Why can't CPU's be bigger? why do they insist smaller = better?
Motherboards are still fuckhuge anyway, so why shink if you can fit a tonne of more power in a bigger CPU?
>>53156445
Compensating for something op?
>I don't understand how CPUs work
>>53156445
Heat might be one reason
>>53156445
Because even though the one on the left is smaller it has billions more transistors. CPUs are one of the few areas where smaller means better. The smaller the manufacturing process, the less power it consumes and in turn less heat it puts out. Also dies can't just be as big as you want
I had ome of these big things like 4 years ago, found it somewhere. I think it was a dual core, special server cpu, dont remember.
Sold it for 21€
>>53156527
*right
>>53156527
Interesting, but with a bigger CPU, couldn't you fit billions MORE transistors thus = better performance?
>>53156527
then why don't they just make these smaller ones bigger
>>53156510
Because it wouldn't be easier to dissipate heat with a larger surface area.
Because those things are so fast that a few µm physical distance can have a noticeable effect, since electrons literally need longer to travel those distances.
>>53156574
see 1.7%
>>53156593
>>53156576
/thread
>>53156445
Basically, the speed of light. The signal needs to go all the way through the chip in the time it takes the CPU to go through one clock cycle, which is very little time when your CPU is running at 4GHz.
>>53156575
>>53156574
>Also dies can't just be as big as you want
>>53156640
>>53156527
Wow, I was sure it was just a fad, to make things smaller and consume less power. It's still a bit funny to hear shrinking something can produce bigger performance.
Actual Computer Engineer here.
*Cost: This one is huge. Increasing die size not only means that you get less chips for every wafer because of size, but also means that you are more likely to get a manufacturing defect on a specific chip. Once you get big enough, the chip would be almost guaranteed to be faulty. This greatly reduces yields, thus greatly increasing costs
*Propagation delays: This will fuck you hard in many situations. Signals take time to travel.
*Heat/power consumption/power dissipation: This one is pretty obvious.
By the way, the actual die is much smaller than what you see in OP's pic. Pic related is quite a big one.
>>53156706
thread over
>>53156706
And by the way, instead of doubling die size and transistor count, the logical thing to do (which is what is done in practice) is to just use two chips (or whatever many). See servers with multiple CPUs. This brings by itself many issues, but far outweighs big-ass dies issues.
>>53156706
as a computer engineer what do you think of gamers?
>>53156813
What do you mean?
>>53156826
do you like them hate them think they are idiots for spending 500 dollars on a video card?
I wish Slot CPUs were still a thing.
>>53156882
Daughterboards in general are fucking cool
>>53156882
why?
>>53156882
I still have one around somewhere.
>>53156445
smaller = less energy lost on transporting electrons = more efficient = faster
OP = Fag
I dont get why manufacturers use this shit ball soldering on onboard chips like gpu instead of using pins. Just solder the pins if you dont want hardware removed, but this shit costs me mad dosh. 2 gpus dead cuz of faulty soldering. Also 1 laptop with shit northbridge solder fault.
Fucking
>>53156910
>daughterboards
go get raped by a large blackman you faggot dont say shit like that
>>53156882
They still exist in some server builds I think. That was a few years ago.These days it's co processors and blades I think.
>>53156979
That's what they're called doe
>>53156979
Get rekt faggot
daughterboard |ˈdôːtərbôːrd | (also daughtercard)
noun Electronics
an expansion circuit card affixed to a motherboard that accesses memory and the CPU directly rather than through a bus.
>>53157016
no fuck you the word annoys me dont use it
CHANGE IS BAD
>>53157034
If change is bad then you're the one in the wrong
>>53156869
Why the fuck would he care?
>>53157034
>change is bad
What's changed?
>>53156979
thats what they're called though you fagtron
>>53156869
I really don't mind what people spend their money on. If people spend 1000 dollars on a big dildo, it's their money, if it makes them happy why would I mind?
