The desktop dataset is too large for mobile devices -- it does not fit onto mobile devices with less than 1GB of RAM. Even if it did fit, the runtime would be unreasonably long for our users.
The mobile dataset it too small for desktop devices -- execution times would become small enough...
Geekbench 3 takes advantage of SIMD instructions for some of the workloads.
We build Geekbench with the de-facto standard compiler for each platform since this how most software is built. I'm not aware of any platforms where the Intel compiler is used for a non-trivial amount of software...
FWIW, here's a comparison between the mobile and desktop workload sizes on the A9:
http://browser.primatelabs.com/geekbench3/compare/3594692?baseline=3594646
Mobile is on the left, desktop is on the right.
When Geekbench 3 was released, there was a huge delta between desktop and mobile performance. Running the desktop dataset on mobile devices would increase runtimes to a point where our users would become impatient. Running the mobile dataset on desktops would decrease runtimes to a point where...
For Geekbench 3 append ".gb3" to URLs to download the raw JSON result, which you can then parse with our Python scripts:
https://github.com/primatelabs/geekbench-tools/tree/master/geekbench3
We're reading the maximum frequency from /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_max_freq (which is 2499000 on a Nexus 9). /sys/devices/system/cpu/cpu0/cpufreq/stats/time_in_state shows that the device never goes beyond 2295 MHz, though:
shell@flounder:/ $ cat...
We've put a lot of effort into making sure cross-platform comparisons are valid. Using Geekbench 3 to compare iOS and Android devices is quite valid. Recent iOS devices do quite well in Geekbench 3 for two reasons:
* A7, A8, and A8X chips have impressive processor cores.
* A7, A8, and A8X...
Could you let me know of scenarios where we report the wrong frequency? I'm aware of an issue with big.LITTLE systems but otherwise Geekbench should report the correct frequency under Android.
This is a toolchain limitation.
Dijkstra's data structures contain a lot of pointers, so it's a combination of increased memory transfer and increased cache pressure. I don't know why BZip2 Decompress is faster off the top of my head.
I believe Clang performs the same optimization, which...
Only the AES and SHA-1 workloads show an improvement on AArch32 (thanks to the ARMv8 cryptography instructions). AArch32 doesn't include the extra registers or the double-precision NEON instructions that are part of AArch64.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.