AES test in Geekbench 3 variability
Just to warn that I've noticed the AES test in Geekbench 3 can vary enormously between machines compared to the other Integer tests. I think it must be using the Intel AES facility when available. Not sure if you want to mark AES as quite so important - it is difficult knowing what to do about facilities like that which are good but only sometimes.
Keyboard shortcuts
Generic
| ? | Show this help |
|---|---|
| ESC | Blurs the current field |
Comment Form
| r | Focus the comment reply box |
|---|---|
| ^ + ↩ | Submit the comment |
You can use Command ⌘ instead of Control ^ on Mac
Support Staff 1 Posted by John on 18 Aug, 2013 09:53 PM
Yes, the AES test uses the AES-NI instructions when available, which is why there are large differences in AES test scores across systems. While AES performance is important (it's one of the most widely used encryption algorithms) it's just one of 13 integer tests, so it should not dominate overall performance.
We've also included a Blowfish encryption algorithm test as well to measure the performance of encryption algorithms where dedicated instructions are not available for the algorithm.
Let me know if you have any other questions and I'd be happy to help out.
Best,
John
2 Posted by Anon on 12 Oct, 2014 11:28 PM
When comparing Intel vs ARM, the AES speed can be 300x as fast on Intel, while Integer performance otherwise is 3x as fast.
So I worry that the 100x difference in relation to other tests in fact DO dominate the integer tests.
Example:
http://browser.primatelabs.com/geekbench3/compare/949953?baseline=786009
Support Staff 3 Posted by John on 15 Oct, 2014 04:03 AM
Hi Anon,
In your example the Intel Xeon score is 2.5x higher than the Rockchip RK3288 score. If you exclude the AES test then the Xeon score is "only" 2.3x higher than the RK3288 score.
AES is an important encryption algorithm and hardware cryptography instructions provide a massive boost in performance to both benchmarks and real-world applications. If you're not interested in AES performance then I would recommend ignoring the AES test results. Since Geekbench transparently breaks the overall scores down into individual section scores and workload scores it's possible to re-calculate the Geekbench scores to exclude one or more workloads. I would be happy to provide you with some scripts that help automate this process.
Let me know if you have any other questions and I'd be happy to help out.
Best,
John