• Download this, rename it to linpack.c Build email protected:/benchmarks# cc -Ofast -o linpack linpack.c -lm -mcpu=cortex-a8 -march=armv7-a -mfpu=neon -mfloat-abi=hard -funsafe-math-optimizations -fno-fast-math linpack.c: In function ‘main’: linpack.c:78:14: warning: ignoring return value of ‘fgets’, declared with attribute warn.
• High Performance Linpack (HPL) is a software package that solves a (random) dense linear system in double precision (64 bits) arithmetic on distributed-memory computers. It can thus be regarded as a portable as well as freely available implementation of the High Performance Computing Linpack Benchmark.
• Download Linpack and enjoy it on your iPhone, iPad, and iPod touch. ‎Linpack Benchmark The LINPACK Benchmarks are a measure of a system's floating point computing power. They measure how fast a computer solves a dense N by N system of linear equations Ax = b, which is a common task in engineering.
• The description of Linpack Standard Linpack Benchmark App The LINPACK Benchmarks are a measure of a system’s floating point computing power. Introduced by Jack Dongarra, they measure how fast a computer solves a dense N by N system of linear equations Ax = b, which is a common task in engineering.

The HPC Challenge benchmark consists of basically 7 tests: HPL - the Linpack TPP benchmark which measures the floating point rate of execution for solving a linear system of equations. DGEMM - measures the floating point rate of execution of double precision real matrix-matrix multiplication. The High Performance Conjugate Gradients (HPCG) Benchmark project is an effort to create a new metric for ranking HPC systems. HPCG is intended as a complement to the High Performance LINPACK (HPL) benchmark, currently used to rank the TOP500 computing systems.

This page is not too useful as is, and needs to be fully split/removed/reworked to feature in the modern device->components centric wiki logic.

A10 Benchmarks

CPU

Linpack

Download this[1], rename it to linpack.c

Build

Results

-mcpu=cortex-a8 -march=armv7-a -mfpu=neon -mfloat-abi=hard -funsafe-math-optimizations -fno-fast-math

mcpu=cortex-a8 -mtune=cortex-a8 -march=armv7-a -mfpu=neon -mfloat-abi=hard -funsafe-math-optimizations -fomit-frame-pointer -ffast-math -funroll-loops -funsafe-loop-optimizations

Whetstone/Dhrystone

http://www.roylongbottom.org.uk/linux%20benchmarks.htm (requires File:Classic benchmarks.patch)

Building

Results

./whets (gcc-4.7 -static -O3 -mcpu=cortex-a8 -mtune=cortex-a8 -mfpu=neon -funroll-loops)

./dhry1 (gcc-4.7 -static -O3 -mcpu=cortex-a8 -mtune=cortex-a8 -mfpu=neon -funroll-loops)

./dhry2 (gcc-4.7 -static -O3 -mcpu=cortex-a8 -mtune=cortex-a8 -mfpu=neon -funroll-loops)

Adding -Ofast and -flto:

./whets (gcc-4.7 -static -Ofast -mcpu=cortex-a8 -mtune=cortex-a8 -mfpu=neon -funroll-loops -flto)

./dhry1 (gcc-4.7 -static -Ofast -mcpu=cortex-a8 -mtune=cortex-a8 -mfpu=neon -funroll-loops -flto)

./dhry2 (gcc-4.7 -static -Ofast -mcpu=cortex-a8 -mtune=cortex-a8 -mfpu=neon -funroll-loops)

OpenSSL

How to test

run

Results

Linaro-alip soft-float

ArchLinux-ARM hard-float

SciMark

Build

Results

nbench

build

results

CC=gcc-4.7

CFLAGS=-s -static -Wall -O3 -mfpu=neon -mcpu=cortex-a8 -mtune=cortex-a8 -fomit-frame-pointer -marm -funroll-loops

CC=gcc-4.7

CFLAGS=-s -static -Wall -Ofast -mfpu=neon -mcpu=cortex-a8 -mtune=cortex-a8 -fomit-frame-pointer -marm -funroll-loops

Linux kernel build

setup

tests

OpenBenchmark Phoronix Test Suite

Comparisons with Debian and Raspian on r-Pi vs. Cubieboard 1 and 2http://openbenchmarking.org/result/1308083-UT-1302242BY19http://openbenchmarking.org/result/1308084-UT-1301189RA85

GPU

Results for X11 libraries and framebuffer libraries may differ.

ioquake3

/call-of-duty-dmg-free-download.html. See ioquake3

es2_gears

X11 libraries:

• 131FPS

• r3p0: 195-200 FPS
• r3p0: 58-75 FPS - fullscreen (1024x768)

Framebuffer libraries:?

glx_gears

X11 libraries + mesa:

• 117 FPS
• ~25 FPS - fullscreen (1024x768)

glmark2-es2

X11 libraries:

Video decoding

See CedarXVideoRenderingChart

IO

SATA

This may be limited by the comparatively old, cheap SSD being used.

