CPU
Images are sometimes not shown due to bandwidth/network limitations. Refreshing the page usually helps.
You are currently reading a thread in /g/ - Technology
You are currently reading a thread in /g/ - Technology
How small can these CPU's get anyway?? Is anything smaller than 5nm actually possible at this point???
>>51960901
In terms of feature size certain gate topologies scale down to 3nm, possibly lower depending on a host of factors. Independent from gate length area scaling has plenty of room for improvement since a total transistor including source and drain wells is large relative to the gate itself. Better isolation of components could allow transistors to significantly shrink in size without touching gate length whatsoever.
Ibm has made a 1,8 nm trannysistor, so its possible.
>>51960901
It's possible, but in semiconductor industry the outlook for scaling down is not very positive, you might have noticed this by the delays in 16nm from TSMC or 14nm from Intel.
Problem is not that it's impossible, the yield usually gets lower, and high yield is the main thing that drives the semiconductor industry.
Ultimately, you are limited with the wavelength of the light you are using for the photolitography, but they have workarounds for this (polarized lighting), and also they use UV and stuff.
Funny thing is even with devices scaling down, CPUs are actually not getting smaller. The actual die is getting bigger, because there is no where to go but to increase the number of cores as frequencies are pretty much around silicon's power density limit.
>>51962457
>Funny thing is even with devices scaling down, CPUs are actually not getting smaller. The actual die is getting bigger, because there is no where to go but to increase the number of cores as frequencies are pretty much around silicon's power density limit.
Oh boy.
Mainstream desktop die sizes:
45nm Lynnfield 296mm2
32nm Sandy Bridge 216mm2
22nm Ivy Bridge 160mm2
22nm Haswell 177mm2
14nm Skylake 122.4mm2
Die sizes are decreasing, even while more media accelerators, co processors, and progressively more GPU logic is added. Clock speeds are limited by leakage, and self heating. If the transistor operated at an extremely low drive voltage with no leakage then the only hard limit to clock speeds would be getting heat out of the device. Tons of chips out there hit 5ghz, and there isn't any sort of clockspeed wall that would prevent us from scaling higher if power and heat allowed. Design the arch to be stable at high clocks and it would function without issue.
How do they make computers out of the same stuff boobs are made from?
>>51962677
silicon and silicone are two completely different materials
>>51962457
>CPUs are actually not getting smaller
Don't be that guy. The CPU die is getting smaller, but demands for stronger iGPUs are going up, which is why ~50% of Skylake's die size is taken up by a larger iGPU. The CPU area itself is immensely small, but Intel bundles increasingly larger iGPUs on the same silicon.
w-why don't we just make b-bigger CPUs to fit more on it and make it faster instead of making the size of the transistors smaller? why are CPUs the size that they are?
>>51963073
electricity doesn't move instantly anon
the larger the distance, the greater the latency, they can only be big OR fast, not both
>>51963139
>an actual answer from a nice anon
wtf is going on with my 4chan
>>51963073
>w-why don't we just make b-bigger CPUs to fit more on it
First of all don't type like a fucking retard, secondly target die size for an IC on a given node is governed by dozens of factors like yield, percentage of dark silicon, and etc. The smaller a die the more candidates fit on a wafer, larger dies are more prone to defects, and you get less candidates per wafer.
> and make it faster instead of making the size of the transistors smaller?
Your entire line of questioning is based on an enormously incorrect assumption. A smaller transistor uses less voltage to switch, and that lower drive current allows you to reach higher clocks at a given power level. Smaller features improve performance per watt. Big transistors are not faster, that notion is just stupid.
>>51963139
Latency can have a noticeable impact to performance with SRAM cells, the logic itself has higher gate delay the larger the structure is, but this doesn't have much of a pronounced affect that a user would ever notice.
>>51963139
nice anon
>>51963276
not nice anon
>>51963139
Nice answer for us less tech saavy to understand.
>>51963276
Faggot who thinks he is better than everyone else. Gives an answer which still doesn't make sense unless you research the field and understand wtf a DIE / WAFER / DARK SILICON is.
GO DIE
>>51963370
>>51963318
Stop shitposting, retard.
>>51963276
>gives answer to someone that obviously doesn't understand CPUs at all in a way that only someone who understands CPUs would understand, yet you would never have to explain to anyway as they already understand CPUs and would never ask such a question
is this douchebag fucking serious?
