Transistors Will Stop Shrinking in 2021

>>55723522
Eat shit, singularityfags.

>>55723522
>10nm is the smallest we're going to get until the semiconductor material changes completely
Then the semiconductor material will change completely. The industry is far too large with far too much money to throw at research for everyone to just throw up their hands and say "welp we hit the limit of silicon better close down shop"

>>55723522
find or fabricate a replacement semiconductor

>>55723522
>>55723551
>>55723562
Graphene processors when?

AMD will save us

>>55723551
If 10 nm is the best we can do they can just keep pumping out millions of units of chips each year per demand desu, not being able to improve doesn't mean stopping production desu

>>55723551
A die shrink costs billions by itself, then it takes years of research and costs billions to do a change in the materials used in transistors since the lithography changes so much
>>55723574
graphene likes to be graphite too much for anything useful.

>>55723624
It does mean the end of doubling performance year over year, and the end of solid power consumption drops.

>>55723609
this just wait for year 2653

>>55723562
They've been researching this forever.

There has been working lab samples of graphene and diamond semiconductors since more than 20 years ago, but the progress has been very slow even as the industry has rapidly approached the silicon node shrink wall.

File: eienlol.jpg (87KB, 346x347px) Image search: [Google]

87KB, 346x347px

>>55723750

>>55723522
I, FOR ONE, WELCOME OUR NEW QUANTUM COMPUTER OVERLORDS

File: 1446639350036.png (92KB, 295x229px) Image search: [Google]

92KB, 295x229px

>>55723522
It's all over

File: 1468900329618.png (784KB, 899x899px) Image search: [Google]

784KB, 899x899px

Absolutely coincidentally, 2021 is the year that the last white man in the industry has been replaced with a diverse team of women and minorities.

File: aRPw3xM_700b.jpg (76KB, 700x875px) Image search: [Google]

76KB, 700x875px

moar coars

And they say programmers job is future proof lmao

File: tumblr_n073c7pqFb1sgwvw2o1_501.png (10KB, 248x203px) Image search: [Google]

10KB, 248x203px

>emulate 5nm transistors on 10nm transistors
>everything goes twice as fast

File: q.jpg (27KB, 700x394px) Image search: [Google]

27KB, 700x394px

Hi there

planck scale transistors when

File: Tech-brain.jpg (93KB, 440x228px) Image search: [Google]

93KB, 440x228px

so its official then singularity is a pop fiction lie then

unless quantum computing doesnt turn out to be a meme computers wont get anymore powerful after 5 years only bigger

>>55723750
>implying AMD isn't going bankrupt by the end of this year

File: 131435356316.jpg (289KB, 698x400px) Image search: [Google]

289KB, 698x400px

>>55724592
not programmers because "monkeys" but the rest it will be more needed because of everyone cant live without tecnology, and the robots and the 3D stuff will take over the world. the rest of the jobs (not all) will not survive more than 30 years.

I am not joking, read my words carefully because its true

File: different-transistor-topologies.jpg (479KB, 1195x894px) Image search: [Google]

479KB, 1195x894px

>>55723522
>10nm

LOL

new transistor designs could in theory bring density improvements without major changes to the process itself, TSMC 16FF and samsung 14LPP are essentially planar 20nm except with finfet transistor design to give planar 14nm equivalent density.

>>55724680
>he fell for the d wave meme

isnt cannonlake 10nm set to release next year (2017)

>>55725521
10nm Cannon Lake will be mobile only. Their next desktop platform will still be 14nm.

It doesn't matter, we'll get past the limitations by then.

Already moving onto the next medium behind the scenes.

http://www.pcmag.com/news/345934/microsoft-stores-200mb-of-data-on-strands-of-dna

This trend has been obvious since 2000. I'm old enough to remember when the jigaherz were going up a ton every couple years. Now we've been stuck at 3.5ghz for ages and it's "more efficient architecture".

>>55724680
>I know nothing about quantum computers

>>55725589

we've been stuck at 5ghz for ages, not 3.5ghz. at least on intel's side, AMD only got >4ghz clocks consistently with bulldozer and probably won't replicate that with zen.

I honestly doubt that 10nm will be it for silicon, the physical limit is 7/6nm, and there is to much money in getting to that limit that they wont just sit at 10 while there is still a smaller process.

we however WILL switch over to another material, graphene is one of the more promising ones, as in labs hitting 50-100ghz is not hard on air, the issues are the stability of the material, and the fact we would be going backwards to a 30~ nm process, considering i'm on a 45nm chip and its running just fine, being able to run the chip at even 20ghz over air would EASILY make up for the larger process.

