I don’t know the principles behind how doping concentrations are chosen, but I’m sure it’s optimized for speed. Also, you can compensate for Vth variaton using body bias, but it’s basically impossible to do this per-transistor. You can do it for large blocks of transistors, which allows you to compensate a bit for systematic variation (due mostly to optical aberrations in lithography), but there’s nothing you can do about random variation. Also, there’s effective length variation, which I don’t think you can compensate for using body bias.

What you’re talking about is “approximate computing,” which is a hot area in research right now. If you can tolerate some errors, then you can get a massive increase in performance.

So, first other countries start dropping the dollar as the international reserve currency. Now they’re going to stop buying our products and services. Our economy is going to hell in a handbasket.

There are a number of factors that affect the value of technology scaling. One major one is the increase in power density due to the end of supply and threshold voltage scaling. But one factor that some people miss is process variation (random dopant fluctuation, gate length and wire width variability, etc.).

Using some data from ITRS and some of my own extrapoliations from historical data, I tried to work out when process variation alone would make further scaling ineffective. Basically, when you scale down, you get a speed and power advantage (per gate), but process variation makes circuit delay less predictable, so we have to add a guard band. At what point will the decrease in average delay become equal to the increase in guard band?

It turns out to be at exactly 5nm. The “disappointing” aspect of this (for me) is that 5nm was already believed to be the end of CMOS scaling before I did the calculation. :)

Incidentally, if you multiply out the guard bands already applied for process variation, supply voltage variation, aging, and temperature variation, we find that for an Ivy Bridge processor, about 70% of the energy going in is “wasted” on guard bands. In other words, if we could eliminate those safety margins, the processor would use 1/3.5 as much energy for the same performance or run 2.5 times faster in the same power envelope. Of course, we can’t eliminate all of them, but some factors, like temperature, change so slowly that you can shrink the safety margin by making it dynamic.

The simulated universe hypothesis is based on the seeming odd coincidence that our universe’s operation looks identical to information theory.

The problem with that hypothesis is that people seem to forget that our concept of information theory is a function of the universe it was developed in. Thus, it’s no coincidence, and the congruence of physics and information theory is not evidence of simulation.

What I want to know is, what kind of applicant pool do these companies have. If their hiring diversity is the same as their applicant pool, there’s not all that much they can do except maybe try harder to recruit in communities with higher proportions of minorities. If the minority applicants that they get aren’t as well qualified (objectively), we shouldn’t encourage them to hire less qualified applicants. Anything else would be reverse discrimination, which would also be wrong.

Maintaining higher diversity avoids a monoculture and increases the diversity of thought, which is good for problem solving. But you can’t squeeze blood from a stone. (Well, except for blood stones.)

They can be compared in that there are ethical considerations in both cases. As I said, abusing the supercomputer is a much more extreme case. In many ways, my examples are victimless crimes, while the supercomputer case had a far more tangible impact. In a relative moral scale, the supercomuting case was much more severe and would therefore have a more severe penalty. My whole point, I guess, is that even victimless crimes are cases where an ethical person should think twice before taking action.

I do feel compelled to point out that “victimless crime” is a loaded term that is abused by some people who want to poke their noses where they don’t belong. Some people would, for instance, want to say that growing your own weed and smoking it is a “victimless crime.” I’m not sure what the current laws are, but since this doesn’t involve interstate commerce, it’s not a crime at all, and it might be called a crime in the first place only because someone objects in general to smoking pot. Even if it were a crime, technically, I think the ethics in this case depend on the broader impact. If smoking pot improves your over-all wellbeing and doesn’t negatively impact your functioning in society, then it’s probably a good thing. If you’re neglecting your kids because of it, then it’s wrong. The ethical failure, however, is not in smoking pot but in failing to moderate the impact of your choices — it’s just as bad to neglect your kids playing video games online.

In the case of abusing equipment that was bought by tax-payer money, even if it’s “victimless,” it’s still unethical because you’re acting beyond your rights with respect to this asset that you have been trusted wtih. In other words, the ethical failure is not in the use of the equipment, per se, but rather in a breach of trust with respect to how you’re expected to use it. It’s one thing to borrow the company truck to go grab lunch. It’s entirely another to borrow company A’s truck to go do consulting work for company B, even if you’re not on company A’s clock at the time and you use your own money to fill the gas tank.

