Matisyahu's Blog

I’ve been following what the x86 working group have agreed upon and one of the big changes was the inclusion of FRED that has made an appearance in Panther Lake and will be appearing in the Zen 6 based CPUs rumoured to be coming out in late 2026 or early 2027. At the moment I haven’t seen any discussions over whether FRED made its way into Windows yet (we may not see it until 27H2 with the new system based around 29000) however Linux have had the necessary patches merged with the kernel.

Over on Phoronix they have released a FRED benchmark with the latest Linux kernel (link) and the results so far look very positive. It would be interesting to see how over time the performance improves once APX is merged along with memory memory tagging. Although there was an attempt at a more radical change with the x86S the decision was to create the x86 Advisory Group where they probably took many of the ideas explored with x86S and see how they could deliver those changes without being disruptive. As others have pointed out, the cost of legacy support is very small and most modern operating systems don’t even touch the old legacy stuff.

I love seeing Ed Zitron pop the hype cycle when it comes to AI – every time I see what is happening in the AI space I can’t help but get the feeling of the accounting scandals of the 2000s. Promises of datacenters and funding but not actual contracts being signed, GPUs being shipped to datacenters but not being plugged in because there isn’t the infrastructure needed. Then add to that is emerging technology in the form of TPUs that Google have demonstrated to be more efficient at inference work as well as building the LLMs thus questions about the long term viability of brute forcing GPUs when there can be bespoke hardware that can do it more efficiently in much the same wat that bitcoin mining now uses Application-Specific Integrated Circuit (ASIC) instead of GPUs because they’re more efficient.

The thing that is also ignored is the fact that AI is a means to an end not an end in itself – you integrate it into your products to make those product do things better. No one buys AI for AI sake but they do so because it enables them to do something that either they couldn’t do it before or if they could do it before it was a laborious process., I also question long term whether many of these stand along AI companies will last long term and I would hazard to guess that many who are ingesting into them are making the big gamble that a bigger player will buy them out in much the same way that Loopt was bought out by Green Dot. One thing to keep in mind – AI is more than just LLMs and there is plenty of legit and interesting work that is being done and we’re still very much at the beginning but unfortunately far too many on Wall Street appear to lack a well tuned BS detector to pick up when promises are being made with very little grounding in reality.