Stockfish on iPhone 5s closing the gap with Desktop computer

[Lion]

Hello all,

I have an i7 980x OC at about 3.8 GHz.
Using Stockfish, I hit about 9'000 kn/s.
As the new iPhone 5s is 2x faster than the previous model, it should hit about 1'200 kn/s which is less than 8x slower than my desktop.

8x slower would be around 3x70 ELO = 210 ELO.

That is really not much, especially considering my computer is OC.

impressive times in front of us. Looking forward to the new iPad 5.

rgds

[mwyoung]

Hello all,

I have an i7 980x OC at about 3.8 GHz.
Using Stockfish, I hit about 9'000 kn/s.
As the new iPhone 5s is 2x faster than the previous model, it should hit about 1'200 kn/s which is less than 8x slower than my desktop.

8x slower would be around 3x70 ELO = 210 ELO.

That is really not much, especially considering my computer is OC.

impressive times in front of us. Looking forward to the new iPad 5.

rgds

My guess is it would be less then 210 elo. Since the Iphone is hitting its 1,200 kn/s using only 2 cores. You need to use time to depth in your calculation, not pure nodes per second.

[Lion]

U are maybe right... so that would be even more amazing !

[Lion]

Ok, i tested on my 5s and hit 1'148 kn/s !

[Lion]

For those who are curious, with iPad Air => 1'200 kn's

[Milos]

Hello all,

I have an i7 980x OC at about 3.8 GHz.
Using Stockfish, I hit about 9'000 kn/s.
As the new iPhone 5s is 2x faster than the previous model, it should hit about 1'200 kn/s which is less than 8x slower than my desktop.

8x slower would be around 3x70 ELO = 210 ELO.

That is really not much, especially considering my computer is OC.

impressive times in front of us. Looking forward to the new iPad 5.

rgds

My guess is it would be less then 210 elo. Since the Iphone is hitting its 1,200 kn/s using only 2 cores. You need to use time to depth in your calculation, not pure nodes per second.

Your guess is wrong since on PC I'm sure nobody anymore uses engines with less than 1-2GB of hash. On Iphone with its limited memory you can maximally assign 128MB without the rest of the system being compromised in performance.
8-16x larger hash size on any TC faster than bullet brings at least 40 Elo.
In addition SMP implementation (OS part) on the phone (especially on Iphone) is much less efficient than on the PC (it's enough just to compare cache sizes of ARM and i7 )

[jhellis3]

ARM is an ISA not a chip. And you are quite mistaken regarding Cyclone's caches.

[Milos]

ARM is an ISA not a chip. And you are quite mistaken regarding Cyclone's caches.

ARM is not just ISA, ARM is CPU architecture overall, it would be like saying Intel is ISA , but in context of this thread ARMv8-A is exactly the same as i7, main CPU architercure (Cyclone is just stupid Apple marketing name for slightly modified Cortex A53).
And regarding caches, the important one for SMP L2 (L3 in case of i7) is much smaller 1MB for both cores in iPhone's ARMv8-A compared to 12MB for all cores in i7 980x.

[jhellis3]

I am afraid you are quite wrong and are continuing to misspeak.

[Milos]

I am afraid you are quite wrong and are continuing to misspeak.

How convincing .
I can only remain silent under such strong argumentation.