Did you know that Intel has said that transistor sizes will go as small as 6nm in the future but after that any smaller is unviable due to quantum tunnelling? This means that in another 3 years (assuming 6nm is that far away) we are going to see a complete halt to any further CPU and GPU development and most of you guys that spend lots of money on things like 980ti's won't know what do with your money any more.
>>53096646
>we are going to see a complete halt to any further CPU and GPU development
No we won't. Fuck for 30 years people having spouting this shit shut the fuck up already.
It's time to start stacking CPU dies vertically.
>b-but the heat
through-chip water cooling: http://www.pcgamer.com/researchers-have-moved-liquid-cooling-directly-onto-the-chip/
>>53096866
>le AMD logic
>moar coares !1
no. just no.
>>53096901
Stop repeating retarded memes. More cores has no drawbacks unless you count retarded developers. And we'll be getting less and less of those if a high core count will be the norm.
>>53096933
>More cores has no drawbacks
lie
>>53096963
Like ?
And don't bring up shit that only applied in 2005.
>>53096972
The fact that its difficult to multithread programs after a certain point - even impossible a lot of the time.
When microprocessors can get no smaller, computers will just get bigger.
All those people claiming mobile is the future will get a shock when they realize that their phones are already reaching the limit of performance and efficiency.
>>53096993
Really ? Because every worthwhile cpu-heavy application can handle 24 threads just fine.
>>53097020
why dont i7s benefit games over i5s then?
>>53097049
>retarded developers
ames are more about the gpu anyway so who gives a shit about them
>>53097049
because games are gpu bound.
you know what a gpu is?
it's a microprocessor with thousands of cores.
>>53097006
Smartphone devs will get a shock when they suddenly have to write performant code
>>53097084
>>53097073
only the graphics are about the GPU. if you want decent AI in games in the future you're going need some nice CPUs.
>>53097006
I think mobile has a good way to go yet, phones are still gaining dramatic increases in performance every year.
The problem with scaling performance out to multiple chips is that such setups tend to perform worse than a single chip unless the software is explicitly written to take advantage of this capability.
You are also limited in how large a single chip can be because of latency, since electricity isn't infinitely fast, once a chip gets large enough, parts of it can actually become desynchronized from one another.
Using photons in light-based CPUs seems to be the future and scientists already have working samples, so the hope here is that technology matures in time.
Why is /g/ so obsessed with childish video games?
>>53096646
>bonces off
>thiner the barrier
You might be right about CPUs, but graphics rendering is highly parallelizable, so you would still see progress there with the moar cores approach.
>>53097126
I'm sorry there's no space for fun in your life anon
And the op wasn't even purely game based
>>53097105
AI is the easiest shit ever to parallelize if you have coroutines you fucking faggot.
Non-programmers need to stay the fuck out of /g/.
>>53097904
whatever fag, it was a bad example. I'm sure CPUs will become an issue though even for game.
>>53097098
True, at least I know I will.
>>53097998
Except they're not. It took until fucking node.js got mainstream for most developers to learn that you can even do concurrency without manually managing threads, and now that Go is becoming popular across the industry they have learned of a proper concurrency system (CSP).
The vast majority of CPU-heavy tasks can be parallelized. The problem until now has been shit graphic APIs (solved by Vulkan) and shitty ignorant developers.
>>53096646
just trash silicon and move to another shit.
Massive parallelization and core specialization is the future. Nobody is getting past the 4 GHz barrier.
>>53097126
vidya is the only interactive entertainment medium.
Games prepare us for life.
>>53098088
7nm is where intel claims they are not using silicon. They don't mention a replacement. It won't be for 2-3 years, in any case. Probably longer. https://en.wikipedia.org/wiki/7_nanometer
>>53098046
look at the first response in this thread for example. tl;dr It's extremely difficult to create an RTS game that uses more then 2 threads.
Hence why even the best of the best CPUs today slow down starcraft 2 before the GPU does.
https://www.reddit.com/r/starcraft/comments/3q0v3z/sc2_runs_on_1_core_blizzard_mention_multicore/
We'll get more cores.
>but anon writing multithred programs is so hard!!1111
That's why functional programming is slowly getting more popular.
>>53096646
I'd rather have more Ghz than more transistors.
Photonics are the future.
>>53098149
Possibly germanium? That shit was used early on until it turned out to be too expensive for commercialization
>>53098142
>vidya is the only interactive entertainment medium.
That can't be true.
>>53097126
What's wrong with games?
>>53098245
Ehh, it kind of is. Atleast the only currently relevant one.
>>53098235
Indium gallium arsenide seems to be everyone's best guess.
>>53098166
Let me repeat again: NON-PROGRAMMERS NEED TO GET THE FUCK OUT OF /G/.
The faggot you linked is still talking about coordinating parallel tasks, which becomes FUCKING TRIVIAL when you have a model like Go's take on CSP, where channel communication between goroutines allows a subroutine to dispatch the parallel tasks easily and not continue until they get the OK from both, which is much simpler than dealing with a bunch of mutexes (what he means by "synchronization mechanisms").