>>55725623
Yeah, I meant on average. Where is my stock 20ghz chip dammit

>>55725623
don't know that yet, but yea, honestly doubt them hitting 5ghz will be a non silicon lottery thing. that said, 4.5 is all you really need to alleviate any reasonable bottlenecks, and to that end, 3.5-4 can also do just as well, its once you hit 4.5 its diminishing returns all round.

>>55724623
Kek'd

>>55723522
For performance, wouldn't a higher frequency (on good architecture) be more important? For parallel tasks you can just throw more threads at it. I obviously understand the power/price push currently in the industry, but maybe they'll chase one when the other is at a dead end

>>55725851
The problem is with the end of silicon scaling, we cant go faster. The substrate simply cannot handle it, and these wide and slow architectures limit the clock speed further.

Going narrow and fast again (Neburst, Bulldozer) wont work because heat and power consumption would give us another painful bitchslap.

Different transistor structures, 3D chip construction, and maybe switching to GaAs or something similar are the only real ways forward, and all of them are going to be expensive as hell, which in turn will drive prices up for us.

Chlorine processors when?

Since humans will eventually master time travel, why haven't the humans from the future come back in time and given us future technology?

>>55725568
That's storage you dumbass.

We're talking about processors.

>>55725986
There is a theory for that

>>55725986
You can't actually travel back in time. You can only see a mirage from the past.

Like we do when we look at the starts right now.

>>55723522
Why should the industry try introducing newer fabrication methods? The manufacturers benefitting from this all have zero competition. Intel has monopoly on x86, Nvidia has monopoly on graphics accelerators. AMD barely releases anything that could drive the market forward.

This is also why earlier forecasts show more favourable node shrink predictions: a few years back, AMD had more competing products. Few years ago they still had high-end CPUs (even if they were high end only in their own way). Today they have... one new videocard, and a bunch of low-end APUs. recycled from 3 year old tech.

If there was any actual competition, you can bet your ass that Intel would sink billions into further shrinking nodes. They abandoned tick-tock because of this too: there is no gain in R&D when they are a half decade ahead of the competition in their x86 designs.

>>55723535
>implying the singularity won't start on a Pentium 4 running gentoo

>antisingularityfags
You niggas don't even know what the singularity is. Shit, I fell for the bait didn't I?

>>55725986
If FTL travel is possible on humanity's lifetime, will we have powerful enough scopes to see the history of humanity in its entirety?

>>55726156
>techno cultists

>>55726124
>what is that a-lot-more-lucrative-than-PC embedded SoC market?

>>55723745

They will still have plenty of leg room to improve performance and lower power consumption. Just because 10nm is the smallest they can make the chip, doesn't mean they have the optimal architecture figured out yet.

>>55726267
Pretty much this, the only reason lithography has advanced aside from Intel developments is the SoC market
Aside from the SoC market, lithography is used to produce pretty much any electronic, be it SSD's or whatever ASIC out there

Fuck it just make a bunch of architecture improvements and high speed die interconnects until then.

>>55725986
1. We can only travel a certain distance from the inception of time travel

2. We are the first instance in space and time and thus, there is no "future" us to meet.

3. Time travel will never be created

>>55723551
This.

Basically when the new material is stable enough everyone will use it.

Much like how Carbon Fiber and Bonded Alum took 30+ years for the auto industry to actually start using it.

>>55725289
Hei lysefag

>>55726615

if theoretically time travel is possible im sure there will most certainly be some ban on going in the past and changing shit to create outcomes beneficial to yourself. in other words with the exception of God time travel is impossible.

File: 1446439277109.jpg (10KB, 234x215px) Image search: [Google]

10KB, 234x215px

>>55723626
>graphene likes to be graphite too much for anything useful.

>mfw my AMD Zen 2 CPU overheated and turned into a pencil

>>55723551
5nm and you get quantum tunneling.
3nm is barely possible with different geometry for the transistor.
We're running into a brick wall with transistor sizes, the industry will stop the shrinking (unless quantum computers take off).

>>55726758
>5nm and you get quantum tunneling.

Insulated channel.

>3nm is barely possible with different geometry for the transistor.
GAA.

>We're running into a brick wall with transistor sizes, the industry will stop the shrinking (unless quantum computers take off).

Optoelectonic devices can scale below conventional transistors. "Quantum computers" are still using chips fabbed in silicon, the more qbits you want the larger the die will be. Guess what that means? Qbit chips will scale exactly like every single other CMOS chip. It isn't a solution to area scaling, you idiot.

Stop regurgitating things you see other tech illiterate retards post here.

File: laughing Dalian.webm (63KB, 240x135px) Image search: [Google]

63KB, 240x135px

>>55726680

Doesn't Intel have plans for a 5nm CPU?

