bcrl

What everyone forgets is just how expensive context switches are on modern x86 CPUs. Those 512 bit vector registers fill up a lot of cache lines. That's why async tends to win over processes / threads for many workloads.

100 nanoseconds from RAM is correct. Latency != bandwidth. 3 nanoseconds would be from cache or so on a Ryzen. You ain't gonna get the benefits of prefetching on the first 100 bytes.

The main difference between algorithms and humans is that software feels no guilt. Traditional human building superintendents were once happy with a 10% increase in rent from one tenant to the next, and would feel guilty when doubling prices. There are plenty of small business owners which take pride in delivering affordable prices to their customers across many market segments. Not a trait that is present in large corporations.

One of the Behind the Bastards podcasts touched upon the fact that in the rental property market the 2 cloud vendors peddling software to manage properties could collude on behalf of landlords. Collusion by humans is fairly limited in scale, but when you're a "platform" every price can be set based on the prices of millions of listings -- what was once impossible for humans is now trivial.

I feel obliged to link to the Behind the Bastards podcast "Why is rent so damn high?" It's a lengthy 2 part piece, but quite enlightening. https://podcasts.apple.com/us/podcast/part-one-why-is-the-re...

Personally, I far prefer black over grey. Grey is really hard to read across a variety of lighting conditions and devices. The older you get, the more important contrast becomes.

Rogers is perhaps best described as a confederacy of independent acquisitions. In working with their sales team, I have had to tell them where there facilities are as the sales engineers don't always know about all of the assets that Rogers owns.

There's also the insistence that Rogers employees should use Rogers services. Paying for every Rogers employee to have Bell cell phone would not sit well with their executives.

That the risk assessments of the changes being made to the router configuration were incorrect also contributed to the outage.

I remember having to figure out how port magazine and book BASIC code between the various dialects between the various spaces I spent time in. In the 1980s, my uncle first taught me BASIC on an IBM PC. At school we had an Apple ][+ while at home I had a Coco II and later an Amiga. Another friend had a Vic 20, while another had a Commodore 64. Then came QuickBASIC, QuickC, Microsoft C, C on the ICONs, Aztec C, gcc... 6502, m68k, 8086, i386... Learning about the quirks across systems so early on turned out to be an invaluable experience.

That is heavily contingent on the assumption that the dependencies between services are well documented and understood by the people building the systems.

Parker is what Larry McVoy advocated for Linux back during the early days of multiprocessor scaling work. The idea was basically to treat an MP system as a cluster. Everything old is new again!

Personally, I would never agree to give up SMP CPU coherency. Multiprocessor systems are hard enough to debug with hardware cache coherency that adding in entirely new unpredictable non-deterministic behaviour would lead to more developers losing the rest of their hair prematurely. And it would likely introduce an entirely new class of security issues that nobody ever imagined that would require even worse performance draining software workarounds.

Some things are best done in hardware.

That's similar to the total outage of all Rogers services in Canada back on July 7th 2022. It was compounded by the fact that the outage took out all Rogers cell phone service, making it impossible for Rogers employees to communicate with each other during the outage. A unified network means a unified failure mode.

Thankfully none of my 10 Gbps wavelengths were impacted. Oh did I appreciate my aversion to >= layer 2 services in my transport network!

This is the kind of article that deserves to be posted on April 1st, preferably with an accompanying RFC published at the IETF.

Utilities require long term planning. A 22% increase in 1 year is an extreme outlier by any normal grid planning standard. Where I live, the high point in the last 5 years was a 2.8% annual increase. ( See https://www.ieso.ca/power-data/demand-overview/historical-de... ) Building large amounts of new capacity takes years to decades.

If hyperscalers keep pushing that kind of increase they will eventually learn that the supply and cost of energy becomes a limiting factor. When demand outstrips supply, prices go bonkers and rolling blackouts become a fact of life. See the California grid crisis back in 2000-2001, although in that case the supply shortage was the result of blatant market manipulation by Enron.

It is economic. Significant amounts of advertising dollars were once spent on ads in local newspapers and TV stations. That is no longer the case. Expecting media content to remain high quality when the revenue stream that once supported investigative journalism has dried up makes no sense.

Any time you read or view a piece of content, follow the money trail that got it in front of your eyeballs, ears or fingertips -- it may well be enlightening.

*If the bandwidth of the Analog Front End (AFE) and Analog to Digital Converter (ADC) / Digital to Analog Converter (DAC) doubles as well. In the real world the AFE of any wifi radio has a fixed bandwidth, with the ADC sampling rate and accuracy being fixed as well. The end result is that doubling the channel width in a wireless network requires a received signal strength that is roughly 3 dB more in real world devices. This constraint is quite visible in data sheets for most wifi cards like here: https://compex.com.sg/wp-content/uploads/2024/01/wle7002e25-...

Almost nothing in tech is sustainable outside of gold recycling.

You may need a third access point in the middle of the house then which would enable you to further reduce tx power on the access points while maintaining good reception of client transmissions by the APs, as clients won't roam unless the other access point is of a sufficiently better signal strength. With only 2 access points the signal level in the middle of the house may be too low for either AP to maintain good reception. As always it depends on the materials in the walls, the placement of large metal obstacles like ductwork or appliances, the antenna orientation on the access points.... You would have to measure the signal levels and build up a heat map to figure that out.

Of course with 3 APs instead of 2 the best layout is different.

People are almost always better off more cheap access points than fewer expensive access points, but that's not how most regular folk reason.

