The concept of "Many-Core" and the obstacles to achieving it. Plus using silicon to boost wireless performance.
Slightly raining on the multiple processor vision, or “Many Core” as Weinrib referred to them (a phrase I would suggest you will hear with increasing regularity), was the fact he was always forced to come back and remind us of their dependence on software.
Domain-Specific Parallel Programming, ie: software that efficiently allocates different tasks to the different cores, is absolutely fundamental to the Many Core vision. Without it Weinrib said, you’ll see little better performance from having two cores over one (something we are currently witnessing from software not designed for dual core systems) and poor proportional performance even from the likes of future four, eight and sixteen way systems.
The message: the software must move at pace with the hardware. We don’t want another delay like we got with 64bit. “There is no point escaping one bottleneck to run into another” he added. Microsoft take note!
Unfortunately, limited flight availability was about to put the kibosh on the other two talks scheduled (thankfully they were going increasingly off consumer issues), but it is worth mentioning a few points made by Soumyanath Kurshnagurumurphy who was also back that day. He was moving through communications history at incredible pace, starting with Anderson and Marcone and ending up in the here and now in less than the space of five minutes.
He had a point though, which was to say that nearly 100 years from the first radio transmissions to the WiFi of today we are still using the same basic analogue principles. “Marcone was a smart guy,” Kurshnagurumurphy said, “but he shouldn’t be able to look at where we have got to today and be able to recognise basic similarities, which he could”. Kurshnagurumurphy proposed an entirely different direction: to use silicon to digitally assist analog: “It should be fine tuning the signal, helping to maintain the quality and the frequency,” he explained. ”We should not be worrying that moving to the other side of a room will mess up our signal”.
Just like speaking with Pankiewicz the previous day, it seems the original technological breakthrough is not the problem, the consumer preparation is. Kurshnagurumurphy said Intel already has technology in its labs to do what he proposes, but that miniaturisation, power efficiency and production costs are the big challenges. “If I had it here I could plug it into your laptop now and show you how it can deplete a battery in 20 minutes,” he joked.
It was time to be whisked away to the airport. One slightly intimidating short flight to Copenhagen on an ancient Dehavilland Dash 8-400 Turboprop (above – for the benefit of my Uncle John, who was a pilot) and a jet to London and I was home. Ready to strap icepacks to my technologically overloaded brain.