The Gordon Moore’s law vs actual transistor count
Found on twitter video: x.com
Because since there is not a central control, evolution at the moment is more physically unstoppable natural behavior rather than a human-controlled evolution. Of course, there is this tech limitation. But progress driven by evolution at the moment looks unstoppable.
For the game industry market: The presumption that the 2020 GA102 3080Ti will be around 40.000 million transistors, tot. And we need to learn and migrate to the job system to take advantage of the upcoming 128 upscaling core count…
I suspect that even if scaling is basically dead at what would be called 2-3 nm, there’s still significant optimization options left;
a) architectural (cpu vs gpu type of thing)
b) factory optimization ($ per die mm2 could go down significantly; imagine a 20x bigger chip running much dedicated silicon for random tasks that’s unpowered a lot of the time)
c) paradigm shift to something other than roughly 2D silicon chips
God forbid software people would improve their tools to write efficient code as well. Code bloats as much as hardware allows it to, so if hardware would be stagnant there would be more value in optimizing software (as hardware won’t make optimization ‘useless’ by brute forcing your shit program).
Moore’s law isn’t just about increasing the density of transistors. It’s about increasing the density while decreasing the cost per transistor. It’s because of this that there are two ways that Moore’s law can die rather than just the one way most people seem to believe.
Most people believe Moore’s law will end when increasing transistor density is no longer achievable, but in my opinion the actual death will be when increasing density is still viable but decreasing the cost isn’t. At some point we are going to hit a die configuration that can be achieved but only at great cost (ie very low yields). To date the closest we’ve come to this is Intel’s 10nm process. It was bad enough to the point they started buying up 7nm as fast as they could to be done with it.
Moore’s law gets declared dead and not dead all the time. The Schrodinger’s cat of laws. Actual performance increases is more a factor of competition than tech limitations. Intel was happy shipping everyone barely incrementally improved quad cores for nearly a decade, until AMD figured out how to make good CPU’s again.
Because I’m not so sure of that. @Ryiah I think you mean the cost to feed them. And if it happens cloud services will get an advantage since they can afford that for us. Or just go crypto by itself.
I agree with you that Is also the connection density an important part of the equation.
Common we all know that Moore’s law is not about transistor only! right?
Gordon Moore’s law is also suggesting under the hood Drawing’s unstoppable progress driven by the invisible hand as a natural consequence or artificial extension of the evolution of the natural brain cerebral cortex density.
2020: human 16~80 billion float potentiometers, vs GPU 40 billion booleans gates.
