I am pro-truth and some of you did not like hearing it at the time. 45nm Moorestown was relatively poor, the datasheet specs said so and so was the delivered performance/power as tablets/phones with it died on the vine. I said at the time Intel would have to increase its performance/power by a factor of 2 with Medfield to be competitive against ARM chips and they more than delivered. Once Medfield appeared it was obvious it would be a player but not the knockout blow a lot of you thought due to the early Intel PR. Again balance in the discussion was needed which I helped to bring. However no more caution is required as ARM has showed its hand with A15 and there is so much low hanging fruit to be had with 22nm Silvermont on both the architecture and process front that it won't fail to be a compelling phone chip, probably taking the performance/power crown from Apple's A6.
I am thinking of initiating a pairs trade long INTC / short ARMH using options. Will probably go with Jan 2015 leaps. Do you think the two stock prices could move to reflect the competitive shifts in play sooner (say Jan 2014) or is that too optimistic?
Are you familiar with BOINC ? The BOINC client has now been ported to ARM and there is a growning set of Dhrystone MIPS and Whetstone MIPS that measured from (I expect) similarly compiled source code. I don't like Dhrystone MIPS (Weicker, author has also given it thumbs down) or the Whetstone benchmarks but we can directly compare the numbers as possibly being similar.
My Linux Ivy Bridge Core i5 3450 3.10ghz measured BOINC
3,151 FP MIPS (whet)/CPU and
17,516 INT MIPS (DMIPS)/CPU.
They have ported the BOINC client to run on Android and run the benchmarks on phones, tablets, laptops, desktops, workstations and servers. The DMIPS and Whetstone FLOPS are generated from small programs that run out of cache so the bus structure should not affect results.
There are several BOINC STATS sites that accumulate the CPU data and AVERAGE CREDITS PER CPU SECOND for the larger less comparable workloads but the results are rather interesing.
It seems like a 1.6ghz ARM v71 is about the same as an ATOM N450 1.66ghz. The metrics are PER CPU.
0.000935 AVERAGE CREDITS/CPU SECOND
ARMv7 Processor rev 0 (v7l) @1600MHz
Intelx86 Family 6 Model 6 Stepping 5 (Mendocino – 0.25 µm process August 24, 1998)
Intelx86 Family 6 Model 8 Stepping 6 733MHz (Coppermine – 0.18 µm process October 25, 1999)
0.000938 AVERAGE CREDITS/CPU SECOND
Intel(R) Atom(tm) CPU N450 @ 1.66GHz (Pineview 21 December 2009)