/tech/ - Technology★

CPU development will stagnate eventually and programmers will have to finally stop being pajeets and optimize their software.

>>900821

Text editors will take 100 gigabytes of ram in the next 20 years and you will be saying the same thing.

>>900820 (OP)

Moore's Law doesn't say anything about CPU clock rates.

>>900820 (OP)

>The Failure of Moore's Law

More's law isn't a law it's an observation about miniaturization of components that is supposedly correct because of physics.

>What will happen then?

They'll find a way to make 3D transistors (which is already the case).

They'll use a material which cost 0.10$ more than the one they usually use.

More botnet.

>increasing the cache, adding more cores, and adding more threads.

Sounds like progress to me. a higher clock speed is not the only technology that can improve a processor

>>900858

Did you even look at the graph you posted? Its measure clock speed and performance per clock cycle. We have many more cores now that can do way more processing compared to older CPUs.

>>900820 (OP)

> Doesn't know what moores law is

Nothing about clock speed you ignorant nigger. Its about the number of transistors on a die and going by the number of cores being squeezed onto CPU's Moores law is still in action.

It will fail when the transistors become so small that electrons will pass through them ie. quantum tunnelling which we aren't too far away from.

>>900860

Isn't that just an illusion of performance? yes, it's possible to have a double-digit amount of cores per die now (and multiple threads per core) but if intel were to release a processor that was a single core without hyper-threading (made using the technology of today) - it'd probably barely outperform the Pentiums of the 2000s.

>>900863

>Is having more performance just an illusion of performance

WTF are you talking about? webservers, databases, all kids of shit are many many times faster. Next you say:

>What if we take the technology of today, cut out everything that makes it modern, then compare it to something old, it would barely be better

well no shit

>>900862

Sorry, I conflated what Moore said with what David House said. But more transistors = more CPU power seemed like a tacit part of the law to me.

>>900862

>It will fail when the transistors become so small that electrons will pass through

<Moores law about transistors will fail when we stop using transistors

I geee smart one here

>>900866

I'm sorry that I failed to explain it well (I've been drinking). I mean that while adding more threads/cores per die is certainly more performant (and the "modern" way to optimize CPUs) the performance itself comes from the cores synergizing - the point being that single core/single thread performance has stalled. I still don't think I explained it well, whatever.

>>900873

It sounds like you are trying to talk about Amdahl's law. Even raytracing eventually caps out performance above a certain level of cores, even though it it 95% parallelisable.

There's a great video on youtube from an actual chip designer, google Sophie Wilson - The Future of Microprocessors

It does explain that 3d chips won't be a panacea because it doesn't solve the underlying problem of heat generation (can make it worse).

>>900873

There's no need to apologize, I think I understand where you're coming from. To the pedants lurking this thread, don't read any further.

Sure, we can technically maintain Moore's law by making bigger dies with moar cores. Sure, performance for some problems will scale with the number of cores. But something about this feels wrong.

Don't any of you remember when computers were actually taking leaps and strides on a monthly basis? You could buy a new computer every year and it would blow your old one out of the water. And look where we are now. I'm using the same desktop I bought 5 years ago, and what's my incentive to upgrade? Better TDP? A couple more FPS for muh gaymen? A shiny new botnet?

Moore's law represented the idea of an ongoing technological revolution--now it's just a biannual opportunity for Intel to jerk themselves off. If Moore's law is still alive, its spirit is long gone.

>>900915

>Sophie Wilson

>it's a tranny

No.

>>900963

>taking leaps and strides on a monthly basis

They still are. Just not for consumer shitware game tech. Look at servers. They get better and better all the fucking time. Four terabytes of ECC ram in a 1U rackmount server is not crazy anymore.

woah as if we have reached the physical limitations of raw speed of flipping transistors on silicon and we have to focus on optimizations instead

>>900867

Well I guess that depends on what you mean by CPU power. If you mean Instructions Per Second than a multicore CPU would be plenty powerful the problem is that all our single threaded languages and code are dependent on IPS being driven on a single core based on clock speed and we hit the thermodynamic limit on that a while ago.

>>900869

I suspect that sounded better in your head.

>>900820 (OP)

Look up the power wall. We can't get much faster without parts melting. That's why we began using multiprocessing

>>901153

http://www.edwardbosworth.com/My5155_Slides/Chapter01/ThePowerWall.htm

>>900821

Oldfag here, they didn't bother to optimize shit when the average PC had 32MB of RAM, they're not gonna bother now

>>901141

Well we can go back to vacuum transistor technology and that'll give it another 10-50 clock speed boost.

>>900863

>it'd probably barely outperform the Pentiums of the 2000s.

Did you forgot to check your nanoseconds ?

>>900820 (OP)

Simple: Go back to the 6502. Computers went wrong when they made them for niggers.