First off, I was wrong when I wrote that the NEON unit is required; it is in fact optional. I regret this error.

Apple assembled an all-star team to produce the A4 chip. The Alpha and StongARM both captured my imagination back in the early 90’s, and the engineer who led their designs, Dan Dobberpuhl, apparently works for Apple now. If anyone could get a quad-core processor shoe-horned into the iPad, it would be his team. Here are a couple relevant articles:

“Dan Dobberpuhl is responsible for the design of many high-performance microprocessors, including the PDP-11, uVax, Alpha, and StrongARM.” — http://queue.acm.org/detail.cfm?id=957732

“Apple is working on something big. But it’s too soon to know what, exactly, Steve Jobs & Co. are doing. But it is not too early to know who is doing it. And that, ultimately, may be more important.” — http://www.forbes.com/2009/04/28/apple-steve-jobs-technology-enterprise-tech-apple.html

Some comments back to you. Yes, I don’t see why NEON would not be utilized. As for the vector processing unit option? Isn’t this one of PA Semi’s strengths? Your comments about cores and power usage are all very insightful and shade this “wonderlust” to find and discover the true details of the A4 all the more fun. Thanks for the great feedback!

It isn’t saying anything more really than what we stated previous.

“However, maybe the secrecy is because the A4 isn’t so custom after all? Perhaps it is simply an ARM Cortex A9-based multicore chip that happens to be quadcore. Does that sounds plausible? It does to me after reading up on ARM’s latest and greatest.”

BSN admits it only got confirmation that Apple is a “member of the family” of ARM that is. That says nothing really. “We were told that ARM licensed its CPU and GPU technology to Apple.” Actually, PA Semi had the broadest architectural license from ARM prior to Apple acquiring them.

Given that Apple and ARM go way back, I’m sure Apple could have acquired an architecture license. ARM does not require that this be public knowledge, and I’m sure that Apple would want to have as much control as possible over any hardware project they undertake.

The Samsung S5PC100 widely reported to be in the iPhone 3GS includes a Cortex A8 with NEON, and I don’t see why Apple wouldn’t use it. Dropping it now doesn’t make sense to me. Perhaps they came up with a better vector processor, but I think it would have to remain compatible.

The following press release on ARM’s website indicates that a single A9 core (with L1 cache) built on TSMC’s low-power, 40 nm process requires just 250 mW: http://www.arm.com/news/25922.html. Since the size of an L2 cache (if any) is not specified, I don’t see how a power figure could be quoted as a single number and not as a range.

We may be overestimating the A4’s power efficiency. Supposing it is a quad-core design with 1 MB total L2 cache, graphic acceleration, and other co-processors, then a peak power consumption under 4 W should be possible at 1 GHz. Typical power consumption, however, could be vastly lower. Given the iPad’s huge 25 Whr battery and large, heat conducting aluminum case, short periods of high power draw may be acceptable (but not acceptable for an iPhone).

Yet the lure of multiple cores from a power perspective is clear: four cores running at 1/2 frequency use only 1/2 the power of two cores running at full speed. In other words, a quad-core chip running at 1 GHz would be as powerful as a dual-core running at 2 GHz but use only half the power. (SMP’s are hard to program, but fortunately Apple’s working on that…) Pages 22 and 27 among others of this PDF (http://www.jp.arm.com/event/pdf/forum2008/t1-1.pdf) include some interesting slides regarding multiple cores and power consumption.

So, I’m still hoping for a quad-core chip with lots of engineering surprises that accomplished the feat, but there is scant evidence so far to indicate this.

According to the ARM website on Cortex A9 the L2 Cache Controller can provide access up to 2 MB (up to!) in high frequency designs (presumably the A4 is a high-frequency design).
http://www.arm.com/products/CPUs/ARMCortex-A9_MPCore.html

I read in another post that the ARM Cortex A9 aimed at smartphones had a power footprint targeted at 300 mW…the limit imposed by cell phone manufacturers on chip makers. It was a direct quote from an ARM product manager.

The patent application description above sounds pretty interesting. Apparently smartphones with quadcores will be aplenty by year end…so this is the year for quadcore smartphones. I just can’t figure out what else the “4″ can mean in Apple’s A4 designation. Have any ideas other than “quad”??

ARM NEON Processing Media Engine? Yes, of course! It’s a required component of the ARM v7 A & R architectures, and the iPhone 3GS already has it by virtue of it’s Cortex A8 processor.

2MB of cache under 300 mW: I don’t develop ICs, but that sounds rather optimistic to me. Cache RAM is very power hungry. I frankly would be surprised if the A4 has more than 64 KB L1 and 256 KB L2. If it’s really a quad-core, those figures might get doubled.

Which brings me to speculation about the chip having four cores. Die space is not an issue, since the A9 core is tiny compared to even the Intel Atom. However, you have to feed all four of those cores with data or they’re just wasting space. To increase memory bandwidth, you’ve got to increase cache size and use a faster memory interface, another power hungry part of the system. I definitely think quad-core iPods, iPhones, and iPads will eventually arrive, but it’s probably too soon. (I hope to be wrong about this!)

Apple has a lot of financial incentive to use the same chip across it’s entire line of mobile consumer electronics, so I expect the A4 is already designed to work in the iPod and iPhone. I don’t expect a hot-rod; I expect something more like a Prius.

Patently Apple spotting an interesting patent regarding CPUs: US patent application 20080288759 describes a method for detecting memory hazards (cache misses) after speculative execution of code. The method described to do this is novel (disclaimer: I am not an expert in the field) in that it combines minimal hardware changes and a smart compiler. This would allow Apple to use a smaller cache to obtain the same performance while reducing both power consumption and cost.

Of course, I was also hoping that Apple’s subtractive display patent (7,586,472) would make it into the tablet, but no such luck. Such a display should be able to use ambient light and thus cut backlight power—surely the largest single power draw of the iPad.

I suppose I’ll just have to wait until the iPad is released before learning anything more for certain about the A4.