Moore’s Law, it ain’t just for the rich–Or is it?
Posted by Jon Peddie on March 29th 2012 | Discuss
Mopping up; that’s what we used to call it when we would go to the next process size and integrate some of the peripheral components. It was also called jellybean removal. Companies like National, ST-Micro, TI, and SMSC hated it because every time there a mop-up they lost business for some jelly beans.
Recently Microsoft did it in the Xbox 360, taking the Power PC, the ATI Xeonos from 65nm to one SoC in 45nm. In the process they eliminated a power consuming front-side bus, and lowered cost and power consumption.
Also in the near past Intel combined the GPU and CPU die in their multi-chip Arrandale into one chip and introduced the Sandy Bridge. All of a sudden the historical evolution of process density of the so called Moore’s Law seemed new and grave new predictions were made about adjacent components being mopped up. Little things like a three-billion transistor GPU. What is missed in these forecasts of doom for the adjacent devices is an understanding of the trajectory of development of the adjacent device and the mopped up device(s).
When Microsoft made their super SoC it was a logical and smart, and difficult thing to do. Microsoft did the right things, but they could only do the right things because they were designing for a static (performance-wise) game console. One of the main attractions to the lazy game developers and publisher of the console is that it doesn’t change, it delivers the same MIPS and FLOPS in 2012 as it did in 2002. Many game developers love that because they don’t have to invest in new software tools or learn anything new – the performance stays flat, their costs stay flat.
So mop up works really well in that situation. However, if the platform is evolutionary then the mopper-upper has to be Solomon like a balance the mopping up with performance improvement. The mopper-upper shouldn’t love one child more than the other, but, cultures being what they are in organizations, they do love one child more than the other. In the case of Intel, they love their x86 child more than their GPU child (which has often been cruelly described as stunted in its growth). AMD on the other hand loves tier GPU chip more than their x86 child (which has been scornfully referred to as a weakling).
In the astonishing mobile market the mopper-uppers were at it a decade ago and almost invented the term SoC. Those companies, like Apple, TI, Qualcomm, and others have been pretty balanced in their efforts moving the GPU along about as fast as the CPU, if not faster. Part of the reason for that is they don’t have a technologic culture, they are buyers of technology getting the CPU IP from one company and usually the GPU IP from another. In the case of Qualcomm and Nvidia where they have their own internally developed GPU IP, they do tend to be a bit more biased toward the GPU. In the case of TI and Qualcomm, they also have a bit of a bias for DSPs.
But regardless of one’s bias to do mopping up you have to have the mop, and the bucket, and soap and all the other stuff. In the semiconductor biz those things are simulators, design tools, debuggers and compliers, oh, and a fab that can build the work of art they produce. That stuff ain’t cheap. In fact it’s so expensive it becomes one of the primary barriers of entry to a market. The other barrier is the mopper-upper him - or herself – the designers. SoC (i.e., mop up) designers aren’t standing on street corners waving at pickup trucks looking for day jobs. And they aren’t cheap— rare things seldom are.
So it’s an expensive proposition to be in the game, unforgiving, unrelenting, and uncheap—mopping up isn’t for poor companies. That’s why the little ones like Matrox and VIA don’t do it. They also do not manufacture in the latest process because of the associated costs and talent needs. The irony is the poor companies freeze their designs because of expense and fight against downward price pressure, while a rich company like Microsoft mops up to reduce costs. And the other rich companies like AMD, Apple, Intel, TI, and Qualcomm mop up to keep on the technology curve.
ASIC suppliers like AM D (ATI) and Nvidia mop up very little but push the process curve, often a half step ahead of the CPU suppliers. Possibly the last major mop up by the GPU suppliers was the integration of the RAMDAC. Most of the changes made by the GPU suppliers is in the architectural design.
And all of those things are expensive, so maybe Moore’s law really is just for the rich. Maybe that’s why the most of the new companies are either IP or software companies.
Previous entry: “Steady as she goes” for the workstation market in Q4’11