Hacker News Reader

Comment by trhway | Hacker News Reader

>the exponential rise in the price/performance of computing power looks like it is not going to stall out any time in the next 10-15 years.

research is good, yet it isn't enough for progress. You need driving force like market competition. With AMD out of the game (hopefully temporarily) Intel, with Ivy Bridge, went of to sell smaller chips (instead of doubling the number of cores on the same area chip) for the same money and with Haswell they decided that they can sell it much longer before introducing new generation (Broadwell). We need AMD back, or much faster ARMs, or whatever to start breathing into Intel's neck again ...

View on HN

True, and I love AMD and wish they were still competing effectively. No doubt if they were, the recent (relative) stagnation in chip price depreciation would not have been so pronounced.

But it's important to remember how innovations such as these change the landscape for competition. As a large pseudo-monopoly, Intel can be complacent, but not indefinitely. While the barrier to entry is very high in chip fab, and maybe insurmountable by anyone assuming current chip designs, there are a lot of very well capitalized electronics corporations in the world. If they believe a new manufacturing process will be required for the next generation in chips and Intel's margins are too high, they may well seize upon the opportunity to go after it.

This type of research it exactly what is needed to help sew the seeds for those next generation processor technologies that will help shake up the competitive landscape and dislodge the stodgier players who sit on high margins and care more for next quarter rather than the next decade.

Market competition is good, but it isn't enough for progress either. Current microprocessors are quickly reaching fundamental physical limits. We will still see speed increases from architectural and algorithmic development, but not from faster (...5Ghz) processors.

The next revolution is widespread GPGPU programming, which can yield great increase in performance when leverage correctly. Java 8 already have libraries for HSA and Java 9 will support it out of the box. And this is AMD lead technology.

While i'm vaguely excited for HSA i'm pretty skeptical of anything that needs to be 'leveraged correctly' there's a history of sufficiently smart compilers not materializing.

Depends. Multithreading is a common method nowadays despite sufficiently smart compilers not existing. It just requires someone who knows what they are doing, just like good multithreaded code.

As a consumer I personally haven't seen the 'killer app' needed to require such horsepower. Rather, it seems Intel is focusing on the next logical thing -- low power, more efficient chips.

Reply on news.ycombinator.com