Sunday 27th July 2014: 9.51pm. Link shared: http://arrayfire.com/arrayfire-on-nvidia-tegra-tk1
Some interesting benchmarks for my quad core nVidia Tegra K1 Jetson TK1 board @ 2.3Ghz fitted with a SSD SATA drive. I compare them, quite unfairly, to a quad core Intel Xeon CPU E3-1230 v3 @ 3.3GHz which is a Haswell unit:
$ sysbench --test=cpu --cpu-max-prime=8000 --num-threads=4 run
Tegra: 13.53 secs
Xeon: 1.64 secs
(the Xeon is 8.25x faster)
Tegra storage performance:
$ hdparm -t /dev/mmcblk0p1
Timing buffered disk reads: 208 MB in 3.01 seconds = 69.20 MB/sec
$ hdparm -t /dev/sda
Timing buffered disk reads: 590 MB in 3.00 seconds = 196.65 MB/sec
$ hdparm -t /dev/sdb
Timing buffered disk reads: 244 MB in 3.01 seconds = 80.99 MB/sec
(the first is for the eMMC and is about right, while the second should hit 240 MB/sec as it's a SATA II interface with SSD, and the third is a SSD connected via USB 3.0 and is very slow, something I've found is typical in ARM USB 3.0 implementations)
Compiling Boost.AFIO using GCC 4.8:
Tegra: 88 secs
Xeon: 25 secs
(the Xeon is 3.52x faster)
I don't need graphics, but according to http://arrayfire.com/arrayfire-on-nvidia-tegra-tk1/ the Tegra's GPU is about half the speed of a fairly ancient laptop GeForce GT650M, so a modern desktop PC graphics card would be about 10x faster. That's actually pretty good given a desktop PC graphics card will burn 20x the electricity. Incidentally the Tegra's GPU is about equal to low end Intel Haswell onboard graphics, the Intel HD 4600, which certainly makes you think and is vastly quicker than onboard graphics in pre-Haswell CPUs.