Geekbench ARM vs x86 confusion

[ScratchCat]

This question has been on my mind since apple started their push for better single thread performance.

Why is it that a passively cooled ARM chip such as Apple's A10X can reach geekbench scores of 3900/9430 at 2.4 GHz while my i5 4690k scores the same at 3.5 GHz but consumes 35-44w at the package level?

According to Wikipedia Apple's A10 (Not X) already uses 3.3 billion transistors while a i7-6950X uses 3.2 billion. Can the GPU make such a difference? Are these benchmarks even comparable or has Apple found the optimal architecture design in less than 6 years?

 

 

A10X source https://www.notebookcheck.com/Apple-A10X-Fusion-SoC-Prozessor-Benchmarks-und-Specs.227170.0.html

https://en.wikipedia.org/wiki/Transistor_count

[sazrocks]

That's ARM for you. Think of it this way. Which is better for cutting vegetables, a Swiss army knife or a kitchen knife? The kitchen knife of course. But which can do other things? The swiss army knife. ARM cpus are more focused in their purpose and have a reduced instruction set, thus allowing them to be more efficient. On the other hand there are many fringe cases that x86-64 can certainly do that ARM just cant do properly. Its about versatility vs efficiency.

 

Or of course the scores could just be different on mobile devices.

[ScratchCat]

7 minutes ago, sazrocks said:

-snip-

That makes sense. I guess also one can compile thru java byte code on android fot each architecture while on x86 you only have a single binary.

Thank you, I can't give rep as button is broken.

[sazrocks]

Just now, ScratchCat said:

That makes sense. I guess also one can compile thru java byte code on android fot each architecture while on x86 you only have a single binary.

Actually its usually the other way around, at least in the case of Java. When running programs on bare metal you have to worry more about the processor architecture and whether it is x86 or x86-64(amd64). For Java all of your programs run in a VM so you don't really have to care about what processor is actually running the VM or its architecture. 

[Coaxialgamer]

Those transistor counts can't be compared. 

6950x Tcount includes cpu and cache

A10x Tcount includes cpu, gpu, cache, modem, comm and everything else in the SOC. 

[ScratchCat]

2 minutes ago, sazrocks said:

Actually its usually the other way around, at least in the case of Java. When running programs on bare metal you have to worry more about the processor architecture and whether it is x86 or x86-64(amd64). For Java all of your programs run in a VM so you don't really have to care about what processor is actually running the VM or its architecture. 

What I meant was companies want many people to use their software but the customers might use older CPUs, therefore they would not use newer instruction sets which could boost performance.

In regards to Java, I didn't know that, interesting fact.