Anyway, I play from times to times (even developed games myself) so yeah. What I do find disgusting is playing with <60fps or with less than native LCD resolution.
Also I obviously find things like pic related retarded, or stupid marketing terms like "MILITARY CLASS" and the usual 13 year old edgy hardware like https://i.imgur.com/rPQPD9b.jpg
or 99% of computer cases
Please just give me a nice, plain, mechanical keyboard with MX Blues.
>>53156979
That's what they're called, negro.
>>53156882
>wanting to throttle your CPU to PCI-E speeds
>>53156974
It's not the ball solder that's the problem, it's the fact they use shit lead free solder because eurofags are ignorant fucks that think that there's some health risk.
>>53157126
It still connects like a normal processor you dipshit
welp this thread derailed hardcore about 25 posts ago.
>>53157102
>>53156974
It's done that way because it's cheaper of course. But what really happened is they got forced to switch to lead free solder some years ago (I think around 2006) and that really fucked their shit. They got their shit right now, and with recent hardware it's very rare to get soldering issues.
It's pretty much an issue I think with 2006-2010 hardware because it took them some years to figure out how to properly use lead free solder without it failing
>>53156574
>Interesting, but with a bigger CPU, couldn't you fit billions MORE transistors thus = better performance?
Yes and no, The whole point is incremental development. Intel does a Tick-Tock, they work the architecture and then shrink it. You want to increase performance and decrease Heat/Energy demand.
By taking current technology and filling a massive die, for reference the pentium pro. It used a .5um to .35um process and consumed 45w at its peak. While only shittiing out 200mhz it would probably get single digit passmark scores.
the pentium pro/2 die is nearly 200mm2
You then take a modern architecture like skylake which uses a .14nm process (0.014um) and fill the old pentium pro die with moar coars or just have pipelines to alaska and back. I can't find numbers for skylake but i did for broadwell (82mm2) and at least half that is the eGPU & Memory controler so we will say 41mm2, it also draws much more power, up to 95w in performance versions. Besides the CPU will limit power to itself to give the eGPU power or vice versa just to stay within the TDP
Pentium pro Watts per mm2 = ~0.225
Skylake watts per mm2 ~2.317
I am just gonna take a shot in the dark but it could draw well over a kilowatt if not into the tens of kilowatts.
You would have a CPU that draws a self destructive amount of power that would be impossible to cool.
>>53156445
>why do they insist smaller = better
Sell you smaller cheaper to produce shit for the same price as before
>>53157227
>I am just gonna take a shot in the dark but it could draw well over a kilowatt if not into the tens of kilowatts.
>You would have a CPU that draws a self destructive amount of power that would be impossible to cool.
Jesus fuck. Welp scratch my post with a bag of nigger dicks. Guess shilltel is right for smaller=better and amd fallowing
>>53157213
Thats why my acer and gpus died, theyre 2009 models
I hope my 2011 acer doesnt have the same fate
>>53156706
Also, surface area decreases in ratio compared to the rest of the chip.
>>53157034
>I DONT LIKE THIS
>REEEEEEEEEEEEEEEEEE
>>53157303
Gamegate in the nutshell.
>>53157303
Computers where made by a gay man who liked incest.
>>53157332
well that came out of the fucking blue.
>>53156882
you literally don't though. the voltages, wattage, and phasing of modern CPUs is WAAAAAAY too finnicky to be trusted to a slot, not to mention that massive number of pins you need to map.
I mean fuck, that CPU barely gets all the pins on and it's only got 242. imagine how dense you would have to make the connections for an 1150-pin slot.
>>53156445
i agree op
a real man likes his processors blazing hot sucking electricity and loud as a cement mixer
Bigger cpus are slower because the signals take time to travel.
Sure, you can shove more shit in, but it will not make your single thread go faster.
>>53157290
Intel was never right about anything. It's common sense.