Plus his example is quite shitty since it's a very obvious synchronous operation. After route calculation if you have a sudden collision mid-path it's your physics engine that's supposed to tell you of the event and dispatch a new route generation. You can't generate a new route for a collission in position x,y,z if you haven't yet collided.
>Hey guys you can't asynchronously print user-input before he runs the print command therefore concurrency is not possible senpai xP!!!
What a fucktard.
>>53097049
But they do. Stock i7 is more stable with less frame drops than an oc i5.
>>53096646
>quantum tunneling
so ghosts are plausible, they just have to be very small?
>>53098376
no, this isn't true because hyperthreading increases temperature without providing any benefit in games. With higher temps, the CPU no longer operates in turbo mode.
>>53097904
That depends on how your AI is laid out .
If you've got multiple agents doing multiple things then it's fucking easy. If you have a central system that controls your agents then you're not that safe. Especially if you're using heuristics . Or you're making a strategy game AI.
why don't we just make the CPUs bigger? They're too small now anyway.
>>53098413
forgot pic
>>53098417
In other words:
If you design your AI like shit, you can't make it run concurrently.
If you design it well, you can.
Sounds like we're back to good developers/shit developers.
>>53098378
Quantum ghosts aren't anything to worry about. As soon as you observe them, they disappear.
Core efficiency > more cores
>>53098432
because that increases the latency of signals getting from one side to the other
bigger chips = slower clock speeds
>>53098357
Hahaha wow it's not easy to trigger you autists. I will never learn to program, and I will always be on this *technology* board posting whatever I want. Maybe you need to go to a programming board, since this is a technology one.
>>53098438
That depends on what you're trying to do. I guess if you're making an everything simulator then you can run into issues but most people wouldnt bother because game development isnt about creating complex systems - it's about entertaining someone and that can be done fairly easily.
In fact most Game AIs are simple enough that they dont require anywhere near as many operations as a modern core can put out.
>>53098448
perhaps, but they're only there if you observe them
>>53098478
The real world runs concurrently anon. An "anything" simulator could be done with a consequence system that queries distributed central nodes (one per grid, for example) that keep track of objects to get those objects and trigger their responses asynchronously.
>>53098413
Bullshit bullshit bullshit. Buy a better cooler.
https://youtube.com/?#/watch?v=EhaB1dqYv_I
>>53096646
obviously once we get to that point we will start seeing more of an investment in different cpu technology, or even an entirely different way of thinking of a "cpu". I can't see it just being the end of the road
>>53098465
>Be a retard
>hurdurr look at how retarded am I
>>53096646
They used to say that about 65nm.
http://www.geek.com/chips/quantum-mechanics-in-flash-memory-547896/
>>53098166
Just make one thread per unit anon. ezpz
>>53098575
Damn, 2002? I was rocking a 1 GHz Athlon Thunderbird and thought it was the hottest shit since sliced bread back then.
>>53098519
whyd you ignore my pic
>>53098438
you need to take into consideration that there simply isnt that many above 140 IQ people in the world and we aren't going to be able to make these elaborately taylored programs that can be paralled to the nth degree. You need to design technology around what people are capable of creating.
>>53098913
Why'd you ignore my video? If your chip is thermally throttling your cooling solution is inadequate. Quit using the stock POS and invest in a quality solution that won't throttle the CPU under heavy load. Duh
>>53098966
they call it turbo mode because it's not expected to always operate at that frequency. It does it only when it can. And in Australian summer, that's never.
>>53098943
I think the problem is that programmers have all been white males historically. If we want better software we have to include more women and minorities.
>>53098999
Waste of good trips senpai
>>53098999
why, that would lower the IQ drastically.
>>53098983
Too poor for AC? Sucks to be you. And turbo gives you what, 200mhz bump? Wow, what an improvement. A small oc will disable turbo anyways.
>>53098221
but we hit that barrier 15 years ago
>>53099146
>10 years ago was 30 years ago
>tfw you probably won't be able to play dwarf fortress with a 24x24 sized map and a thousand dwarves with more than 100 fps in your lifetime.
>>53099267
anders breivik had no problems doing that years ago.
>>53096646
>we are going to see a complete halt to any further CPU and GPU development
By this logic once the tallest possible building has been built we will never have a building that can house more people.
The only difference will be that chip manufacturers will stop caring about transistor size and will care more about the thing that actually matters, architecture. It will just be like what happened at the end of the pentium days - we couldn't push clock speeds any higher so people thought that processors weren't going to get any faster. All that happened is the intel marketing department found another arbitrary statistic to fight with other manufacturers about and architects continued to make processors get faster.
>>53100200
if archiecture matters so much then why has x86 existed for over 30 years?
>>53098357
>NON-PROGRAMMERS NEED TO GET THE FUCK OUT OF /G/.
>>>/prog/
>>>/8gag/prog/
technology is not solely programming.
even in this thread, there is only one tangent conversation that referenced programming.
fuck off neckbeard.
what if instead of making the chip smaller, we just put more of them on a board
like a non-shit Xeon
>>53097020
You just cant make some algorithms parallel.
