Heya! Sorry, I would've added some links but I was on mobile and still am.
As I hinted at, this is mostly coming from anecdotal wisdom I've heard. The hardest evidence seems to be Apple's chips increasing more (10-20%) in GB6 vs GB5 when compared to how much other manufacturers such as QLC and INTL increased. As another comment said, GB5 is well known to be quite accurate with SPEC (industry standard), thus this discrepancy with GB6 and GB5 is a bit concerning.
Of course synthetics are always skewed (sometimes because a chipmaker tries, sometimes not), most famously Cinebench which has lost most credubility for favoring SIMD perf far too much.
Also might be worth it to check out the top comment to see an example of how a choice can bias syncthetics but not real-world.
I would recommend reading the Geekbench 6 internals document, they explain the rational behind the change.
> Geekbench 6 uses a “shared task” model for multi-threading, rather than the “separate task”
model used in earlier versions of Geekbench. The “shared task” approach better models how
most applications use multiple cores.
> The "separate task" approach used in Geekbench 5 parallelizes workloads by treating each
thread as separate. Each thread processes a separate independent task. This approach scales
well as there is very little thread-to-thread communication, and the available work scales with
the number of threads. For example, a four-core system will have four copies, while a 64-core
system will have 64 copies.
> The "shared task" approach parallelizes workloads by having each thread processes part of a
larger shared task. Given the increased inter-thread communication required to coordinate the
work between threads, this approach may not scale as well as the "separate task" approach.
Nothing about this is biased towards Apple. GB6 simply scales worse with more cores due to increased inter-core communication requirements.
This is correct. AMD and Intel CPUs had many slower cores. GB5 made them look better. Apple has fewer cores but more fast ones.
Most applications can't utilize many cores. Thus, usually, consumer applications perform better with fewer but faster cores than many slow cores. Geekbench is a consumer CPU benchmark.
I think you may be a few steps behind, what you're explaining is the rationale for the ST and MT individual scores. This is something all modern benchmark software has.
Hallo! That was really interesting to read through!
I just wanted to clarify that by bias I mean a skew towards one side that causes scores to misalign with SPEC. In this case, this bias against high-latency ICC is one of the causes of such 'bias'.
Thanks for the thoughtful and informative response though.
The team at Nuvia, before they were acquired by Qualcomm, did the analysis: https://medium.com/silicon-reimagined/performance-delivered-...