>>55726854
>GAA
how much a benefit will this realistically offer end users?

sure leakage and switching speeds would improve, but it's not like there are massive density gains to be had, nor the speed benefits to be had from increased densities.

>>55723522

I don't get it.

Why do people associate transistor size with performance ? The actual size of a CPU's die is small, and in a PC as well as many other applications you aren't constrained with space anyway.

>>55727157
The orientation of the structures is entirely different. Instead of having enormous source and drain wells with a small switch in the middle, GAAs can be oriented vertically, or horizontally, and require a much smaller pitch metal contact. They facilitate enormous back end scaling, and drive current is in the mv range. They're the crowning gem of conventional CMOS designs.

>>55726854
>quantum tunneling.
>Insulated channel.
This is the dumbest thing I've ever read

>>55726854
Insulation won't have a noticeable effect on tunneling, go brush up on quantum theory friend. Tunneling can occur through regions of infinite potential, aka perfect insulation

>>55727228
Shorter distances means faster processes. Making the die bigger can only help so much.

>>55727228
Smaller transistors mean faster switching and lower voltages (which means less heat). Die shrinks increase yields, and are also necessary to keep the quicker clock synchronized across the entire die.

>>55725603
>Not having a quantum core in your cpu
>2029

>>55727268
>I'm a retard
Cute

>>55727277
Literally thats what SiGe channels are used for. The entire design behind a "quantum well FET" is to have a device completely insulated so that errant charge cannot tunnel out of the channel. A normal FinFET structure with a SiGe channel is matter of factly a QWFET device.
Tech illiterate redditor retards who spout buzzwords are in no position to discuss the nuances of advanced IC design.

>>55723522
That's good. This will force developers to make efficient OS and applications. You'll no longer need a new CPU every 6 years because some new OS update needs it.

>>55727236
but isn't the entire premise of this thread that lithography scaling is arguably becoming untenable?

how much do things like smaller pitch metal contacts matter for things like wire delay if the metal layers aren't getting denser?

>>55726677
>im sure there will most certainly be some ban
How the fuck do you enforce something like that? Fucking time police?

>>55727293
Logic core
GPU core
Quatum core
Neuron core


What did I miss?

>>55727432
FPU.

>>55727329
Time Enforcement Commission

>>55726677
Allahu Snackbar

>>55727329
time travel, if invented is almost sure going to require nation-scale or even global-scale resources and effort to perform.
Which means enforcing would be as simple as just not doing it for anyone.

>>55723535
This isn't really bad news for the tech singularity. Yeah transistors are now shrinking slower but performance of processors is getting better and more energy efficient.

The whole point of the tech singularity is that a computer becomes self aware and starts improving itself to become smarter than a human.

All this requires is understanding of what algorithms make humans conscious and smart. To do that we just have to simulate a human brain and find out those things in simulations.

Modern computing requirements estimates for a human brain simulation are about 1 exaFLOPS. We have 0.1 exaFLOPS of computing power with that recent chinkshit supercomputer.

Anyway I'd say we're perfectly on route to the tech singularity by 2040-2050.

>>55727455
Okay then how do we unify and program all this?

Imma take this opportunity to ask a question that doesnt deserve its own thread, i saw somewhere that transistors themselves are not 14/16/xx nm but instead that thats the size of the die containing the transistors, could anyone clarify or post some source where i can read about this?(i dont know how to word it for google) thanks

>>55727526
How long before rich people send themselves into the future with fast spacecraft?

>>55723522
>ctrl-f "gallium"
>no results
InGaAs is an alternative. Intel says 10nm is the end of the road for silicon alone, but gallium arsenide buys some time.

>>55725964
Ch-Chlorine processors, anon?

>>55727554
now
t. Elon

>>55727537
no. it measures the half pitch of two memory cells

>>55727533
kek give me a fucking break, singularityfags

>>55727611
as soon as chlorine becomes an III-V semiconductor

File: 1846295639503.png (134KB, 465x235px) Image search: [Google]

134KB, 465x235px

>>55727611

>>55727533
I definitely think we will have AGI by 2050 for sure. I believe the whole AlphaGo thing from Google Deepmind is already promising considering its accomplishments.

>>55727277
You might want to brush up on quantum theory. Tunneling probability through a potential barrier decreases with the height of the potential

>>55724623

Holy shit it's been a while since I've seen a troll science post.

>>55726012
I've been privy to some work building transistors out of RNA, like most biological structures they don't have good heat resistance though.

File: ibm-whiffs.jpg (33KB, 915x686px) Image search: [Google]

33KB, 915x686px

SAVE US, IBM!

>>55725986
They will. But we haven't hit the point they travel back to yet.

What does this say about semiconductor testers?
I'm hoping it's business as usual for us.

