Apple A12 SPEC CPU2006 results

[Anton Ertl]

Andrei Frumusanu has published SPEC CPU2006 speed results for the
Apple A12 and Apple A11 (and other Smartphone CPUs) on Anandtech:

https://www.anandtech.com/show/13392/the-iphone-xs-xs-max-review-unveiling-the-silicon-secrets/4

For these non-throughput benchmarks, the A11 and A12 blow the
smartphone competition away without consuming more power. What's more
interesting, is that they are competetive with the CPU that has the
highest SPEC CPU base result (while probably consuming quite a bit
less power). If Apple went into the server CPU business, they have
the potential for being quite successful also there.

And if you think that this is due to the 7nm process they use for the
A12 (Vortex core), note that the 10nm A11 (Monsoon core) is pretty
close to the A12 in both performance and power.

Here are some SPECint base speed results (rounded to integers)

A12 A11 9810 8895 845 835 CTX CTX2 7601 8176 6146 1270
45 41 20 12 18 13 8 20 31 50 53 53 400.perlbmk
29 24 16 11 16 12 7 14 24 27 32 34 401.bzip2
45 34 20 10 15 11 11 27 35 25 48 55 403.gcc
50 35 15 10 11 11 10 45 40 43 88 97 429.mcf
39 30 20 15 17 15 9 16 24 31 38 38 445.gobmk
44 40 29 18 25 16 5 22 28 35 110 110 456.hmmer
37 31 18 13 15 12 9 16 24 34 42 42 458.sjeng
113 94 56 32 49 28 6 76 69 102 7860 2400 462.libquantum
67 61 37 21 32 22 12 27 50 67 80 82 464.h264ref
36 25 11 8 8 9 7 26 23 41 41 38 471.omnetpp
27 21 14 10 12 10 8 16 20 27 44 44 473.astar
57 47 23 13 19 15 8 28 35 67 92 101 483.xalancbmk

Data from:

https://www.anandtech.com/show/13392/the-iphone-xs-xs-max-review-unveiling-the-silicon-secrets/4
A12: Apple A12 (Vortex core)
A11: Apple A11 (Monsoon core)
9810: Exynos 9810/2314 MHz (M3 core) (the 2704MHz result takes too much power)
8895: Exynos 8895 (M2 core)
845: Snapdragon 845 (Cortex-A75)
835: Snapdragon 835 (Cortex-A73)

https://www.anandtech.com/print/12694/assessing-cavium-thunderx2-arm-server-reality
CTX: Cavium ThunderX
CTX2: Cavium ThunderX2

https://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/14
7601: EPYC 7601 3.2GHz Turbo
8176: Xeon 8176 3.8GHz Turbo

https://www.spec.org/cpu2006/results/res2017q4/cpu2006-20171114-50707.pdf
6146: Xeon Gold 6146 4.2GHz Turbo (Skylake-SP)

https://www.spec.org/cpu2006/results/res2016q1/cpu2006-20160108-38634.pdf
1270: Xeon E3-1270v6 4.2GHz Turbo (Kaby Lake)

For evaluating the hardware, it's probably best to compare among the
Anandtech numbers (i.e., up to 8176), where compiler effects do not
appear to affect the results as strongly.

OTOH, if we want to see where compilers have the most effect, just
compare the 8176 numbers with the 6146 numbers.

In 7 of 12 benchmarks, the A12 beats the 8176, in 2 they are equal, in
3 the 8176 beats the A12. The A12 consumes only 3.64W on avarage
while running these benchmarks, which is substantially better than
what the 8176 consumes (it has 165W TDP for 28 cores, i.e., 5.89W/core
if all are active, but in that benchmark typically only one core was
active, so it used a higher clock rate that consumed more than 5.89W).
So an A12 core is about as fast as a Xeon core, at significantly less
power.

Of course, if you think that the compiler effects that broke
libquantum will substantially speed up the applications you use, go
for Intel and its icc.

- anton
--
M. Anton Ertl Some things have to be seen to be believed
an...@mips.complang.tuwien.ac.at Most things have to be believed to be seen
http://www.complang.tuwien.ac.at/anton/home.html