash rate of the Blake2b algorithm is measured directly, eliminating the random aspect of typical benchmarking solutions. This leads to much quicker benchmarks, while providing more accurate and reliable results.

Tests over 10 seconds in duration should provide reasonable accuracy, providing there are no resource-intensive background apps running.

This tool is only compatible with OpenCL-enabled hardware such as graphics cards. You may also need to install an OpenCL driver for your device if it isn't detected. For benchmarking incompatible devices, try using the facility included with the Nano Work Server.

How to use

Download the latest built jar, or clone the repository and build the project from source using Maven.

This program requires Java 8 (or above) to be installed on your system, and possibly an OpenCL driver.

You can run the program from the command line like so:

java -jar npowbench.jar [options...]

Command line options

Option	Description
`--device <platform:device>`	Specify OpenCL device ID and platform ID
`--duration <seconds>`	Benchmark duration in seconds
`--threads <threads>`	Thread count (global work size)
`--local-work-size <size>`	Set the OpenCL local work size
`--difficulty <diff>`	Output calculations for the specified difficulty threshold (as hexadecimal) Multiple difficulties may be provided
`--kernel <variant>`	Specify which nano-node kernel variant to use Supported versions: NN1, NN2
`--kernel-file <path>`	Specify a custom OpenCL kernel program file The program must be compatible with `nano_work`

Example

Command

java -jar npowbench.jar --device 0:0 --duration 15 --difficulty fffffe0000000000

Output

=========================================================================
                      NANO PROOF-OF-WORK BENCHMARK
             https://github.com/koczadly/nano-pow-benchmark/
=========================================================================

Benchmark parameters:
   Device:                     Intel(R) HD Graphics 530 (#0)
   Platform:                   Intel(R) OpenCL HD Graphics (#0)
   Threads (global work size): 1,048,576
   Local work size:            Default
   Kernel program:             nano-node Blake2b V2 (as of ff424af)
   Kernel function:            nano_work(attempt, result, item, diff)

Running benchmark for 15 seconds... Please wait...

=========================================================================
                            BENCHMARK RESULTS
=========================================================================
Benchmark measurements:
   Total time elapsed:     15.009 s
   Total computation time: 15.008771 s
   Batch computation time: 33.502 ms
   Hash rate:              31.299 MH/s
   Total computed hashes:  469,762,048

Expected performance for difficulty thresholds:
   fffffe0000000000 [receive]: 3.7312 work/s (268.014 ms/work)
   fffffff800000000 [send]:    0.0583 work/s (17.153 s/work)
=========================================================================

koczadly / nano-pow-benchmark Goto Github PK