Why are companies not able to build processor that can operate on 9GHz for example by joining CPUs in Serial formation...

Why are companies not able to build processor that can operate on 9GHz for example by joining CPUs in Serial formation that every processor takes part of work on itself.
By this logic we should be able to build 1000 GHz or more processing CPUs

Attached: cpu serial.jpg (756x471, 50K)

Other urls found in this thread:

homepage.cs.uri.edu/faculty/wolfe/book/Readings/Reading04.htm
en.wikipedia.org/wiki/Amdahl's_law
twitter.com/NSFWRedditImage

Adding more cores does not reduce the time to perform single operation.

never post again

That's still x amount of cores @ 3GHz doing work independently, it's never the same as having 1 core operate at 9GHz. You'll still have to optimize the shit out of programs to efficiently do thinks in parallel with multiple cores.

if you have 2 horses (each 25HP) you don't have a horse with 50HP
You can't just add cores and sum clock frequency of it and call it a core with the n of all cores added up.

how can one be this retarded
read about hazard in digital logic

homepage.cs.uri.edu/faculty/wolfe/book/Readings/Reading04.htm

In those companies there are people working that are many times smarter than you and came up with a pleothora of solutions to make cpus faster. Do you think your brainlet tier idea was not considered even once?

Yes but you could group it as having 1 horse that performs as 50HP
Its important work they do so 2 horses (25HP each) perform work as 1 horse that has 50HP

because someone else might tought of it, it doesnt mean it was tried to be implemented

If each horse is 25 horses when 2 horses would be 50 horses which would be 50x25=1250 horses.
Divided by two that's 625 horses per horse which is more than 25.

/thread

2 horses = 25 horse power
25 * 25 = 625 horses
625 * 25 = 15625 horse ...etc.

So I create infinite loops and infinite energy.

How much LN2 would be necessary to make a 10GHz overclock?

wow
go away lol

t. retard

Attached: 12860531641161.jpg (640x480, 133K)

3 x 3 Ghz is not 1 x 9 Ghz retard.

Technically yes but a 50hp can't exist because of limits. So you can't compare it to something that doesn't exist.
1HP =/ 1 horse You fucking underage summerfags

Were you living under a rock???
I have amd ryzen 1700 with 16 threads
Each thread runs 3GHz base clock
So 16 * 3 = 48GHz!!!

And when intel releases 28 core 5GHz CPU as they promised and demonstrated their prototype it will be 28 * 5 = 140GHz!!!
Innovation of intel corporation will be first to break 100GHz barrier!!!

Make sure to support intel for your best computer experience!!!

>So 16 * 3 = 48GHz!!!
Technically this concept is true, you do have that much capacity, but you will almost never see a workload that could actually use it in a way that results in that kind of productivity.

Also, threads aren't the same thing as cores, you don't get double performance from another thread, so it would be more like (8*1.5)*3=36 or something along there.

Sending data between CPUs like you're talking about takes a lot more time than you expect. Transfer via RAM is about 100ns (just executing the next instruction would take about 0.3ns), and if you try to go via a bus you've got other problems (n^2 routing and bus delay).

>Moar cores!
You -can- just stack CPUs together on the same chip, but they still can't talk to each other fast enough, so you'll share data probably through the cache. A cache hit is about 5ns, which is a lot better than going out to RAM, but you're not going to execute a single thread any faster that way

>Moar threads!
Actually, we can run multiple threads on the same CPU at the same time. Most x86 CPUs will have at least two CPU threads ("hyperthreading" in intel-speak) that do this. But, same thing -- this is only possible because you know there's no data dependencies between the threads, and so you can cut out the hardware that fixes these depenedencies.

>Just run more at once!
If there's two instructions that need different hardware, we actually do run them at the same time, and just fix things so it looks like they were in sequence at the end. Again, doesn't work with data dependencies.

>Get rid of dependencies!
We've got register renaming, so if the dependency isn't actually real, we'll execute everything fast. But sometimes you just can't get rid of it, there's a sequence that has to be executed in order.

>Let the compiler do it for you!
This results in VLIW. It turns out that trying to get rid of all your dependencies this way is (1) NP-Hard and (2) usually not very effective. They tried to build computers like this, and eventually we got a GPU. Good at some things, but not purely sequential code.

Some old hipster computer architects are designing something called the "Mill Architecture" that tries to sidestep all of these problems, by replacing a bunch of register hardware with a FIFO. They say they can get 50 GHz... but long odds those chips will ever see the light of day.

Is the bottleneck heat? So if for example Intel release a 1 core CPU, could they clock it at a higher speed? As the CPU doesnt have to run 8 cores or whatever.

Only this guy gave decent answer even tho I don't understand half of it.
All you other fags just say : hur get out, idiot, hur like you are all top experts in field of CPUs .
I hate smart asses that can't explain whats wrong that know just to say idiot.

>tfw no cpu trap gf

>something called the "Mill Architecture"
Vaporware, those faggots have put more time and effort into simulating CPU processes than making an actual fucking CPU, they arre just investor scamming at this point.