>>55725986
What if they already did?

whatever happened to metamaterials?
also, the talk about GHz made me remember about terahertz EM waves... wonder if we will ever control and use fast photonics circuits

>>55725986
Might happen, but I dont think we would be able to go back in time.

>>55728068
Jews are still alive. They didn't.

>>55727533
>All this requires is understanding of what algorithms make humans conscious and smart. To do that we just have to simulate a human brain and find out those things in simulations.

You say that like it's no big deal. How would you even find, and relate to human sentience, algorithms, let alone understand them? How would you simulate a human brain? How would you turn that into human readable "algorithms"? How, if you found these algorithms and they even exist in quantifiable form, artificially recreate the plasticity of the brain i.e. create self-improving software? Plus a whole bunch of other shit. No one alive today will be around to see any of that shit. It's not like we suddenly make a computer powerful enough that it's like "oh hey it's fucking sentient, time for singularity". There's a lot more that has to go into it beyond processing power.

>>55726854
>GAA
That's exactly what I'm referring to.
>Optoelectonic devices can scale below conventional transistors. "Quantum computers" are still using chips fabbed in silicon, the more qbits you want the larger the die will be. Guess what that means? Qbit chips will scale exactly like every single other CMOS chip. It isn't a solution to area scaling, you idiot.
Are you fucking autistic? I'm not saying quantum computers will solve the area scaling problem. Holy fuck how dense can you be? I'm saying quantum computing will improve our processing capabilities by using a fundamentally different computing mechanism.
Fucking retard aspie thinks I'm referring to area scaling lmao.

>>55728147
The field is called artificial intelligence and we've already made significant progress.

>graphene is one of the more promising ones, as in labs hitting 50-100ghz
Hearing the idea of a 50ghz cpu makes me want it NOW

Duh? Fucking atoms aren't much smaller than a nm

>>55728830
We have, but the second AI winter is coming, when people realise that neural nets are just neural nets.

>>55723551
>we can't go smaller than quarks, unless physics are redefined
>Then physics will be redefined because money!!
(you)

>>55728942
This. We still don't know if there's quantum-level processes involved in intelligence.

File: aliums.jpg (111KB, 960x720px) Image search: [Google]

111KB, 960x720px

>>55729016
>Implying the ayyliums won't feed us know-how

File: lesbians[1].jpg (30KB, 460x413px) Image search: [Google]

30KB, 460x413px

>>55729052

what about biological components like that so-called DNA computer tech?

>>55723626
>A die shrink costs billions by itself, then it takes years of research and costs billions to do a change in the materials used in transistors since the lithography changes so much.

You don't think fabs are willing to pay that? And you think they aren't already underway with the research?

>>55729016
>>55726758
Well we don't have to keep using transistors. Optical computing is a good candidate

>>55723574
Graphene is a meme.

>>55724755
https://www.pcper.com/news/General-Tech/Red-Team-new-black

>ASCI White
>The world's fastest supercomputer from November 2000 to Nov 2001
>12.3 TFLOPS
>106 tons
>6 MW electricity, half of it just to cool the beast
>$150 M in current dollars

You can now buy a single GPU that beats it modestly OC'd, @$1200 and fits inside a shoebox.

And if you SLI 3 of those, you beat Earth Simulator which was the top supercomputer until 2004, and still made it top 10 in June 2006.

My question is: How long until we get Avatar -tier graphics @60fps that doesn't cost you a kdney? And how badly affected are GPUs by stalling Moore's law?

File: image.jpg (47KB, 534x491px) Image search: [Google]

47KB, 534x491px

Nothing's stopping anyone from developing new lithography techniques....

If you want to push ahead fabrication you gotta consider the bottleneck in production.

Optics of narrowing a beam to such a small resolution.... And then, thickness, and Layer depth, and then cooling.

So... If one could print circuits in another form other than lithography, and come up with a solid way to cool this. On novel materials/composites...

And we combine that with the latest in "quit" hardware neural networks or whatever is the best circuit for the desired task, cram as much into a 3D space as possible and hooray.


If that's not enough, just look into growing Giant genetically engineered brains in vats and connecting them to computers wetware/hardware.

I dunno anything

>>55727533
Not happening in your lifetime, true singularity is at least 100 years away, we don't even understand our own brain and thought process.

>>55729636

As long as there's plug and a guy whose only job is to look out for killer robots, we ought to be fine.

>>55725851
power dissipated is square of current, and current scales linearly with clock speed. Also, you need more voltage for faster clock, which in turn also increases power... and increases current as well (due to gate charge increasing with voltage), so it becomes a cubed relationship...

>>55729613
You can always fit more cores on a gpu because its parallel, no?

>Figure out how to make 10nm manufacturing reliable
>Wait 5 years
>Someone figures out how to make memristors
>Feels good

>>55730253
Yes, but increased die size is bad. More heat, harder to cool, much more expensive to produce, etc.