>>53156445
because signals have to travel from one side to another
and more silicon = less cpus per wafer
>>53156445
Because it is better. Physically smaller = superior. In terms of energy usage, and performance.
It's also why multi-cores are preferable to multiple CPUs. Albeit you can have multi multi-core CPUs as well (which is what high performance machines tend to have now) but for now just assume it's 2 or more single cores vs 1 multicore with an equal amount of cores to your multi-CPU setup. The cores of the multi-core CPU are closer together, and confined to a smaller physical space, therefore they can communicate with each other faster than the multi-CPU setup. The multi-core also draws less energy, and produces less heat. I am oversimplifying things though.
>>53157951
>Physically smaller = superior.
sup whitey
>>53156640
This guy beat me to it.
Basically if they are made any bigger the electrons won't arrive at their destinations during the clock cycle in which they are expected. It will lead to unpredictable behavior from the CPU.
>>53157290
>Guess shilltel is right for smaller=better and amd fallowing
>amd following
AMD isn't following shit. This has been the industry trend for literally decades now, long before you were even born.
>put 2 cpus on board
>works perfectly fine
>make cpu bigger
>durr impossibru
You're just getting jewed into buying 'power-efficient' CPUs. Cooling and power usage are not a problem anymore.
>>53158112
faggot
>>53156445
Nigga, you can't be srs
>>53156445
Let the Admiral explain:
https://www.youtube.com/watch?v=JEpsKnWZrJ8
>>53158003
Remember the brief period where a new process meant higher clock speeds? The future seemed so bright back when we didn't know shit.
>>53158112
It's been explained multiple times in the thread why CPUs can't just be as big as we want, and the reason was neither cooling nor power usage.
>>53156445
google it you fucking retard
>>53158263
But those reasons suck.
>somehow putting 2 cpus on a board is viable
>somehow distributed computing is viable
>somehow putting more transistors on a wafer isn't
>>53158337
>those facts suck
Well they're why it's the way it is so just fucking accept it.
>>53158337
>I disagree with the speed of light
Alright. I'll be sure to inform whoever it is that enforces that speed limit.
>>53156445
Ive got 6 of those pentium pros in a box including the black one with 1mb cache. They were beasts in their day apparently.
cuz then you'd get more cores for free.
jewtel wouldn't like that.
Smaller cpus = more cpus per wafer = more profit per single wafer.
Also smaller transistors = lower power consumption. That's just how silicon works.
Reminder to invest in IBM:
http://arstechnica.com/gadgets/2015/07/ibm-unveils-industrys-first-7nm-chip-moving-beyond-silicon/
www.cnet.com/news/ibm-spends-3-billion-to-push-the-far-future-of-computer-chips/
>>53156882
Not quite slot, but the 2nd CPU module for the Dell Precision T7500 is a daughter card. Kind of obnoxious, but it does exist.
(LGA 1366 / 6x DDR3)
>>53157126
>>53157174
Heh. Clueless youngsters trying to be cool on the internet? Priceless.
>>53156882
TFW when no L2 cache chips. Hopefully that is a Mendocino.
>>53156575
because if they jew the shit outta you with 2 or 3 billion transistor on ~120mm^2 why would they consider doubling the count and area? (if other technicalities as keeping heat in check could be managed).
if they could make something decent enough for half the size they would definentely take that chance to double the profits on a single wafer.
tl;dr.
it's to jew you out.
>>53158263
those reasons are mostly bullshit. The main reason is to maximize profits per wafer.
And you can a bigger cpu OP (e- series, more than double die size and transistor count). It's just that instead of shelling 300 bucks out it would be 2k bucks. Not to mention some trade off in freq specs.
>>53162141
The 5820K is barely any more expensive than Skymeme and cucks it in both physical size and performance.
>>53161792
A lot of servers and workstations use daughter cards.
>>53157102
>>53157202
Let's face it, if you don't own one of these you are probably a virgin.