>I don't understand half of it.
Are you retarded? Why are you here?

the problem isn't the design of the cpu, it's about the type of job you're running
many kinds of jobs are serial in their very nature, where tasks depend on the output of previous tasks, these kinds of jobs cannot be accelerated with more cores
splitting jobs to run on many cores is a software problem, which is already a thing. OSs already manage running many applications across many cores, and certain software like video encoders will perform tasks in parallel across a number of cores

but wait aren't cpus basically providing electrical impulse? So if there are more impulses(bits) that means more processing power per second?

>tfw your question gets ignored

Attached: 1504238465906.jpg (495x362, 19K)

So, you're implying that instead of having 3 units of time per a operation, we have 3*3?
I can understand if you mean we divide 9 by 3 and do those 3 things simultaneously, just as the 9 would in the first place, but that's not how it works. You would have to know the end result on one before the second one could even start.

This entire thread belongs in /sqt/

Slow it down to one operation a second. The processor takes a number, x, and adds 2 to it. The other processors now need this x+2 for the next operation so they have no choice but to wait.

You can just use a 3GHz clock for all three

nah, op just wants more bipelining.

Isn't it important how many Hz (electrical impulses) processor generates

Attached: a.jpg (456x853, 239K)

Hello AMD CPU designer.

you'd need liquid helium for that

en.wikipedia.org/wiki/Amdahl's_law
tl;dr: using 3 cpus doesnt make it 3x faster, also there is a limit to how much faster you can get with parallel computing

High frequency = more heat. You don't want more than 3GHz except for extremely special applications. Even for muh gaming you shouldn't go over 3.5GHz unless you want to boil eggs in your PC.

ehh... my i3 runs at a fixed 3.7ghz and it barely gets over 40°C.

>using power as a measure of frequency or speed

Epic

CPUs are obviously cooled, the point is it would emit less heat on a lower frequency.

Y. E. S.

the implication in OP's post is that there is no synchronization overhead

How do you keep them 1 clock cycle out of phase with eachother at an affordable price?

There's data inside the CPUs that are needed to know the result of the next operation.
For example if you do:
load 10 to a
add 20 to a
store a in b
And tell each CPU to do one of the instructions, you would end with a CPU with a as 10, one with some random retard number in A and another with a retard number in a and b, instead of having 30 on both A and B as expected.

which is impossible. also amdahls law tells us that even with 99% parallel code you can get a max speedup of 100x.

>all these posts
>nobody has posted cow2beef.exe

Attached: 1531542666395.png (1341x1334, 390K)

laughed too hard at this

how to learn how to SIMD

well, even if.
it still wouldn't be viable because you would need n-times faster clock speed and probably shorter rise and fall times.
higher clock speeds need have higher switching losses, which would negate your hypothetical power savings gained by having lower duty cycle

Fucking frogposter. There is more then one limiting factor, one being the speed of light and another being the size of silicon atoms.

>2018
>IT took almost 50 posts for someone to post cow2beef
Nu-Jow Forums is garbage

You can do 9 Ghz. You can do low voltage. But not both at same time.

this

disregard that, super-scalar cpus ,(aka. everything made in last >20 years) work nearly exactly like that.

your drawing doesn't work because the blue chip is doing nothing while red and yellow are working i.e. 3 chips are working only at 1/3 capacity = no increase in total processing power. There are two immediate & obvious solutions:
1) make it so that all three cpus can work independently at full capacity at the same time = parallel multicore execution (already exists)
2) make it so that when the code is "handed off" from the blue one to the yellow, the blue one immediately gets a new piece of work = pipelined execution (already exists)
Neither solution gives you 100% straight multiplication in power because non-trivial computer program are never so cleanly cut for for execution.

ITT: Neanderthals cant into Pipeline processing, smash rocks together and screech.

kek

The short answer is yes
but not with any hardware that is actually in use and I can't think of a use-case that would requre this

Heat scales exponentially with frequency

3/10 got me to reply

>not even the full pic

That was very clearly a joke man

because 99% of aplications arent made for more than 1 cpu, fuck, even a lot of programs still dont use more than 1 core

9Ghz = 3 magma

Yes. The higher the frequency, the more heat it generates and the more power it requires (both for cooling and operating). Heat also reduces the lifespan of the CPU. So it is usually better to optimize the way the CPU handles instructions than it is to run the cpu at a faster rate. This is done in many ways, such as more cores, the use of cache, the use of dedicated circuits to perform specific instructions that would usually take a long time such as aes instructions and so on.

Actually horsepower has nothing to do with horses.

i saved this thread for a reason you cunts, answer me
how to into parallelism, specifically SIMD

/thread

This

based

Why can't companies just put us on a near-lightspeed spaceship so time flows much faster for the processors?

You can’t expect 6 niggers to fuck a woman and expect her to born 6 babies

Attached: puppies.png (388x330, 292K)