Is Your Next Computer Already Outdated?
Weekly Wheaties #2612
In this newsletter:
📝 Post: Is Your Next Computer Already Outdated?
🗞️ In Case You Missed It: Tech and AI Headlines
😎 Pick of the Week: Guilty Food Pleasures
📦 Featured Product: Brother Laser Printers
📝 Is Your Next Computer Already Outdated?
In the last few years, with aggressive investments in chip production to keep up with the artificial intelligence computing power needed, we’ve seen chip manufacturers working harder and harder not only to speed things up, but simply to keep up with demand.
However, AI, at least from the software side of things, appears to be following a hardware rule - Moore’s Law. This simply states: the number of transistors that can be packed onto a single integrated circuit (chip) double every two years. As I’ve talked a little about how CPUs are manufactured and why they get slower over time the last two weeks, Moore’s Law is the push behind the manufacturing aspect of this.
Although it’s not a physical law but rather an industry trend, it has seemed to slow in recent years as engineers approach the physical limits of manufacturing on the wafers. There are companies trying new things all the time, with one of the most recent examples being Apple bringing its chip manufacturing in-house. Essentially, the low-hanging fruit win (a physically larger chip) has physics issues to deal with - heat from more power needed, latency as the pieces get further apart, but mainly due to more defects showing up the larger the chip gets.
What we’ve actually witnessed in the historical context of technology can happen in a few ways to benefit the business and consumer sides of this.
First, the chips can actually be built in new innovative ways to compute faster. You may have even heard of the nanometer process before. The size has become smaller and smaller over the years. There are tons of manufacturing processes that allow this, but let’s just focus on the fact that the mixture of speed and/or power in regard to manufacturing affects the computation speed.
Second, these manufacturing processes, or the physical product used to create the chips, can become exponentially cheaper. Meaning, the cost to produce twice the computing power today costs the same or less than it did for the power of just one chip (2 years ago). This chart from the NVIDIA keynote last week shows how power and performance affect cost.
Third, and similarly to the previous example, the power required to produce the same compute can be halved. This can be due to a new manufacturing process, or even a new type (or cheaper) power source. This is why we’ve seen many larger companies (that require tons of computing power) invest in remote areas where electricity costs are lower, or the gain from solar power is higher than in their current location. We’ve seen recent examples - just in Louisiana - of this with Amazon, Meta, Hut8, and many more.
Next, this is kind of a cheat, but one other way to allow CPUs to keep doubling is simply a mixture of the above mentioned ways. When combined together and produced in massive quantities, every little bit adds up, and can help save money. We’ve seen this in play over the last few decades. Phones and laptops are simultaneously getting smaller, faster, and using less power (assuming a steady price point). Alternatively, TVs are becoming larger and less expensive, while offering better and better quality. So the technology innovations are still coming at lightning speed.
But there is one more side to this argument - the software side. While faster computing is allowing AI and softwares to keep getting faster and faster, developers are figuring out how to optimize their code. Older systems are being hacked to run newer softwares. A fun test of this is when developers are hacking newer (very basic) hardware to run the computer game Doom. This is only done for fun, but it has become a staple hack to show what these products are capable of. An older (and confusing) example of this is when a programmer hacked a pregnancy test.
When ChatGPT was originally released for the wider public, the computing power required for a basic prompt would satisfy hundreds of prompts for the newest models. At least for their basic/fast model. This is also exponentially better than the earlier version mentioned. LLMs have also become more innovative in how they operate, with many models offering a version that can be downloaded and used on a single laptop — without an internet connection. The same model would have required a server farm and internet access 5 years ago, but the newest iPhone has enough power to run it locally. The Google AI Edge Gallery app mentioned last week is a perfect example of this.
When hardware and software can innovate together, crazy cool things can happen. Moore’s Law appears to still ring true, but time will tell for how long. Or at least in what way. Until then, we’ll still seemingly be paying the same price for a computer that could theoretically be twice as fast for the same price in two years. Keep that in mind next time you purchase your next big tech product!
Which leads to the question: If computers keep getting twice as powerful every few years, when is the “right” time to upgrade?
🗞️ ICYMI: Tech and AI Headlines
Amazon Employees Say AI Is Just Increasing Workload. A New Study Confirms Their Suspicions
Apple unveils AirPods Max 2 with H2 chip, upgraded noise canceling, and more
Introducing GPT‑5.4 mini and nano - Fast and efficient models optimized for coding and subagents.
Introducing the new full-stack vibe coding experience in Google AI Studio - a quick way to build web-based apps. Check out this Magic Coloring Book I built in under a minute, with no coding required!
Google Fiber will be sold to a private equity firm and merge with a cable company
😎 POTW: Guilty Food Pleasures
Do you have a go-to food item or local spot in town? Perhaps some of these can spur an idea or encourage you to visit a local restaurant!
Reheat Leftover Pizza With These Techniques For Maximum Flavor
The Hands Down Best Hole-In-The-Wall Restaurant In Every US State
📦 Featured Product
If you’ve ever asked me for IT help in the past, chances are I prefaced each interaction with, “I don’t do printers…” Hate is a strong word, but I think it’s okay for me to use it towards printers. I’m also very digital, and only print when I have to. However, if you are in need of a printer for your home, I do have two suggestions, depending on whether you need full color or are okay with black and white only. If this is for your business, just go ahead and pay someone else to handle your printers, and possibly even rent one. It’s worth it! If anything, I typically suggest staying away from the ink jet style. Which leaves the more expensive, but longer-lasting, laser printer version. Brother is a very well respected brand here, and the basic HL-L2405W Wireless Compact Black & White Printer can’t be beat. In their web store, they offer other versions with scanning capabilities and full color, but there are a few to choose from, so I’ll just let you browse.
