Quantum Computing – A Compute Odyssey

Silicon is here to stay for a long time.

• Everyone is getting excited again about the prospect of Quantum Computing providing the almost limitless compute needed for super-intelligent machines, but the reality remains that this is a single step on a very long journey that will keep silicon in the driver’s seat for many years to come and way beyond 2030.
• Google has published a research paper in Nature detailing the creation of a new type of Quantum Computing chip that Google claims is far more stable than previous generations.
• This is important because greater stability means a much lower error rate meaning that more of the resource can be used for calculations rather than discarded.
• The research paper (see here) is a demonstration of a theory that was proposed back in the early 1990s and has taken over 30 years to be demonstrated in a physical system.
• This should cause the cool-aid drinkers and technology media to pause for thought in terms of what the realistic outlook for quantum computing going mainstream is.
• Publication in Nature is important because this paper has been reviewed by a panel of Google’s peers and rivals meaning that it is much more than the marketing-press-release-dressed-up-as-a-scientific-paper technique preferred by Open AI and its peers.
• The improvement in performance achievable is jaw-dropping as the paper claims that its new chip can complete in minutes a task that would take a normal supercomputer 10²⁵ years.
• 5 years ago, Google made a similar claim but this time the task that took minutes would have taken a normal supercomputer 10⁴ years.
• This represents a mind-boggling increase (10¹⁹x) in performance and promises to usher in an era of virtually infinite compute power at a very low cost per unit of compute.
• However, there is a catch which is that all of this horsepower is completely useless for any task other than very specific types of mathematical operations which have no real practical use.
• Furthermore, if one creates a translation system to enable practical tasks to be carried out on these machines, all of the benefits over normal supercomputers drain away leaving the quantum machine as inefficient and expensive.
• Therefore, the real problems of quantum computing stretch far beyond achieving performance and are mainly centred around making practical and economical use of it.
• Consequently, I see the technical problems being overcome now as being analogous to the first attempts to migrate from vacuum tubes to transistors back in the 1950s and 1960s.
• Even though Moore’s Law has long ground to a halt, silicon looks like it is going to be the medium of choice for compute processing for many years, if not decades to come.
• Hence, this is not the end of TSMC, Nvidia et al nor is Google about to become the sole owner of the one computer that can outdo all others.
• Instead, what we are witnessing are the first steps being taken on the quantum computing odyssey that I think has a very good chance of one-day replacing silicon in its entirety.
• Whether I am around to witness this transition is another question, meaning that for all intents and purposes, quantum computing remains almost un-investible.