>>55726680
Underrated comment

>>55729613
>>55730253
GPUs can always be scaled up in die size, but

Once transistors stop shrinking it's all over.
You can't reduce power draw in a significant way without sacrificing double precision heavily.
You can't pack vastly more transistors into the same area and thus get your doubling of performance for free.

GPUs will become more expensive because more RnD is going to need to be invested for the same or less improvements.

Production costs will go down if we're stuck on the same node for a long-ass time but that's a pittance compared to actual core design.

>>55730314
We obviously need cpu production revolution, something entirely new

File: not-a-jew.jpg (6KB, 319x158px) Image search: [Google]

6KB, 319x158px

>>55727533
>human brain simulation are about 1 exaFLOPS

I like how singularityfags seem to think that the brain does floating point math at some point during cognition.

File: i7-6700k-gta-v-2[1].png (21KB, 652x474px) Image search: [Google]

21KB, 652x474px

So with all this talk about the degradation of microprocessor development, how is it that nobody has mentioned Intel's 4790k outperforming the 2-gen-newer skylake 6700k ? By outperforming, I mean with the daily/gaymen use of single thread power and FLOPS, primarily. And look at this quickly-picked benchmark - GTAV is as fast on either one. This is AMD Bulldozer-tier BS and this is coming from a team red fanboy.

>>55730392
>he can't divide 9 by 7
>laughingmathematicians.jpg

>>55730410
How many times can you do that per second, friend?

>>55730410
He literally specified 'during cognition'
Of course brains are capable of floating point else arithmetic would not have gotten very far

File: 1455346611601.jpg (36KB, 500x395px) Image search: [Google]

36KB, 500x395px

>>55726758
Quantum Computing will not be good for rendering in games.
It will mainly be used as a separate chip on video cards to handle physics and A.I.

>>55730402
tfw I have a 3570.. for once I did a good buy

>>55725630

50-100ghz is impossible unless you shrink the chips, you need to be able to communicate from one point on the chip to other point on the chip in a single clock. With 50-100ghz you would need to break the speed of light.

>>55727611
Well, we have the option to make storage devices with them that are vastly superior to helium:
https://youtu.be/ZcU-sZJkh_U

If you can't make the transistors smaller, you should just make the processor bigger. That way. You can fit more transistors on it.

???
Profit

>>55723551
never going to happen.
going to take todays chips and "stack them" kind of like they're doing with HBM

we need to find a material that can handle sub-10nm transistors while also having the properties of silicon that cause it to be fairly resistant to electromagnetic/radiation interference. Otherwise shit's going to be way too fragile at that scale. I've been wrong before though, i remember worrying about that shit when we were approaching 0.18 µm on silicon but it turned out to be a non-issue anyway...

>>55731005
Heat.

>>55730890
at 5ghz light only has time to travel 59.95849 millimeters or about 2 inches between each tick

pretty interesting stuff.

>>55729016
>Then physics will be redefined because money!!

Yes

>>55723522
That article took me 5min 38 seconds to read. It didn't even mention graphene. I guess it really is just a meme. Looks like 3D chips are the way forward for now, unless one of the big four chip manufacturers make a different breakthrough and choose another way forward.

>>55731257
So get a bigger heatsink with a bigger fan.

With the end of Moore's Law and larger dies, we might see 140mm or 160mm fans become the standard for CPU coolers instead of the pissy ass 120mm nonsense we have now.

>>55731318
>>55730890
so you enclose the die inside a static warp shell, or bubble if you prefer, and poof whered my laws of physics go

>>55731442
It's not really the "end" of Moore's law. The amount of Transistors will continue to double, it'll just be on a 3D lattice.

> The memory industry has already turned to 3D architectures to ease miniaturization pressure and boost the capacity of NAND Flash. Monolithic 3D integration, which would build layers of devices one on top of another, connecting them with a dense forest of wires, has also been an increasingly popular subject of discussion.

>>55731442
200mm is the way to go. Always has been.

I think it'd be totally fine to return to a larger fab process to see gains in clock speeds and better heat dissipation. Miniaturization has hit a technological wall and without some form of construction via nanomachines, the apparent alternative does seem to favor switching to a different material. Space program could help with this cuz asteroids n sheit.

>>55725986
what if the multiverse theory forces you to live in a version where you dont have time travel as at some point 1000 years into the future you invent it you go back but create another universe where you went back with timetravel.
you cant give it to yourself but when you invent it you can start a new pararell reality where you jumped tech 1000y forward.

