The SURE house used 9100 kwh for a year compared to 130,000 kwh for the average NJ house. That is obviously quite the improvement, but my 1970s rambler, 25% larger, only used 27,000 kwh last year - in North Dakota. About 18,000 kwh of that was for space heating. My monthly energy usage beat the SURE house three months out of the year.

I guess it depends on what your definition of output is. For a math algorithm, you generally wouldn't expect a change. There are still some exceptions to that. The compiler vendor could correct a bug in a library that your code took into account. Various libraries could become more compliant to a standard, or be upgraded to a newer version of the standard and have very subtle differences. Ideally your unit tests would catch this, but you would have to run your test on target, with the actual libraries. This is easier said than done in some embedded systems.

Another example, you may be depending on the fact that a calculation takes X amount of time for you to go off and do something else or the time it takes to run may be an integral part of your output timing. Ideally, you don't want to code things that way, but sometimes you have to due to hardware constraints. Optimizations can cause things like that to break.

Regression tests are never perfect. I recently debugged an issue where if we received an interrupt during a one instruction time window, the system had an issue. It worked fine on an older compiler but we were forced to upgrade for other reasons. The new compiler generated code that now had the issue due to improved optimizations that reordered the code. There is no reasonable way to unit test that kind of issue. It required the entire system to be running, and running on target.

Compilers improve their optimization algorithms over time. I've personally helped debug issues due to this on a number of occasions.

I wouldn't say that I've had a lot of long conversations in the men's room at work(in the US), but over the years I've had plenty of "Could you stop by my desk and look at what I am working on?" sort of interactions.

Manufacturing constraints are a big one. There are a number of other big ones people rarely take into account. Things like ESD protection, robustness over a wide temperature range, parts going obsolete, emissions and susceptibility, regulatory requirements. I spend a huge chunk of my time dealing with these kinds of concerns and I am mostly on the software side of aisle.

I mostly write C and C++ code these days, but I've worked a lot with guys using Simulink and Labview. Both are good tools, but like any "programming" language, there are issues. Some of them stem from the lack of software processes being used by people who have never been exposed to a typical software development process. Simple things like version control, or "code" reviews. My experience with Labview in particular, is that the ease of changing the graphical representation encourages people to quickly tweak something to make it work. I've ended up in the situation where I asked what they changed to make it work, and they can't tell me. This is as much a lack of rigor in the process as it is in the people using the tool. They can often do amazing things, and quickly, but they tend to be difficult to maintain.

My experience with Simulink is a little different. The guys working with Simulink tend to be domain experts, often with PhDs. They really know their stuff. When they can generate a good model of the system they can do AMAZING things. The downside is you can spend millions of dollars developing and validating the system models. It is why you see Simulink used in projects with tough control systems problems, which coincidentally often have large budgets, like the automotive world.
Sometimes, the code generated by SImulink isn't efficient, which can drive up memory/CPU costs. When you are shipping hundreds of thousands or millions of a product a year, adding a few cents per unit really adds up. Sometimes the cost of spending an extra million or two writing the code in C(or optimizing the Simulink after the fact) is worth the expense. My opinion is that tools like Simulink will eventually take over the control systems market.

Another reason is cost. You can get a C compiler for free. Matlab/Simulink is a 6-figure expense per developer. Labview isn't that expensive, but it is still not cheap.

You may not be working with a signal that is in the 1 GHz range, but your rise/fall times may be in that range and if you want to verify that you meet timing requirements, you will need a higher speed scope.

Cash is faster? You must be joking. For the vast majority of my credit card transactions these days, I swipe the card while they are ringing up my purchases and walk away as soon as they finish. Most of my charges are under $50, so most of the places I shop don't even require a signature. Even when they do, my signature takes far less time than handing them cash, them fiddling around getting me change out of the drawer and handing it back.

25/3 Mbps for $50? I wish! I get 5 up, 0.5 down for $53.50 a month. I have exactly one broadband option. Aren't monopolies great?

I think you should go back and do some actual calculations. About 30 seconds of googling tells me that standard, commercially available solar panels for making roughly 700 kWh a month would cover about 400 square feet. The combined area of my garage and relatively small house is over 2000 square feet.

The battery back on the base Tesla S is a 40 kWh battery pack. With a 400 square foot system, it should produce enough energy to charge a Tesla battery pack about 17 times in a month. That should get you about 2500 miles in a month.

Seems like plenty of room on my roof to charge an electric car, if I wanted to. I would just need to solve the problem of my car not being there during the day when the panels produce most of their energy.

So computer science, and electrical engineers should learn base 2, base 16, and maybe base 8. I would even go so far as to make everyone learn converting between base 10 and one other base so they learn the concept. 99% of people are never going to need to do base conversions, and of the remaining 1%, they are pretty damn unlikely to need to convert between base 5 and base 7. Humans work with base 10 for good reason.

Uggg. I had several teachers in college that wrote their own "textbooks" for their classes(electrical engineering). They were extraordinarily smart individuals, but their writing sucked. They were desperately in need of a technical writer and an editor. The ones that didn't completely suck were not any better than the normal books I had for my other classes.

When I added onto my house, they of course used the arc-fault breakers. About one out of every three times you shut off the vacuum cleaner when it is plugged into one of these outlets, it trips. Two different corded drills do the same thing, only they are closer to 50% of the time. I replaced the breaker with a different one. Same thing. Very annoying.

These days, I'm not even sure a paper receipt will last 4 years. I've got a bunch of them that have faded significantly in a years time.