12-02-2020, 10:08 PM
I haven't seen a comparison of the two A72 cores vs the four A53 cores. Just how much faster are the A72s?
So I did a benchmark of compiling the kernel with just the two A72s enabled vs just the four A53s.
Values in seconds. Lower is better.
So for compiling the kernel, an A72 core is about 2.5x faster than an A53 core. The big group put together is about 1.26x faster.
Which is pretty much as predicted from the per-core speeds, as (2 cores * 2.5x) / (4 cores * 1x) is 1.25x. As we can see, the kernel compile scales very well to four cores, but not perfectly, as there are still some bits that only use one core. So the faster single threaded performance of the A72 helps. But very little.
One might guess from the schematic that the big core group draws about 2x the power of the little core group. So if you want to see how many kernels you can compile before the battery dies, just the little group, it's more efficient (peripheral power draw not counted, so not really).
So I did a benchmark of compiling the kernel with just the two A72s enabled vs just the four A53s.
Values in seconds. Lower is better.
- A53x4: 2888 elapsed, 10985 cpu
- A72x2: 2285 elapsed, 4377 cpu
So for compiling the kernel, an A72 core is about 2.5x faster than an A53 core. The big group put together is about 1.26x faster.
Which is pretty much as predicted from the per-core speeds, as (2 cores * 2.5x) / (4 cores * 1x) is 1.25x. As we can see, the kernel compile scales very well to four cores, but not perfectly, as there are still some bits that only use one core. So the faster single threaded performance of the A72 helps. But very little.
One might guess from the schematic that the big core group draws about 2x the power of the little core group. So if you want to see how many kernels you can compile before the battery dies, just the little group, it's more efficient (peripheral power draw not counted, so not really).