Home → Geekbench →

Geekbench 2.x.x Versions Results Comparability ? 2.2.0 vs 2.2.7 ??

• Edit

jj.gud

18 Aug, 2013 08:45 AM

I ran both versions on my Apple Cube G4 7447A@1.7GHz (Sonnet upgrade).
The aggregated 32 bits scores are VERY different :
2.2.0 -> 743
2.2.7 -> 865

The differences come from a very different number of benchmarks executed by both versions :
2.2.0 -> 26 programs
2.2.7 -> 39 programs
The 13 additional programs executed by Geekbench 2.2.7 are all multi-threaded versions of the original 13 Integer + Floating Point programs.

First, I am surprised of this huge increase in number of programs between both 2.x.x versions, especially since it is not mentioned in the versions history.
Second, I don't understand the purpose of executing multi-threaded versions of programs on a single core / single processor CPU (7447A G4). It doubles the weight of the programs executed twice (single + multi threaded) and skews the aggregated scores, making them non comparable from one version to the other.

Or is that just a bug in Geekbench 2.2.7 ?

1. Support Staff 1 Posted by John on 20 Aug, 2013 04:54 AM

   

   Hi JJ,

   Originally Geekbench 2 ran both single-threaded and multi-threaded tests regardless of the number of cores present in the system. A change was introduced around Geekbench 2.2.0 that changed the behavior so that multi-threaded tests were only run on multi-core systems. That change was reverted in Geekbench 2.2.6 ("Re-enabled multi-threaded tests on single-core systems.").

   Let me know if you have any other questions and I'd be happy to help out.

   Best,
   John

   • Edit

2. 2 Posted by jj.gud on 20 Aug, 2013 05:54 AM

   

   Hi John,

   Thanks for the explanation, I had missed in Geekbench 2.2.6 Release Note "Re-enabled multi-threaded tests on single-core systems." and there was no corresponding Note for the 2.2.0 opposite change.
   Due to the huge differences of Scores, it however makes many of the registered browser results non comparable, unless you filtered out all 2.2.0-2.2.5 results.
   Ah well, maintaining backward comparability while correcting/evolving the benchmarks (or the benchmark aggregated metrics...) is quite a challenge ;-)

   Regards
   JJ

   • Edit

3. Support Staff 3 Posted by John on 29 Aug, 2013 07:33 AM

   

   The change was the result of a bug introduced when we introduced stress tests (see "Added a new hardware stress test" in the 2.2.0 change list). We may want to filter out 2.2.0-2.2.5 results for single-core systems. I'll look into making that change.

   • Edit