>>55730443
The human brain is naturally capable of trig and physics calculations (behind the scenes) - we've been firing arrows and throwing spears at things for millennia. It's an almost automatic process of analyzing depth of field and motion of target along with learned trajectories of a given mass at x size and shape.
Granted, this is all deep down in the motor function systems, but my point still stands.
Higher level math, somewhat beyond algebra, is a simple function of correlating the expected outcome with our natural sense of complex motion. It is instinctual.

>>55731570
I've got it. We need a cpu that mimics the biology of a brain, rivals the clock cycles of synpases,and all it takes to keep it cool are hair follicles

>>55731537
Time travel in a multiverse theory is de facto travel through infinite similar universes. You wont be affecting your concrete past or future, but affecting the past or future (or both) of a completely different universe.

This thread is now /sci/ mumbo jumbo tier

>>55731005

see
>>55731318
>>55730890

unless you go 3D (qube) you cant make CPUs bigger.

>>55731632
also my math was based on light in a vacuum so it's actually less

>>55725986
Are you stupid or something? The future has not happened.

Intel has the biggest R'n'D funding in the entire world, I'm sure they'll figure out something.

File: WpncJAB.gif (3MB, 250x238px) Image search: [Google]

3MB, 250x238px

>>55731748
>Summer, 2020
>New chipset, socket 9001, an improved version of socket 9000 released in Q1 2020
>New line of processor, codename Columbia
>7 cores, 6 virtual cores on top of physical cause logical core async will work somehow in this imaginary future (screencap this leak!!!)
>locked multi, locked voltages, cock blocked locked clocks
>13nm, TDP 355w, 6.66Ghz
>TJ Max 80 degrees c, Idle speed-step temps @ 65
>MSRP $13,000

>AMD
>Summer, 2020
>Nothing

This is the future you chose.gif

>>55727533
>understanding of what algorithms make humans conscious and smart
wtf are you talking about. humans dont run on algorithms. our brains were brute forced to work through evolution.

File: amdintelcores.jpg (81KB, 800x1000px) Image search: [Google]

81KB, 800x1000px

meanwhile nothing can make use of more cores

0nm transistors when

>>55732090
Based fucking devs, man.

2006-2008
Dual/Quad cores will provide massive leaps in performance for everyone!!!

1976-2005, 2009-2016
Single-threaded count is #1 most important feature because fuck parallel offloading. Shit is too hard for our code monkeys and who the fuck uses cinebench every day (the 10 people seeding your game of thrones episodes).

>>55729613
Nope. GPU teraflops are FP32. All serious computing uses double precision. You still need several high end Xeons to get 12.3 TFLOPS of FP64.

>>55729635
Hehe it says retarded

>>55732090
>I don't understand the purpose of multiple cores or hyperthreading

blame shitty poo in loo programmers for not taking advantage of it

do you really think a one core cpu that can only run one thread at a time would be just as good

>>55732399
well a single cpu core can run multiple threads, how well depends on the workload. i could easily play an older video game and have a youtube video open on a single core of my i5

>>55731318
>59.95849 millimeters

Just say about 60 millimeters you nerd

>>55732389
>graphics rendering
>requiring more than fp32 precision

it's like you didn't even read his post

>>55732399
> being this underage
why do you think multiple cores are somehow better for multitasking?
they only exist because they have failed to scale up single cores
there is literally no advantage for programmers and end users

Could moving away from x86 help at all? I mean, even if we can't fit more transistors on the die, it can't be that something we started using decades ago is the best way to be utilizing the transistors we do have.

>>55724680
>>55724074
>b-but quantum computers
They require a lot of cooling equipment, to the extent where most people likely won't be able to run one in their home. Have fun using the botnet/cloud for everything.

reminder that samsung is already selling a > 15 TB SSD

>>55731748
you mean samsung

>>55723562
Or just stop trying to shrink everything so that heat / bleeding issues don't occur.... The average desktop is relatively large... Increasing the physical CPU processor size on a motherboard won't harm anybody...

File: tumblr_na8jp5Ku7b1sped3xo1_r1_500.jpg (56KB, 500x464px) Image search: [Google]

56KB, 500x464px

>3d architectures
>scalability
>hundreds of small cores instead if four big ones.

That will force them to progress intensively rather than extensively.
Just think about it, we have GPUs that outperform CPUs by like many times. a CPU is about 100-150MFlops gpus are like 5-10 GFLOPS.
It's about fucking time to abandon x86 and x64.
I wouldn't be surprised if they could develop an architecture capable of emulating an x86 system faster than it is at this point.

>>55723522
WTF is going on on this graphic? In 2015 they've predicted that 14nm would come out in 2019??? Intel started selling 14nm CPUs in 2014!

>>55733653
Graphics processing is highly parallelizable. Stuff the CPU does is usually not. You have laterally thousands of cores to achieve the 5-10GFLOPS performance on a GPU. Each core has a shitty single thread performance.

