Improved computer performance has been driving the boom in deep learning and artificial intelligence.
50% of the productivity growth for the world for the past 40 years in the US and the world came from the massive improvements in computer performance.
It is vital to make the investments to continue the rate of computer hardware performance to maintain economic growth.
Vastly faster hardware is driving progress in AI
SOURCES- DARPA, Youtube, Dale Jorgenson
Written By Christina Wong
Product Leader who is passionate about turning “what’s impossible” into innovative realities. Building a new digital SaaS platform to transform the $12+ billion CiscoCustomer Experience business.
8 thoughts on “Economic Growth and AI Progress Depends on Better Hardware”
Good point on the watt.hour.
That point had somehow escaped my notice, even though I knew the basic facts that go into the relationship. But I somehow never put it together.
The inferior serves no useful function other than to bootstrap that which is superior. After which, the continued existence of the inferior is a detriment to the new system.
The actual factor that we are relying on is flops/$. Not features/nm.
So humanity has looked for alternatives such as multicore chips, quantum computing, spintronics etc.
Of course we haven’s seen any of these work yet. I mean work in the sense of being available in the shops with higher performance/$.
You go for energy efficiency using Non-Von Neumann architectures so you can go 3 dimensional without overheating. You liquid cool. Eventually you basically end up with a human brain.
Arguably FLOPS per watt-hour are now a more important metric, because over the life of a server machine that’s now your biggest cost.
Non-Von Neumann architectures will get us another 200x better in FLOPS efficiency at most, even past this death of Moore’s Law.
Very little beyond that without essentially copying living brains, though.
If that doubles every 2 years we hit another wall at 15 years from now.
Wonder what’s gonna happen when hardware power runs up against the cold, hard laws of physics? We can’t get much smaller, and physics always wins in the end.
Of course, humanity never thought of a contingency plan for this.
Obligatory:
A clever line I once saw said that the last quarter of the 20th century, and maybe the first half of the 21st, can be described as a race between Moore’s law and medicare.
Will the ever increasing productivity of high tech (represented by Moore’s law) stay ahead of the every growing costs of our aging, more demanding society (represented by medicare).
Up till now, technological growth has (just) stayed ahead. But now the champion Moore’s law has ground to a halt and realistically that’s the scariest thing in the world right now.
All the other “scary monsters” out there. Pollution, poverty, plagues, even war. They were all being held at bay and many were gradually being overcome because tech was pulling ahead of costs. Without a steady stream of new tech, we’re in trouble.
And all our new tech. Electric cars, Space flight, genetics, medical breakthroughs, telecommunications… they are all depending and accelerated by the ever falling cost of computation that is based, in the end, on the fact that chip capability/$ doubles every couple of years.
Good point on the watt.hour.
That point had somehow escaped my notice, even though I knew the basic facts that go into the relationship. But I somehow never put it together.
The inferior serves no useful function other than to bootstrap that which is superior. After which, the continued existence of the inferior is a detriment to the new system.
The actual factor that we are relying on is flops/$. Not features/nm.
So humanity has looked for alternatives such as multicore chips, quantum computing, spintronics etc.
Of course we haven’s seen any of these work yet. I mean work in the sense of being available in the shops with higher performance/$.
You go for energy efficiency using Non-Von Neumann architectures so you can go 3 dimensional without overheating. You liquid cool. Eventually you basically end up with a human brain.
Arguably FLOPS per watt-hour are now a more important metric, because over the life of a server machine that’s now your biggest cost.
Non-Von Neumann architectures will get us another 200x better in FLOPS efficiency at most, even past this death of Moore’s Law.
Very little beyond that without essentially copying living brains, though.
If that doubles every 2 years we hit another wall at 15 years from now.
Wonder what’s gonna happen when hardware power runs up against the cold, hard laws of physics? We can’t get much smaller, and physics always wins in the end.
Of course, humanity never thought of a contingency plan for this.
Obligatory:
A clever line I once saw said that the last quarter of the 20th century, and maybe the first half of the 21st, can be described as a race between Moore’s law and medicare.
Will the ever increasing productivity of high tech (represented by Moore’s law) stay ahead of the every growing costs of our aging, more demanding society (represented by medicare).
Up till now, technological growth has (just) stayed ahead. But now the champion Moore’s law has ground to a halt and realistically that’s the scariest thing in the world right now.
All the other “scary monsters” out there. Pollution, poverty, plagues, even war. They were all being held at bay and many were gradually being overcome because tech was pulling ahead of costs. Without a steady stream of new tech, we’re in trouble.
And all our new tech. Electric cars, Space flight, genetics, medical breakthroughs, telecommunications… they are all depending and accelerated by the ever falling cost of computation that is based, in the end, on the fact that chip capability/$ doubles every couple of years.