Share This article

One of the topics that comes up fairly frequently in any discussion about enthusiast computing is the idea of switching to alternative semiconductors. For decades, we’ve relied on silicon — it’s cheap, relatively easy to mass produce, and we found multiple ways of extending its utility as opposed to integrating an entirely new material. But here’s the question — as enthusiasts, what would you pay for a return to “classic” CPU scaling, and how often would you be willing to pay for it?

Struggling towards a new economy of scale

One of the reasons Intel became the industry titan it is today is because it was able to take advantage of an enormous economy of scale. This was particularly true in the RISC workstation market, where multiple vendors competed to sell a very low number of microprocessors in absolute terms. SPARC, MIPS, Alpha, and PA-RISC were all extremely profitable at one time or another, but that profitability rode on tens of thousands of chips — not the millions of processors that Intel was shipping.

The electron mobility of various semiconductors. Silicon is the flat line at the bottom, as compared to the III-V types shown at the top of the graph

There are alternative semiconductors, like gallium arsenide, that could potentially hit far higher frequencies than silicon without shattering power envelopes — but the wafers for such chips could be 5-10x more expensive. The cores themselves might be larger, if Intel had to step back to an older process node to integrate the manufacturing. The end result could be a core that cost far more than a current chip, even if the end result was a much faster processor.

The graph above shows the electron mobility of various semiconductors, meaning the speed with which electrons move through the semiconductor in the presence of an electric field. Silicon is clearly not the best option if performance is the goal.

The AMD R9 295X2 has a massive 500W TDP, but it’s still fairly quiet thanks to its massive watercooling solution.

Over the last few years, Intel has chosen to focus on ramping up CPU efficiency. CPU TDPs, even on the desktop, have trended downwards, even as performance per watt has grown. Haswell and Ivy Bridge-E are good cores, but AMD generated far more conversation with the 5GHz FX-9590 — even if that chip also came with a 225W TDP. Then there’s the R9 295X2 — AMD’s new dual-chip GPU with a liquid cooling solution and a 500W TDP.

All things being equal, manufacturers and consumers both prefer low power solutions, but it’s possible that enthusiast consumers (at least on desktops) typically use TDP as a rough stand-in for noise level than a problem in and of itself. Clearly AMD felt that there was a market for a desktop 500W graphics card, provided that part didn’t sound like a dental drill. Were they right?

Given this, I’d like to ask the following:

1. Take whatever you paid for the last CPU you purchased — how much would you pay for a core that was twice as fast? Do you have a use for the additional performance?
2. What would you pay (in absolute terms) for a manufacturing shift that could return the market to better clock scaling — say, 15-20% a year, with additional architectural improvements, as opposed to the current status quo? (i.e. would you pay $1,000 for a gallium arsenide CPU?)
3. If you’re a desktop user, do you actually care about TDP — or do you just use TDP as a proxy measurement for noise? If you knew you could buy a 225W TDP processor from Intel that was 15-20% faster than the 84W chip, but that both chips would be equally quiet, would you be interested?

Some of ExtremeTech’s writers and editors will be answering these questions in the comments below. For previous discussions, hit up the ExtremeTalk category.

http://www.extremetech.com/extreme/180365-extremetalk-what-would-you-pay-for-a-return-to-old-school-cpu-scaling