Every doubling of the channel width costs roughly 3dB. Shannon's law strikes again!

DECT phones run on the 1.9 GHz spectrum which doesn't get absorbed by water like 2.4 GHz, and will penetrate through many other materials far better than higher frequencies.

Most people place wifi repeaters incorrectly, or invest in crappy repeater / mesh devices that do not multiple radios. A Wifi repeater or mesh device with a single radio by definition cuts your throughput in half for every hop.

I run an ISP. Customers always cheap out when it comes to their in home wireless networks while failing to understand the consequences of their choices (even when carefully explained to them).

The biggest obstacle to AGI is data.

Take the knowledge the average human has about integrating visual information with texture of an object. Nearly every adult can take a quick glance around a room and have a good idea what it will feel like to run your fingers along its surface, or your lips, or even your tongue, and be able to describe the experience. We have this knowledge because when we were infants and toddlers, everything we encountered was picked up, pulled towards our mouth, and touched by our hands. An AGI inside a computer cannot have that experience today, so it will lack the foundations of intelligence that humans have built up by interacting with the real world.

At some point it will become possible to either collect that data or simulate an experience sufficiently accurately to mimic the development a human child goes through. Until that happens, true AGI will be out of reach as it will have deficiencies the average human does not.

That said, a lot of people will try to get to that point using other means, and they'll probably get pretty close, albeit with really weird hallucinations in the corner cases.

Modern engineering exists because of the people that were maimed or died to give us the knowledge that lets us live the way we do today.

Licensing takes many forms. Some are onerous, some are not. Working At Heights training takes 0.5-1 day of training every 3 years where I live, and it's mandated because it saves lives. Yes, it's a form of licensing, but it's an easy box to check because it's accessible.

For example: running experiments on humans without their prior consent is considered unethical. However, large tech firms routinely run A / B tests on their users without providing Informed Consent. If software developers were trained like engineers, they would be ethically obligated to obtain Informed Consent prior to engaging in this kind of conduct. More importantly, when software developers realize they were being experimented upon (as happened in the Linux kernel community a while back), they were justifiably outraged.

I know this isn't going to be a popular position to hold here, but there's a lot of harm being done by unethical practices that are currently widespread in big tech. Ephemeral ads that prey on the elderly would be considered Not Good by anyone that has seen a parent fall prey to them, yet there doesn't seem to be any concern whatsoever amongst the industry giants themselves to prevent this practice. So long as an ad brings in money, it's good to run seems to be the bar for advertisers at present, and I don't think that's good for society.

When there are no ethical considerations given to the consequences of an action, unethical outcomes are inevitable. Fixing that starts with learning, which is something every great developer already does.

Ethics is applicable to many situations outside of life critical systems. Engineers take ethics very seriously because it is important, and I think it is important for software developers as well. The first time I worked closely with a Professional Engineer on a project, it was eye opening as to how the concern of doing things the ethical way shaped the interaction... Everything from identifying potential business conflicts to ensuring data collection for inputs to the design process was done correctly and traceable.

Corporations already push enough random HR requirements on tech employees. I can't see how having a semester or 2 of ethics courses is particularly onerous.

This is why I think that we're getting to the point where software developers need to be trained in ethics and licensed in the same way as professional engineers are around the world. The people building the tools need to be held accountable when design choices kill people.

Directional drilling is not a panacea. There are still places where the terrain is too rocky and aerial construction is the only way to control costs.

People still think they own cars, and that they should be allowed to repair them without the blessing of the manufacturer. They don't expect to be banned from using their car because someone decided that using the DMCA to pervert the concept of ownership was a good thing.

The number of transistors per unit area is still increasing, it's just a little slower than it was and more expensive than it was.

And there are innovations that will continue the scaling that Moore's law predicts. Take die stacking as an example. Even Intel had internal studies 20 years ago that showed there are significant performance and power improvements to be had in CPU cores by using 2 layers of transistors. AMD's X3D CPUs are now using technology that can stack extra dies onto a base die, but they're using it in the most basic of ways (only for cache). Going beyond cache to logic, die stacking allows reductions of wire length because more transisters with more layers of metal fit in a smaller space. That in turn improves performance and reduces power consumption.

The semiconductor industry isn't out of tricks just yet. There are still plenty of improvements coming in the next decade, and those improvements will benefit AI workloads far more than traditional CPUs.

Just because a bubble pops on the economic front doesn't mean the sector goes away. Pets.com went bust a mere 10 months after going public, yet we're buying all kinds of products online in 2025 that we weren't in 2000. A bubble popping is about the disconnect between the forward looking assumptions about profitability by the early adopters in the space versus the actual returns once the speculation settles down and is replaced by hard data.

Damn, that must be why they keep making it harder and harder for independently run mailing lists to deliver to gmail users.

This! The cost of training models inevitably goes down over time as FLOPS/$ and PB/$ increases relentlessly thanks to the exponential gains of Moore's law. Eventually we will end up with laptops and phones being Good Enough to run models locally. Once that happens, any competitor in the space that decides to actively support running locally will have operating costs that are a mere fraction of OpenAI's current business.

The pop of this bubble is going to be painful for a lot of people. Being too early to a market is just as bad as being too late, especially for something that can become a commodity due to a lack of moat.

Move fast and break things doesn't work when safety is involved. Did a properly trained and licensed Professional Engineer review the design of Amazon's drones, their safety features and protocols before they got approval to fly outside of private testing facilities? If not, who would insure that hot mess?