>>53161697
Delete this goy. I'll have you for hate speech.
>>53157227
just a note
.5um = 500 nano
.35um = 350 nano
.14nm process (0.014um) = you mean 14 nano
>>53156445
You can get the same thing and better in the future when multicore and multi cpu will be a mainstream feature.
>>53158337
>more transistors on a wafer
that the whole issue anon, if youre not increashing clock, why added complexity on a single wafer
multi wafer is the most efficient and effective way
you not thinking that ur game physics engine will double framerate with a bigger wafer instead of 2 cpus, right?
some things can only be computed in sequence (the parrallel guessing&discarding method can only go so far) and in the end its the clock that matters
and clock decreases as pathways get longer
>>53157126
>somebody wrote this seriously.
>>53157170
Explain yourself: What are the soundest arguments against RoHS and all lead-free jazz?
>>53163555
Lead free solder is inferior, iirc it was actually what caused the YLoD and RRoD in the PS3 and 360. As it heats and cools the balls stress and start to form little strands that eventually end up shorting with other points in the BGA
>>53163500
speed of light is the bottleneck.
if you want to see the future of computing, at least computers operating on relativistic principles, look at ways to locally increase the speed of light or use particles that travel faster than light such as tachyons.
>>53163614
I know that.
I mean, why lead solder isn't dangerous as eurocucks and ecofags say?
>>53163687
how about preventing infants from eating computers? fuck safety. gib led.
Read this, OP: http://www.edn.com/design/systems-design/4368705/The-future-of-computers--Part-1-Multicore-and-the-Memory-Wall
>>53156593
This. Processing latency goes up with size. Picture throwing a LAN cable to your neighbor's house and then complaining that the speed sucks. That's what your asking for.
>>53163717
>average europoor thinking
>>53163770
you're so fucking stupid.
>>53163813
You mad, bro?
>>53163837
no, i am not mad that you are stupid and wrong about probably everything in life, besides everything you think about this matter.
your reading comprehension is 6th grade level.
>>53157102
>hates stupid marketing and components that add no actual benefit
>has mechanical keyboard
Larger CPUs mean that signals have to go longer distances to get transmitted. That lowers performance. Smaller is better.
Your GPU has a larger die than a GPU because GPUs can run massively in parellel so the distance doesn't matter as much. CPUs are basically designed for just a few cores to have extremely high performance basically.
>>53163666
>tachyons
666
ok lucifer
go back to the mothership, they are about to set course to the nearest wormhole to planet inferno
>>53156813
those gamers go out and spend thousands on hardware every 6 months - year. hardware those engineers develop.
so i would think engineers love gamers.
>>53156706
mr engineer who do they not develop slot based cpus anymore like >>53156882 posted?
>>53163666
speed of light
3x 10^10 centimeters per second
30 000 000 000 so 30 GHz ?
im sure its not the simple to guesstimate this, but would it be limited to a single cluster of transsistors? say, a FP unit, the electrons only need to traverse the FPU and a nbew clock can begin for the fpu at least?
but still, hard to guesstimate how many milimeter that would be, i imagine
>>53164210
why not who
>>53156445
Because the size of the CPU has no bearing on the potential of the power of the CPU.
CPU dies are getting smaller and smaller every year, but the power of the processors is increasing, while the power these CPU's draw is decreasing.
And you can get an ITX motherboard if you want a tiny box, but why would you unless you are the type of person who carries their computer around to LAN parties ever week? They're cheaper, I guess, but they are extremely light in features compared to a full sized ATX board due to its compact form factor.
>>53156530
those had a large amount of gold in them
>>53156445
We do make them bigger, when we need to, but most of the time we don't, and we shouldn't.
The Pentium Pro you're using as an example took all of Intel's manufacturing might to produce, and for every Pentium Pro they could put out, they could have three or four regular Pentiums. The size and complexity made them very hot and mega-expensive, all for gains that the average home shitter would never notice.