>>51962992
That is primarily because Intel cannot into gpus. Iris pro has an enormous slab of onboard ram because Intel suck so badly at it.
In b4 "muh Intel".
Larabee? Never heard of it!
>>51963318
>>51963370
>>51963410
quit trying to make this into a big deal, fucking normalfags and their drama
>>51963472
>acts like a faggot
>"guys just stop cmon just let me be an asshole"
it's time to kill yourself
>>51963454
> Iris pro has an enormous slab of onboard ram because Intel suck so badly at it.
They're using eDRAM because they need bandwidth, thats all there is to it. The need for bandwidth has literally nothing to do with how good or bad their GPU arch actually is. They couldn't work a quad channel IMC into every part, and doing so for a mobile SKU would just increase board complexity which no one wants.
AMD's IGPs are horribly bottlenecked because they're relying on limited dual channel system RAM. Intel used eDRAM to avoid being in the same situation, and because they operate their own fabs and eat their own losses they don't care about how one high cost die yields.
>>51963472
Its all one faggot. The question about making CPUs bigger is old copypasta, and the anon who posted it was trying to stir up a shitstorm from the get go.
Just report all his posts. If we're lucky he'll get a ban and a janitor will purge the posts.
>>51963508
who, >>51963276? i'm >>51963139
>>51963276
>First of all don't type like a fucking retard
You're obviously new. You post in stutter when you're trying to convey that you aren't confident in what you're saying to imply that you are genuinely asking a question. Go fuck yourself you fucking nigger.
ITT one bad egg ruins a good thread
>>51964216
Its one retard.
Report and hide his posts.
>>51963276
>>51964231
pls go
>>51963527
not the guy you replied to, but that "stuttering" is still typing like a retard
what are you now, some kind of weeaboo role-player that cannot tell in- and out-of-character?
go drown in a fire you immitation piece of shit made in china
>>51964299
Its infantile meme shitposting.
>>51964299
you're obviously a fucking commie
>>51964310
That kind of proves his point, but this thread has gone off topic too much already. Sounds like a couple anons ITT at least know their subject.
>>51962582
>>51962457
THEY ARE MAKING OUR BROCESSORS SMALLER AND ADDING BOTNET WHERE HTE OLD DIE USED TO BE!!!11
>>51963518
AMD is three letters
IGP is three letters
AMD's is five characters
the third character in "AMD''s" is M
IGP and M
re-arrange
GIMP
AMD confirmed for gimp'd processors
>>51964809
How do you feel about yourself after willfully demonstrating you can't count?
>>51964836
He can count but how does it feel to not know the difference between letters and characters?
>>51965085
BTFO
>>51965085
>being this much of a laughable retard
Its too sad for words that you're this stupid and don't even realize it
>>51965176
*tips fedora*
Hey dudes, is MIRACLE SILICON-REPLACEMENT GRAPHENE revolutionizing CPU's yet? Every year we're assured that this will be the year of graphene desktop CPU's, and every year we're told to keep positive and wait for more developments when it turns out they can't make it work, FOR NOW...
Better question is, why are motherboard pins so fucking fragile these days?
I bricked my motherboard by accidentally touching the chipset just the other day.
>>51965946
Its pop sci bullshit mostly. Graphene is the hot ticket item for research right now so everyone is jumping in on it. If you start doing research with the material you're pretty much guaranteed to get funding from someone, and when you get a clickbaity article written on your research espousing all the vague possibilities you've discovered then you hit the jackpot.
Lots of materials have X magical property, they just display that property under a certain set of circumstances that make it entirely impractical for most things. Like with graphene lacking a band gap. Its not going to be replacing silicon substrates whatsoever. What might happen is that graphene crystalline structures may be built up on die and exploited, but that is an entirely different usage.
tl;dr is don't follow clickbait pop sci reddit bullshit.
Advances are brought about by millions of man hours of hard work, they shouldn't be taken for granted, there is no such thing as a miracle material, and if it sounds too good to be true then it probably is.
>>51965977
Could have been a tiny ESD.
I've bent pins then bent them back with an ink pen shell, the physical contacts aren't fragile, but a tiny ESD that you're incapable of even feeling can fuck things up.
>>51960901
they will always try to make it smaller.
