Slashdot Mirror

That's two components, at least one of which is based on licensed block designs from ARM, who could simply stop licensing them for future use and BAM, no more A series chips can be made. Go on thinking Apple invented everything in the iPhone, really, go ahead, don't let those pesky facts get in the way. Don't get me wrong, I've got a number of apple products and I do love them, but i don't bullshit myself about what Apple actually does as a company. The AC you replied to is absolutely correct.

Actually, Apple has a "Soft Core" ARM license that actually lets them design their own CPUs (and they do), rather than just licensing IP from ARM. Being a founding member in ARM doesn't hurt, either. There are a few other companies that have such a license (Samsung and Qualcomm being two of them, IIRC); but Apple actually does design A-series chips from the CPU-on out.

There's a great website called LMGTFY... :)

But, as a FYI, Apple was one of the founders of ARM, so it stands to reason Apple would have access to any kind of licensing they wanted.

http://www.linleygroup.com/new... is from 2012, and is pretty accurate.

manufactured in Texas by Apple-owned Fab house, Intrinsity

Intrinsity (back when it existed independent of Apple) was a fabless manufacturer that contracted out to other foundries (including and especially Samsung) to manufacture their chips. Samsung just upgraded its Austin plant for A#-series chip production; Intrinsity never had an Austin plant and Apple hasn't built one.

Rumor has it (ugh...) that Apple was trying to shift assembly to TSMC, but they didn't have the quality/yield.

Hmmm. Interesting. I did not know that! Why would Apple need a fabless semi house? That makes NO sense...

Oh wait! Here's some stuff... I guess I'd always figured that PASemi was responsible for the Axx SoC design. Guess I was only half-right!
HOWEVER, the rest of my post stands. ;-)

I worked in the early 1990s at a startup that interconnected multiple parallel DSPs (AT&T DSP32C) by (Plessey, then Xilinx) FPGA. A lead DSP preprocessed the stream to dice it into parallel tasks, which it distributed through the FPGA to the parallel DSPs. Mainly it decomposed a single color image into independent color channels, transforming color spaces in the master thread for faster processing in the slave threads. The FPGA gradually absorbed more CLU logic from the original app that was prototyped on a 386 and ported to a single DSP, reducing the DSPs to more totally ALU logic. If the company had survived Bush Sr's 1991 recession, we probably would have had a multichip module by 1994 that worked essentially the way the Zynq does now in 2011.

I'm not sure how the AMBA bus scales across chips. But since one of the geniuses at my old company is one of the geniuses at Xilinx now, I expect that one way or another their Zynqs will scale using the onchip tech.

As for power efficiency, the Atom design targets mobile, while Zynq is reported to dissipate under 2W ("typical", maybe not with full FPGA execution), at under $15 a chip in quantity. If Zynq's FPGA executes logic more efficiently (BitsOutput:WattConsumed) than general purpose ARMs over the course of a whole day of diverse smartphone use, then mobile devices might swing over to them in a rush. Indeed Google's Dalvik JVM differs from Sun's JVM in being register oriented rather than stack oriented. If so, then FPGA matches Dalvik more closely than even ARM, and the performance:power ratio could be the best of any Android device.

Actually not really true the current cortex A9 is about 2-3 times as fast as the Atom, you can google the performance numbers.

I'm highly sceptical of this

http://www.anandtech.com/show/3696/intel-unveils-moorestown-and-the-atom-z600-series-the-fastest-smartphone-processor/14

ARM does a bit better on CoreMark (5.7 vs 3.6 for a single threaded app, for dual threaded an N280 does 5.6. Most CPU intensive stuff is probably dual thread friendly these days, even the horrid flash)

http://blog.linleygroup.com/2010/04/arm-outmuscles-atom-on-benchmark.html

Still CoreMark is small enough to fit in the L1 cache. The Achilles heel of ARM systems has been slow memory controllers. Admittedly that's because ARM is aimed at phones where power consumption is a much more serious issue than netbooks, but still.

On something like SpecInt which is large enough to exercise the DRAM interface I'd expect Atom to blow away an ARM from a similar generation. And Intel already have things like Core2 which can be migrated down to low power as you can see with ULV. I'm not convinced Arm can migrate up.

Emulation would likely make things worse - CPU core performance will drop and if a JIT is used it will stress the memory subsystem more because you need to access both the native code and the translated code. An Jazelle like solution implies a greater memory subsystem hit too - you need to keep the code that emulates the instructions you don't handle in hardware in the cache. That's cache memory an Atom could use on the code its actually executing.

In fact Anandtech reckons that Apple should switch from ARM to Moorestown

http://www.anandtech.com/show/3640/apples-ipad-the-anandtech-review/17

I expect Itanium to replace Xeon, but not until 2003, when McKinley and its successors open a performance gap over Xeon.
[...]
Changes in servers never happen fast. But with Itanium now a reality, Intel's dominance is only a matter of time.