The abuse of the supercomputer is an extreme case. But there are other less clear-cut areas. For instance:

- What if I bring my own computer to the university and use their electricity to generate bitcoins?
- What if I bring university-owned equipment (that I have control over) home and use it to mine bitcoins on my electricity?

In either case, something that doesn’t really belong to me (even if I’m in charge of it and have the right to relocate) is being used for profit in a way that is (a) most likely against policy, and (b) not ethical in the first place.

The latter category is the really tempting one. Nobody would catch me, because all the network traffic and electricty usage is at my own home. Any impact on the longevity of the equiment is moot because it would probably go obsolete long before it suffered hardware failure. And of course, I can claim that I’m taking it home for official purposes (nobody would question me anyhow). This is one of those cases where you have to let your sense of right and wrong take precedence over the fact that you're clever enoug to not get caught.

This guy must have been reading the recent stuff on Fortran and decided to jump on the bandwagon.

Fortran was written by engineers and scientists for engineers and scientists.
R is written by statisticians for statisticians.

Well, there you have it. If a language or other kind of tool was developed by practitioners of X for other practitioners of X, it’s likely that it will be better than some other tool that was designed for a different purpose.

Who would have thunk it.

Computers do MATH (and data movement, etc.) really really fast compared to humans. But then again neurons do all sorts of low-level operations really fast too compared to the timescale we tend to think in at a high-level. What we don’t have are algorithms that are both fast and accurate for things like vision and speech recognition, MUCH LESS some form of cognition. (Yes, automatic speech recognition and computer vision are very complex and capable, but they pale in comparison to what humans can do. Imagine a computer trying to make sense of this image: http://accidentalblogger.typepad.com/.a/6a00d8341c575d53ef0163061e18fa970d-pi). Despite what Ray Kurtzweil wants to imply, having the compute power equvalent to the human brain will not magically cause computers to become conscious. We don’t know how to do that, and without that knowledge, we can’t write the necessary code.

So this timescale thing is bullshit. IBM Watson is amazing, but it doesn’t really think a lot faster than a human, in practical terms, and it isn’t exactly something you can have a philosophical conversation with. (BTW, on a completely coincidental note, I work for the Thomas J. Watson school of engineering.)

I've seen a number of papers that squeeze more performance out of x86 processors by being micro-op aware. Some just more carefully choose among instruction sequences, given awareness of the micro-ops that will be generated, while others consider what-if scenarios over what more performance you could get if you could have more control over the micro-ops themselves. For instance, given an ISA sequence and the corresponding micro-op sequence, is there a functionally equivalent alternative micro-op sequence? To some degree, compiler writers have to reverse-engineer the micro-op generation in order to generate micro-op sequences that schedule better. Even Intel compiler writers do this; although they have access to exactly what micro-ops are generated it's still not trivial to get the optimal micro-op sequence.

Actually, a Microsoft engineer developed the 64-bit extension, and AMD adopted it.

This is an honest question. How does the overhead of having all apps written in Javascript affect battery life? There are tools to compile Android apps to native indtead of Dalvik, and the perormance boost is substantial. I’d expect that the performance comparison between Javascript and native would be orders of magnitude. Now, I realize that most of time, phones are either idle or asleep, but all that extra CPU time for every interactive event has got to add up.

Actually, it’s still true that FOSS developers don’t really focus on usability. However, there has been a very gradual pace of evolution, and FOSS has more room to try out new (albeit mostly bad) ideas. Slowly but surely, the good ideas rise to the top. Meanwhile, Microsoft has mostly stood still (cosmetics and METRO don’t count), and as a result, an Ubuntu desktop with LibreOffice is at least as usable as Windows for any cases where you don’t need to install special software (e.g. Photoshop). Where I work (a university), all the computer labs in the CS department run Debian, and nobody has any complaints (that I know of).

Personally, I’d rather use a Mac over either Linux or Windows, but Apple doesn’t address the low end very well, making it probably more expensive to outfit a whole organization. Sure, for a given Mac, an equivalent PC has about the same features and cost, but you can get even cheaper PCs, which are adequate for light office use. I’m not sure that MacOS’s usability superiority (which is a subjective thing anyway) is quite enough to offset other costs.

I swear I remember doing this exact thing when I was a kid.