>>55723522
Just put more cpu in the motherboards and tell devs to stop being lazy.

>>55723522
B-but I thought Intel was already working and moving to alternate materials for their 7nm processors. Please don't let it end like this, I have this awful feeling that processors hitting a wall and people not having a reason to upgrade their computers will push everyone to the cloud.

The singularity for computers would take about as long as it took the human brain to develop through survival and evolution. At least if you're going to have programming or chips approaching the complexity of our neurons.

So on the scale of at least a few millennia of hardcore computing. Of course, the human brain was shaped by nature and computers are just flat out designed by humans and will never evolve.

tl;dr Singularity is never happening. Transistors will hit a limit. Nobody uses processing power these days for anything anyway. We hit the sweet spot back in the early 00s (being able to use a browser without it slowing your computer to a crawl).

>>55734130

Also 'muh AI' is vulnerable to:
- A tornado destroying the building
- The power dropping out
- Someone shooting the server
- Etc.

How would AI defend itself? With one of the goofy boston dynamics robots? LMAO that shit is defeated by 2 stairs.

Technology is so fucking fragile it isn't funny and the more complex it is the more vulnerable it becomes. Go look at the errata list of a modern Intel processor, they aren't even safe for mathematical number crunching because of all the bugs, you think it's going to simulate a brain in your lifetime? It can't even add numbers.

File: my body is ready.png (259KB, 463x462px) Image search: [Google]

259KB, 463x462px

A good discussion thread on my /g/?
Holy shit.

The processor speed is not increasing exponentially anymore, it haven't been like that for a while.
The desktop market is dying, so it only hurts servers and phones.

For me, it gets more exciting.
We have to demand creative thinking from the people who design hardware as well as the people who write software.

Having limits is a good thing.
We have relied on other people to solve our problems for too long.

People should stop being lazy cunts and do their jobs.

File: 1443753297973.png (90KB, 500x501px) Image search: [Google]

90KB, 500x501px

>>55734130
>>55734160

>>55723522

Thats bullshit.

Germanium aint even stable commercially at 7nm so transitioning to that would bankrupt the industry.

They'll find a way to get down to 5nm while finding a way to make germanium affordable.

>>55734210

Go look at the known issues .PDF for a Pentium 4, it's over 100 pages long. A modern Intel i7 is 150-200 pages long.

>>55723551

The industry only does 150 billion in sales for the processor and graphics markets. They're not going to spend 400 billion over even 10 years to retool everything since 1995 so they can move to a new material.

You better start buying silicon stocks because at best the industry is going to eat itself.

>>55729322

No its not. Graphene and Germanium might be profitable some day...

years from now in a Galaxy far far away.

>>55734256
Then why do they use them for most supercomputers?

>hardware stops increasing in speed
>shitty bloatware is frowned upon
>suddenly, a minimum of competence is expected from software developers

What a glorious future.

http://www.gao.gov/assets/680/677436.pdf

>U.S. Government Accountability Office (GAO), which found that the Pentagon was still using 1970s-era computing systems that require "eight-inch floppy disks."

>U.S. Treasury for example, still depends on assembly language code "initially used in the 1950s."

>GAO report says that U.S. government departments spend upwards of $60 billion a year on operating and maintaining out-of-date technologies.

>That's three times the investment on modern IT systems.

And know why? Because old tech is simple and works effectively. Supercomputers are literally a worthless meme to calculate pi. Imagine running any Modern OS in equipment dealing with nukes, what a disaster.

>>55734310

The only thing we've gained from faster processors and more RAM is shitty programmers who still manage to code a website that locks up your computer for 5-10 seconds when it loads because of all the fucking scripting. Literally nothing else.

>>55734336

>COBOL and IBM mainframes
>COBOL and IBM mainframes everywhere

ayy

>>55727533
>To do that we just have to simulate a human brain and find out those things in simulations.

Do you have ANY idea how enormous a task you've just handwaved away in that sentence? The largest neuron cluster we've ever simulated was 10,000, the human brain is 86,000,000,000. And then we have to try make sense of what it does. Oh and we need to extract the configuration from a real brain just so we can build this thing. You don't "just" do any of this.

>>55734325
>>hardware stops increasing in speed
>>shitty bloatware is frowned upon
>>suddenly, a minimum of competence is expected from software developers
>What a glorious future.
underrated post

File: 555050_481375198599453_1131256104_n.jpg (42KB, 499x459px) Image search: [Google]

42KB, 499x459px

Didn't IBM recently come up with a way to reliably use a 7nm process?

>>55730402
That's because the 6700k is what the 4770k was. Kabylake is essentially 6700k but better.