SD Card

NAND

Ethernet

Power consumption

A13 Benchmarks

A13 needs own CPU benchmarks because DDR3 bus is crippled.

nbench

Tested on A13-olinuxino, debian wheezy.

CC=gcc-4.7CFLAGS= -s -static -O3 -mfpu=neon -mcpu=cortex-a8 -mtune=cortex-a8 -fomit-frame-pointer -marm -munroll-loops

A10S Benchmarks

Should be the same as A13.

A20 Benchmarks

CPU

OpenSSL

See Full List On Software.intel.com

Linpack

Compile:

Results

Note: The linpack results suggest that the floating performance of the Cortex A7 core in the A20 is signficantly faster (up to 3x more KFLOPS) than the Cortex A8 used in the A10. Running a flioating-point intensive application (3D geometry processing) seems to confirm that the A20 is significantly faster. https://high-powercomfort348.weebly.com/blog/arma-2-operation-arrowhead-cd-key-generator.

Linpack Benchmark Download

lmbench

lmbench (lmbench-3.0-a9) is an older and not very well known benchmark,but can provides some interesting low-level architectural details aswell as memory speed benchmarks.

The reported latency and parallelism is not consistent between runs,but the L1 data cache and L2 cache size is probably correct. The L2 cachesize is not large, but that is not surprising for a 55nm manufactured SoC(modern higher-end ARM socs manufactured at 28nm for smartphones andtablets have up to 2MB of L2 cache; RK3188 has 512KB L2 cache).

This gives interesting info about the CPU instruction characteristics ofthe ARM Cortex A7 core used in the A20.

• As with most ARM architectures, integer divide slow.
• Integer multiply is relatively fast.
• Floating performance is OK, and single precision (float) is faster than double precision (double).

Memory latencies:

This confirms the cache sizes of the A20, the L1 (data) cache size is 32KB,latency seems to about 3ns, and as buffer size approaches 32KB the latencyreported by the test starts to increase due to cache associativity effects(cache line conflicts). A similar transition is seen for the L2 cache,latency is about 9.3ns, and latency starts to increase as the buffer sizeapproaches 256KB. DRAM latency is about 60ns on the test configuration(1008 MHz CPU, 432 MHz DRAM clock, 432 MHz MBUS clock, 6 cycle CAS timing).With 9 cycle CAS timing latency at 16MB buffer size is 63.8ns.

Memory bandwidth:

Each core has its own L1 data cache, so the L1 cache bandwidth doubles with two cores active.WIth two cores active, the shared L2 cache bandwidth (64K buffer result above) increases compared to whenusing only one core except in the case of copy when one core already saturates the L2 cache bandwidth.With DRAM access (16M buffer), two active cores are able to utilize more DRAM bandwidth compared towhen using only one core (may depend on the lmbench implementation, but is probably a good sign formulti-tasking performance). With a CAS timing of 9 cycles instead of 6, performance is only slightly slower.However, with lower DRAM or MBUS clock, performance will be lower.

GPU

The Mali-400MP2 GPU in the A20 has two pixel processors instead of only one in the A10 (Mali-400MP). Because of that especially fillrate (i.e. high resolution) performance should be higher with proper drivers. You need the newer Mali r3p2 drivers (standard is r3p0) to really take advantage of the improved features of the Mali-400MP2. See the section Optimizing system performance for advanced instructions for using the r3p2 Mali drivers.

Linpack Benchmark Download Free

The following benchmarks were performed with 1280x720 60 Hz HDMI output (32bpp). The window size of of glmark2 is the default 800x600. The device has the memory clock set to 408 MHz (which is lower than some other devices which may impact performance). The CPU governor was set to ondemand with custom settings. The SwapbuffersWait option was set to 'false' in the xorg.conf to eliminate the effect of vsync. The fb0_framebuffer_num in script.bin was set to 3 so that xf86-video-fbturbo can optimally provide Mali GLES integration.

The version of glmark2 (glmark2-es2) used is 2013.08.07. Source: https://github.com/ssvb/glmark2.git. Configure with

You might need to apply a patch like this to the GLES header files for a clean compile:

Performance with standard Mali r3p0 drivers/kernel:

Performance with Mali r3p2 drivers/kernel:

Note that several sub-benchmarks show doubled performance or better. This includes 2D compositing and fillrate-limited tests.

Seagull bartender serial number. After tweaking the memory controller parameters (DRAM frequency 432 MHz, MBUS frequency 432 MHz, CAS timing 6 cycles) the gl2mark2 score increases to 189.

Another benchmark of r3p2 drivers, done on Cubietruck, with r3p2 kernel module, binary drivers and patched xf86-video-fbturbo:

Intel Linpack Benchmark Download