Raspberry Pi 2: Raspbian vs. Linaro (ARMv6 vs ARMv7)

NBench  2.2.3

nbench Integer FP
Raspbian Pi2B 19.80 8.49
Linaro Pi2B 21.0 8.09
Raspberry Pi B+ 11.55 3.88
MIPS CI20 18.08 3.88
Banana Pro 20.23 8.67
ODROID-C1 30.20 10.50

Results are an index relative to a Pentium 90 with 256KB L2 cache.

The Linaro integer results are about 5% ahead, but the floating point performance is actually about 5% behind.

Win some, lose some.

Unix Bench 5.1.3

Benchmark # cores Dhrystone Whetstone Hanoi
Raspbian Pi2B 1 253.8 90.1 34122.8
Raspbian Pi2B 2 506.3 179.4 67640.2
Raspbian Pi2B 4 1006.6 358.7 135452.7
Linaro Pi2B 1 247.9 75.80 31748.40
Linaro Pi2B 2 496.1 150.70 63303.50
Linaro Pi2B 4 988.4 301.30 100654.10
Raspberry Pi B+ 1 142.7 48.9 18790.9
MIPS CI20 1 197.8 58.6 23839.4
MIPS CI20 2 394.5 117.2 47674.7
Banana Pro 1 248.7 89.6 33920.3
Banana Pro 2 490.0 178.0 67205.1
ODROID-C1 1 348.4 113.9 40639.1
ODROID-C1 2 680.6 225.6 81299.3
ODROID-C1 4 1174.3 443.2 148415.5

Results are an index relative to a SPARCstation 20-61 (rated at 10.0)

Linaro is very slightly slower for Dhrystone, about 16% slower for whetstone,  and about 10% slower for hanoi.

hdparm

hdparm media cached read buffered read
Raspberry Pi 2B ADATA 391.1 17.5
Linaro Pi 2B ADATA 383.7 17.7
Raspberry Pi B+ ADATA 159.6 19.5
MIPS CI20 eMMC ioctl err ioctl err
MIPS CI20 ADATA 141.2 16.7
Banana Pro ADATA 323.4 16.7
ODROID-C1 ADATA 714.5 17.7
ODROID-C1 ADATA UHS 690.2 29.3
ODROID-C1 Patriot 690.1 14.3
ODROID-C1 Patriot UHS 699.7 24.7
ODROID-C1 eMMC 709.0 77.2

Results are in megabytes per second.

No significant difference here.

dd

dd media dd read dd copy dd write
Raspberry Pi 2B ADATA 18.5 6.8 10.9
Linaro Pi 2B ADATA 18.6 6.3 10.4
Raspberry Pi B+ ADATA 18.4 5.8 9.6
MIPS CI20 eMMC 7.9 6.7 32.9
MIPS CI20 ADATA 18.5 3.7 4.8
Banana Pro ADATA 17.5 8.1 16.4
ODROID-C1 ADATA 16.5 7.3 8.1
ODROID-C1 ADATA UHS 30.4 8.0 10.0
ODROID-C1 Patriot 16.5 8.9 16.4
ODROID-C1 Patriot UHS 27.8 12.2 23.1
ODROID-C1 eMMC 80.8 11.6 14.5

Results are in megabytes per second.

No significant difference here either.

Conclusion

The application launch benchmarks make me think that the Raspbian is optimized quite well, and other than boot times, I found Raspbian “snappier” than Linaro.

Compiling Emacs took a lot longer under Linaro.

UnixBench and Sysbench showed the greatest micro benchmark differences, so it is possible to get significant improvements by choosing to compile your CPU bound applications for ARMv7.

The rest of the benchmarks were extremely close, so at a guess, most user-level applications will not show a dramatic difference between ARMv6 and ARMv7 binaries.

Related Links

Article Index:

  1. Introduction, Booting & Launching Apps, Compiling Emacs, SysBench, iPerf
  2. nBench, UnixBench, hdparm, dd, Conclusion

 

Pages: 1 2

If you liked this article, please help us grow! "Like" us and let your friends know with the Twitter and Facebook buttons at the top left - We appreciate your support!