Where will Apple take Silicon next?

Bloomberg’s Mark Gurman wrote earlier this week about Apple’s roadmap for its M-Series CPUs. It is some solid reporting, but we did not get too far as it largely confirmed what everyone expected – Apple will extend the Mac products which will use Apple Silicon. The headline says it will go into Mac towers and iMacs next, dogpiling on Intel’s growing perception problem of low performance. However, we revisited the article in its entirety after we read John Gruber’s take on the article. Crucially, he read to the end which noted that:

Apple engineers are also developing more ambitious graphics processors.

Gruber’s take is that Apple is likely to start making GPUs, or potentially GPU-like dedicated cores inside the M-Series. This makes a lot sense. On the iPhone, with the N-Series of chips, Apple has already demonstrated that it has immense internal image processing talent. Images – both video and still – are key selling points of new iPhones. This is pretty important for high-end Macs as well, which are workhorses for many designers, and video and graphic artists.

Bloomberg also reported that Apple is testing out 16- and 32-core variants for its graphics processors for early next year, and 64- and 128-core variants beyond that. This struck a chord with us. As most know, Apple sells iPhones on a similar model – double the memory for $100. As industry participants know, the incremental cost of 64 GB of memory is much less than $100, so this is a fantastic margin booster for the company. Why not replicate that with graphics?

Along similar lines, the Bloomberg article pointed out that:

Chipmakers are often forced to offer some models with lower specifications than they originally intended because of problems that emerge during fabrication.

This is not quite right. When chip companies design chips, especially large chips like CPUs, they know that there will be errors when they get the chips back from the fabs. A very small number of those errors will kill the chip. But much more common is to find that small portions of the chip do not work. And when we are dealing with multi-core processors, the odds are fairly high that the defects will occur in one of those cores. So instead of having a 12 core chip, you have 11 working cores (which for reasons too technical to get into here we usually round down to 10 cores). The chip designers know they will have a relatively small number of parts that have 100% of cores working, a larger number that will have 90% working, and the majority will have 80% (all numbers are rough estimates). And so they build this into their plans. They bucket all these quality specs into separate products. It is the same design, but the 80% parts will be marked as the base tier, and then they can charge a premium for the 90% and 100% versions.

Semiconductor fabs being what they are, with time, these errors can diminish fairly dramatically. So with time, yields reach close to 100% for all the chips. But they maintain the three SKUs, and turn off the extra cores (either with software or fuses on the chip) for the lower priced parts.

Part of the problem with doing this is that it makes inventory and operations more complex. The company has to keep track of all the SKUs and maintain sufficient stocks. It is a headache at best, and can sometimes be a big problem if they misprice one of the SKUs and all demand goes there. Nvidia and AMD already do this SKU bucketing, but they have to manage all those different parts across their supply chain and those of the ODMs and OEMs who build and sell the end product. Most of the time it works, but every once in a while you see gross margins whipsaw one way or the other when things get out of whack. Our guess is that Apple can eliminate this whole problem because they control the end product up to shipment to the customer. This gives them the ability to tack on a nice margin booster with limited cost.

Tying a nice bow on this, it delivers a better user experience. When shopping for a PC, nothing is more frustrating than weighing CPU and GPU add-on options. It never quite seems to make sense, unless you have been reading PC Magazine for years. Instead, Apple will be able to offer Mac customers a simple matrix – how much memory do you want, how many GPU cores do you want? Admittedly, customers will be paying hefty premiums for the advanced options, but chalk that up to the magic of the Apple brand.

One response to “Where will Apple take Silicon next?

  1. Pingback: Apple is Building its Own Modem (Redux) | DIGITS to DOLLARS·

Leave a Reply