D2D: Handheld gaming and how Hardware companies can build software
And of course a lot about AI
Just a reminder that we operate our newsletters on a paid model. Paid subscribers will get three newsletters a month including our China Deep Tech notes and audio recordings of our posts. Paid subscribers will also get early access to the newsletter as well other benefits coming soon. Please subscribe and support our work.
Highlights from our Blog
Handheld gaming is an emerging category of consumer device. Smaller than a laptop, but with an equivalent amount of compute power, these devices have proven very popular with users mostly younger than the demographics of this newsletter. The gaming industry has seen the success of the Nintendo Switch and the “success” of the Steam Deck and sneses an opportunity. This could be an interesting segment for the electronics makers, but, as always, a lot will depend on software.
Unity Technology has had a rough month. The launch of their new pricing scheme sparked a broad, ugly customer revolt, forcing the company to backtrack. We are big fans of the company’s technology, but at the time of the IPO we cautioned that they like to talk a lot about ‘vision’ while their real challenge will be execution. And they have not been executing well.
Electronics company CEOs like to talk about software, but actually implementing a software side to their business often proves much more challenging. We have a few suggestions for how to actually put one of these in place.
If you like this content, you should check out our podcast The Circuit
Semis and Hardware
Intel is spinning off its FPGA unit, the company formerly known as Altera. This is a tricky move. On the one hand, by spinning it off, Intel can sell down its stake and generate some much needed cash to spend on building fabs. On the other hand, FPGAs are an important category, and long-term the company could have built on that. To be brutally honest, Intel has not done a great job running Altera, and returning the company to independence should go a long way to turning around that ship. Interesting that Intel was not able to sell it, which would have been a faster path to cash.
The EU is building a super-computer based on Arm. This would have been unthinkable as little as five years ago. Arm has come a long way to be this competitive in this field.
A big part of Nvidia’s success in AI is rooted in its CUDA software layer. Everyone in the field is exploring alternatives. Chief among those is AMD with their ROCm software. There are signs that ROCm’s appeal is catching on, with this start-up building a fairly large model and then going online to publicly extoll AMD. CUDA’s moat is not insurmountable. And while all of this is true, Nvidia is no longer dependent on CUDA.
A deep dive into Qualcomm’s Hexagon DSP. This functionality has long been a mainstay of Qualcomm’s modem architecture, but is now being tasked with boosting AI inference.
Apple is end-of-lifing its $17,000 gold watch. This is has always been a risk for expensive electronics, and reminds us of Nokia’s Vertu line of mega-expensive, jewel-encrusted feature phones which aged very poorly in the time of smartphones. Apple should start a subscription model - $100,000 for a watch which gets replaced every year. We think everyone would be surprised at how many takers there would be for this.
For years, the smart move for start-ups was to avoid selling to the government. Government procurement contracts can take longer than any start-up’s cash runway. That may be changing. The sector, especially for anything around the Defense Department, has gotten a lot more interesting. And while the contracting cycles still take a long time, the Defense Department is much more motivated to work with start-ups and acquire the latest technology.
Networking and Wireless
The team at Google published an academic paper explaining how they built their optical network. This is an incredible achievement and gives Google a big advantage, but we also wonder how brittle this system will prove to be, and how Google will be able to adapt it as networking fabrics and workloads evolve.
An interesting analysis of how Blackberry failed. Another good take on this came with the movie Blackberry, which we highly recommend both for its take on the tale but also because it is a very entertaining movie. The movie actually provides the simplest and best explanation for what happened to the once iconic device. Blackberry built its fortunes on minimizing data loading of cellular networks, but that lost its luster when operators moved from “charging per minute, to charging per bit”.
We have long been fascinated by networking software company Nutanix, but it has suffered a long struggle to really grow into its strategy. They recently held an analyst day which highlighted the progress they have made, and how far they still have to go.
Networking for AI workloads is emerging as an important battleground area. As AI models grow ever larger, the need to connect all those GPUs becomes a larger challenge. Nvidia is pushing its low-latency Infiniband solution, but faces significant resistance from the hyperscalers who have gotten very good at maximizing their Ethernet capabilities. Meta, among others, is pushing hard to rejuvenate Ethernet.
Dean Bubley’s take on the timing of 6G is worth bookmarking.
Software and the Cloud
Most games today rely heavily on online competition, but building software that scales across dozens (hundreds, thousands) of simultaneous users sharing a common gamespace is an incredible technical achievement. On a related note, here is a big list of game engines, not exactly alternatives to Unity, but fascinating to explore this side of gaming software.
Passive solar-powered water desalination, using easy to procure components and utilizing the revolutionary technology of resting the box at an angle. Very useful if it works.
The South Atlantic island of St. Helena, final residence of Napoleon, is among the most remote places in the world. It now has its first wired Internet link.