Apple a8 vs Apple a9

[superpetro1]

Ok, so I was curious as to how much the advantage is that the a9 architecture has over the a8 architecture. I did some rough math to check:

Info I know:
-Samsung stated that their 14nm process would have a 20% increase in performance compared to leading 20nm architectures. This would be due to the finfet properties of the chips which would reduce leakage.
-the clock speed gain over the a8 is 1.23x (compared to iPad mini 4 at 1.5ghz)

So I did this test using geekbench 3. The iPhone SE had a score of 2556 single core and 4444 multi-core.
It's subscores were as follows (integer, FP, memory)
Single-core:
-2548
-2523
-2638

The a8 was as follows:
-1819
-1699
-1562

Geekbench weights the scores
Integer is 40%
FP is 40%
Memory is 20%

Also, I dropped the a8 memory subscore in favor of the a9 memory subscore, as a shrink in a8 would also probably bring ddr4 support with the new generation.

My math is as follows:

1.23*1.2(1819(.4)+1699(.4))+2638(.2)=2596.18

Basically exactly the a9 single core score.

Would a shrunken a8 core with ddr4 been basically as fast as an a9 core?

 

[MikeA65]

Your calculation is wrong, the 1.23 clock speed gain is the 20% performance increase that Samsung's 14 nm process gives, your double counting it in your calculation.

 

[superpetro1]

Your calculation is wrong, the 1.23 clock speed gain is the 20% performance increase that Samsung's 14 nm process gives, your double counting it in your calculation.

Wouldn't the 20% performance increase be at the same clockspeeds? Apple was not able to clock their CPUs higher on the 20nm node because of leakage, however on 14nm/16nm wouldn't you get the benefits of the shrink as well as the ability to clock higher?

 

[IntelUser2000]

Wouldn't the 20% performance increase be at the same clockspeeds? Apple was not able to clock their CPUs higher on the 20nm node because of leakage, however on 14nm/16nm wouldn't you get the benefits of the shrink as well as the ability to clock higher?

When fabs refer to "performance", they mean transistor performance. Transistor performance, assuming architecture is identical, directly translates into clock speeds.

Imagine a new person at work. Imagine a skilled, experienced person at work. Every minute the latter person will do more.

Of course in the case of CPUs, not only you can get more skilled/experienced workers, but make the clock(time) go faster as well. The process technologies like 16/14nm only speeds up the latter, meaning faster clocks. Getting higher skilled employees are much harder in the world of computers though. 20% faster at the same clock speed in just one generation is extremely impressive, because you are improving it by 20% over your previous best work. 20% clock on the other hand, is easier to come by.

 

[superpetro1]

When fabs refer to "performance", they mean transistor performance. Transistor performance, assuming architecture is identical, directly translates into clock speeds.

Imagine a new person at work. Imagine a skilled, experienced person at work. Every minute the latter person will do more.

Of course in the case of CPUs, not only you can get more skilled/experienced workers, but make the clock(time) go faster as well. The process technologies like 16/14nm only speeds up the latter, meaning faster clocks. Getting higher skilled employees are much harder in the world of computers though. 20% faster at the same clock speed in just one generation is extremely impressive, because you are improving it by 20% over your previous best work. 20% clock on the other hand, is easier to come by.

Ahhh alright, thanks for clearing that up