Is Moores Law slowing Down, or is that just for the Masses?

[MavicFT1P]

Have your say..

 

[grendeljaeger]

I hope it's slowing down, why would we need more laws 'cause I think we need less laws! Seriously I had to look up Moore's law and at some point things will get so small it will have to slow?

 

[Thomas B]

Had to until another major step in miniaturization happens.... or drone size increases significantly.

 

[Johnmcl7]

Moore's law stopped some time (at least in real world terms) ago, Intel themselves some time ago admitted defeat when they abandoned their tick-tock approach to processors as they could no longer make process improvements quickly enough and unusually they're now having severe manufacturing problems with their consumer processors leading to a shortage in supply. The new Xeon Platinum 9000 series looks to have an impressive transistor density with up to 56 cores but it's two separate dies stuck together rather than a new high density die.

Even before that while processors have maintained moore's law with an increase in cores, the per core performance hasn't been going up much at all and for consumer use that's not great as a lot of software is not good at making use of large core counts.

While the processor market pulled back from extreme clock speeds with the failure of the P4 and instead focused on performance per watt, the graphics card market has pushed on with huge dies and massive power demands. Whereas cpu work often doesn't benefit from large core counts, the parallel work of a graphics card has meant they've been able to keep more aggressive growth except they've largely come to a halt now as well. The GTX 10x0 series was released three years and surprisingly Nvidia have only just released their successors with the 20x0 series, even more surprising despite the gap in time is they barely outpace the much older 10x0 cards instead offering support for raytracing which while impressive, has almost no developer support.

Mobile platforms have still been advancing quickly in performance but that's likely to slow down now as it's going to be more difficult to move down to smaller process nodes.

I don't think it's all that bad though as I remember the days of P3's, Athlons, P4's, A64's, C2D's etc. and the relentless pace of performance, a high end PC would typically only last around 18 months being able to run at the top level before new processors and graphics cards would up the ante again. I used to spend a lot of time trying to get games running at their optimum level and speccing up new machines to catch up again. Now my desktop is a staggering seven years old, although most of its been replaced its cpu/motherboard are still the same as its hex core i7 is still extremely capable easily managing 1440p gaming in current titles and chewing through video files. Its 12 threads are often badly underutlised even by software that claims to be multicore optimised so a new system with a higher core count doesn't really appeal.

Normally I wouldn't mind too much if an older system failed as I could treat myself to a newer one but in this case I very much hope it will keep on going as replacing its core components is going to be pricey for not much gain.

 

[frank candor]

Moore's law stopped some time (at least in real world terms) ago, Intel themselves some time ago admitted defeat when they abandoned their tick-tock approach to processors as they could no longer make process improvements quickly enough and unusually they're now having severe manufacturing problems with their consumer processors leading to a shortage in supply. The new Xeon Platinum 9000 series looks to have an impressive transistor density with up to 56 cores but it's two separate dies stuck together rather than a new high density die.

Even before that while processors have maintained moore's law with an increase in cores, the per core performance hasn't been going up much at all and for consumer use that's not great as a lot of software is not good at making use of large core counts.

While the processor market pulled back from extreme clock speeds with the failure of the P4 and instead focused on performance per watt, the graphics card market has pushed on with huge dies and massive power demands. Whereas cpu work often doesn't benefit from large core counts, the parallel work of a graphics card has meant they've been able to keep more aggressive growth except they've largely come to a halt now as well. The GTX 10x0 series was released three years and surprisingly Nvidia have only just released their successors with the 20x0 series, even more surprising despite the gap in time is they barely outpace the much older 10x0 cards instead offering support for raytracing which while impressive, has almost no developer support.

Mobile platforms have still been advancing quickly in performance but that's likely to slow down now as it's going to be more difficult to move down to smaller process nodes.

I don't think it's all that bad though as I remember the days of P3's, Athlons, P4's, A64's, C2D's etc. and the relentless pace of performance, a high end PC would typically only last around 18 months being able to run at the top level before new processors and graphics cards would up the ante again. I used to spend a lot of time trying to get games running at their optimum level and speccing up new machines to catch up again. Now my desktop is a staggering seven years old, although most of its been replaced its cpu/motherboard are still the same as its hex core i7 is still extremely capable easily managing 1440p gaming in current titles and chewing through video files. Its 12 threads are often badly underutlised even by software that claims to be multicore optimised so a new system with a higher core count doesn't really appeal.

Normally I wouldn't mind too much if an older system failed as I could treat myself to a newer one but in this case I very much hope it will keep on going as replacing its core components is going to be pricey for not much gain.

Tesla Valves and Quantum computing? Asks a luddite -sci-fi lover?

 

[Johnmcl7]

Tesla Valves and Quantum computing? Asks a luddite -sci-fi lover?

Technically neither are relevant to Moore's Law which is specifically regarding the transistor complexity which he predicted would double every two years. Some have taken that to mean performance doubles every year and therefore it's a measure of performance but that's not the case particularly in the last few years where we've seen core counts and caches soar but not so much performance for every day consumers.

I assume when you mention the Tesla Valve you're meaning using micro fluidics instead of electronics for computing? It's not something I'm remotely familiar with although after a quick read around, it doesn't seem that well regarded as a mechanical technology never mind anything beyond that.

Quantum computing is something that's been on the go for a long time as well even way back when I was in University we were shown the concept and the effect quantum algorithms would have on the computing world which in effect is destroy it as we know it. Although current processors are incredibly complex with billions of transistors running at high speed with multiple cores and huge caches they're still fundamentally extremely simple with each core taking one simple instruction at a time, computing the solution and sending the output.

Comparatively the human brain is very different as it works by sending huge amounts of updates simultaneously allowing it to process vast amounts of information at any one time. It's not something you're aware of but it's quite staggering how much work a simple action like standing up can take when you consider the number of motor commands that must be sent, all the different sensors sending feedback as well as still sampling and recording audio and visual data while maintaining all autonomic functions such as breathing.

Quantum computers work in a very different way to conventional computers as they don't process single commands at a time but can effectively process many commands simultaneously. It's a technology that seems to have been around the corner for many years now so I don't know if it will ever achievement its potential, same goes with many other AI technologies as well but it's reasonable to think in time we will eventually have a different computing paradigm.