>>55734352
Well, we did also gain the ability to easily run VMs at home, along with the advantages that came from SSDs. I remember having 2 desktops with a KVM switch to change between them years ago because fuck the amount of time it took to restart and change OSes back then. I do agree that most of the potential has been wasted though and one of the things I enjoy is getting older weaker hardware running well enough to do 99% of what I do with my semi modern desktop from 2009. It's amusing when I let someone use my old netbook and they comment on how fast the pages load when compared to their phone, when their phone shits on my netbook spec wise.

>>55724500
*nose erection*

>>55734573
I'll agree that is nice, KVM switches were awful. Games harness the potential more than any modern software, except games are always poorly optimized and will only run acceptably on the coming generation of hardware that is never out yet.

They should have remained with older game engines and just continued to create new games. They always chase engines and now we have less than 5 major game engines that companies use and all are very poorly programmed.

Most of the excellent games from the early part of last decade ran on the Quake 3 engine (I forget which id Tech #). Jedi Knight 2, Wolfenstein... this also happened to be the last game engine primarily designed and programmed by one man almost entirely.

>>55723522
They will only stop shrinking if we continue to use the same materials.

>>55734430
Yes
http://www-03.ibm.com/press/us/en/pressrelease/47301.wss

File: steve_koolaid2.jpg (34KB, 416x405px) Image search: [Google]

34KB, 416x405px

>>55728025
>IBM
>I buy Mac

>>55723609
they can't even safe themselves, and you are talking about saving industry?

TOP KEK

>>55727554
1 month untill Rothschild, Rockefeller and Soros will fuck your mother.

>>55726267
>embedded
>lucrative

kek, embedded is probably the least lucrative market of all as it is extremely competitive and price sensitive. High margin low volume servers are where it's at right now if you want to be rolling in money.

>>55734130
>Of course, the human brain was shaped by nature and computers are just flat out designed by humans and will never evolve.
Nope.
https://en.wikipedia.org/wiki/Evolvable_hardware

We can use a computer to design a computer that will make better computers. This has been going on since at least 2004.

>>55731176
The problem isn't with parallel computing power. You could stack that shit as high as a skyscraper and a sequential program would have no performance boost.

>>55728830
> and we've already made significant progress.
sure we have

>>55736782
Nice meme

>>55736804
not as nice as the singularity meme
go watch another Hollywood movie about AI so you have more psuedo intellectual garbage to spout

Carbon is the answer to our "size" issue. It behaves with other elements almost exactly like Silicon, half the atomic mass, and half the size. Theoretically we could get to Single Atom Transistors easily once the research furthers, and we burn through Carbon's transistor size.

>>55736706
Well, no one says you have to stack separate cores.
Any one core can and should be made on several layers. The biggest advantage of that is locality - imagine a large cpu core, for a signal to travel from one end to another it takes a really long time for the nanoscale. But once you can "cut up" a core into pieces and connect them through the third dimension, one can shorten signal travel times by orders of magnitude, if done right.
It can be compared to "long distance space travel with wormholes", 'cept applied to ICs and for relaying signals instead of moving spacecraft. The fundamental principle is more or less the same.
Thus, chip stacking can greatly increase performance of CPUs by simply allowing much faster clock speeds due to shorter travel distances.
This should also reduce power consumption for the same performance because shorter "wires" == less power dissipation.

File: TheLastMimzy.DVDRipintel2_1_.jpg (23KB, 640x271px) Image search: [Google]

23KB, 640x271px

dont worry guys, intel is going to create an AI bunny and send it to us from the future

File: MZ4.jpg (49KB, 800x600px) Image search: [Google]

49KB, 800x600px

>>55737704
hel;lo

>implying Sandy Bridge isn't all we need

>>55738282
Sandy Bridge Xeons maybe.

>>55738282
Fuck you I want to do things faster.

>>55738584
C O C A I N E
O
C
A
I
N
E

pffft, nanoscale is so 2010's anyways

i'm going picoscale bitches

>>55726067
>You can't actually travel back in time.
there are legit theories involving 2 black holes that think otherwise

>>55723522
save your electronics boys

>tfw my old GPUs and laptops still retain most of their value in the coming years

>>55739705
You mean travel to the future.

>>55725986
Well, i can see 2 reasons. Either there is no future to speak of. That or there has been and John Titor was right and actually from the future.

File: SamsungAndroidFridge[1].jpg (68KB, 550x320px) Image search: [Google]

68KB, 550x320px

>>55733556
Bitch, by 2216 you'll be quantum shitposting from your fridge

>>55740145
>The year of our LORD is 2216
>still living in dark ages and thus needing to use a fridge
what happened to breaking down matter and then re-materializing as ready-to-eat, as-fresh-as-it-gets food, anon?