Computing technology has traditionally led a path where the hardware advances to a point where it can SCALE before the SOFTWARE CATCHES UP. Right after the point at which a new “scale point” has been reached – the point at which a new generation of technology based on some sort of advance in making technology scalable in a new way – a new generation of software is released to take advantage of the new hardware. The hardware takes advantage of that scaling, and then optimizes for a new price versus performance inflection point in the market based on whatever new limits are reached in the current state of the hardware.
One such example of this was the move to the 64bit CPU, which required 64bit drivers for hardware on a 64bit operating system. If you custom built a PC back then and 64bit drivers weren’t available, you had to run a 32bit OS. The 64bit CPU was made available, then the hardware caught up to support it, and in this case, major CPU architectures for desktop use have since standardized on 64bit. Even the ARM architecture is 64bit capable.
Another example, although security related, perhaps, was the transition of the Trusted Platform Module (TPM) from version 1.2 to 2.0. Whether or not the TPM version change drove a new Windows release, or if it was the other way around, and while you can bypass the TPM 2.0 requirement in an installation of Windows 11, security vulnerabilities are not always software related, and can be caused by hardware including the CPU. This still stands as an example where hardware updates can require software updates to use new functionalities.
As computer hardware becomes increasingly complex, CPUs and GPUs have merged already, just like floating point co-processors were merged into CPUs, and now new CPUs with AI features (Neural Processing Units or NPUs integrated into them, greatly optimizing AI performance), software updates will level out. These layering of technologies simplify the functionality of the components into easier to understand concepts (although they also do this because of the vastly independent way each layer works in with respect to the hardware required to perform each layers function).
With computing technology becoming exponentially popular due to the internet, smart phones and cellular technology, and the overall price to get connected very low, it’s safe to say there is an exponential increase in computing demand. The problem is that we too have had an exponential increase in demand for computation tasks on hardware which is not par for the job. Like upgrading from Windows 10 on a PC to Windows 11, it felt slow. The limit for technology is the hardware, and we have hit a roadblock. Stop investing in GPUs. Invest in NPUs.