Questions about Intel's CPU architecture improvement system

jesus christ you're retarded.

>>60792570
Why? I know I'm probably wrong but I'm open to learning. Because I'm genuinely curious

File: 1496157214563.jpg (2MB, 2736x3648px) Image search: [Google]

2MB, 2736x3648px

bump

Did I post at the wrong time or /g/ just doesn't know?

>>60792268
>questioning development department
>just keep shrinking the die
>JUST
Making cpu is hard. It is very hard. It is so hard that the few people who are capable (and willing) of doing it are payed in bitcoins and gold.

If you want to know how does it work, read a lot of books, or go to software engineering course on some decent uni.

Algorithms, instructions, microinstructions, logic, math, compilers, designing circuits, transistors, physic limitations, chemistry and 999 tricks cpu use to be faster while keeping cost low like caching, jump prediction, pipeline, hazards, write back, in memory operations and other stuff i am too lazy to even remember back from school.

Your PC is fucking complex dude. And cpu is pinnacle of it all. It can do so much stuff with so little instructions it is amazing.

>just smaller die
You have to realize that they are at dimensions close to the wavelength of light. It is anything but JUST. To make new processor you have to face so many engineering and technical dificulties and still make it cheap so somebody will actually buy it.

Go read book, educate yourself and stop shitposting on /g/.

>>60792912
I'm not saying it's easy though. I'm just questioning why they do it. Why keep shrinking the die and then sell it as an improvement? When it's improving nothing for the end consumers?
I know making a cpu is really hard, hell making any hardware is really hard. But look at Nvidia's gpu improvement.

So I guess my question is as making a cpu really hard why doesn't intel just take them time and do it? Why make a new cpu every year by just lowering the tdp 10w. Nobody cares about the die shrink. If intel stopped making new processors after sandy bridge and came up with something new instead of shrinking the die and selling it as a new architecture wouldn't it be more beneficial for them and us?

>>60793075
Decreasing die mainly
>decrease cost
>allows to speed up base clock or
>lowers heat amd energy consumption

You say new architecture, but that requires a lot of brain power and practically autism genius at this point. Cpus these days are as optimized as possible and come with something new and better is kind of impossible.

From bussiness standpoint, making whole new architecture is suicide. You need compilers, new operating system and tell to all software giants to recompile (or rewrite from scratch) and release their programs again for new platform. Suicide.

Shrinking die was kind of "easy" and logical step in development, but since the light wavelength has its limits and the electric impulses starts to behave really weird (quantum mechanics and other funny physic), we sort of hitted wall. If not at 10, it will be 9 or 5. Doesnt matter.

Unless somebody comes with something new, we are stucked with this.
>why they do it
Why to make the best, when good is selling better?

>nvidia gpu
Gpu is different game. Gpu unit is super dumb cpu which cant even divide numbers, all its computations have wide error margins and its development could be described as
>add more cores tightly next to each other
You dont care if some computations are off on it, it will generate few pixels off on your 4K screen. If same error happened in your CPU, your data could get lost. Big difference.

Just see some openCL specifications about tolerated float errors.

>why make new cpu every year
It is the same cpu, just different name and higher price tag. Why? Because it generates profit.

>more beneficial for them
No. You want money. Making new cpu costs money and can flop hard. Selling the same shit all over again is much better strategy.
>for us
We dont have comunism, they dont give a fuck about customers. Only our money are important for them.

Dont mix up technology and economics.

New mcu architecture comes out all the time, making existing architecture perform better is hard.

Reducing die size has two benefits, it allows more processors per wafer and it generally increases speed by reducing transistor size, bigger transistors have more intrinsic capacitance, lower capacitance means faster switching speed.

>>60792268
You're actually correct

Intel shuttered their CPU division when it looked like they had a stranglehold on the market with a non-competitive AMD who, by all appearances, were 2-4 generations behind Intel. Why throw money away on improving the hardware?

Intel cut back on R&D and stopped investing into the CPU division. It was more important to get on the SJW train anyways.

Now AMD has come out with a current-gen CPU, because you don't actually need to pointlessly go through those 2-4 in-between generational steps if the tech for current gen exists. AMD now has a competitive CPU, and Intel has an underfunded CPU Division with no real goals or plans for the next 3-4 years except "screw the customer some more" and "incremental updates!"

The knockout punch will come with zen+ though. That's when AMD does a die shrink and adds architectural improvements while Intel is still struggling to get their 10nm process under control, and Samsung+Global Foundries just showed off a 5nm process. Intel is literally fukt.

- AMD's architecture allows nearly limitless scaling with infinity fabric; Intel only has this tech as proof-of-concept, it's not implemented in their current architecture and probably won't be ready for their next one either. Which means Intel cannot compete in servers/HEDT until 2018/2019 at the earliest.
- AMD is moving to 7nm at the same time Intel is moving to 10nm; but Intel is having serious issues with their 10nm process and admitted it may regress performance - I'm sure intel jews were ecstatic at the thought of selling worse processors as brand new, too bad market competition caught up.
- When Intel moves to a 7nm process, AMD will be moving to a 5nm process.

Intel shot themselves in the foot by eliminating long-term CPU development plans and innovation in favor of anti-customer practices and stagnation. Just look how they're rushing out hardware DRM as if it's a selling point of their product rather than a reason not to touch it.

>>60793075
Die size drops if the process granularity drops.
You cannot necessarily make same number of cores faster if you use more transistors because of what x86 is.
Compare transistor count instead of die size.
Oh and 7700k has integrated graphics so actual processing size is even smaller.

>>60793343
Not the anon you were typing at, but this seems like a damn good explanation, and without any Intel vs AMD stuff, either. Well said, sir!

>>60793343
Well then you just proved me right I guess. They aren't innoveting everything and they're cpu improvement is just wrong and they just want to jew us for more money.

>>60793688
Yeah I know moor's law limits what we can achieve on a single core. But my question is to not why not more cores, my question is why not better cores even if they take a bigger die

>>60793799
>jew us more
EVERY COMPANY WANTS TO JEW YOU. That is the point of capitalism. Some companies just want to jew you mildly over long periods of time and some go for quick cash grab and run tactics.

So whatever you do, dont ever become fanboy. Always search market and before you buy anything, ask yourself
>do i even need such a shit?

And yes, intel probably fucked up by ignoring amd. You would have guessed they had spy in amd's R&D, but it seems like he was double jew or simply failed.

So dont except intel recovering in the next few years.

>>60793880
You clearly did not read my post. You cannot make each core better with more transistors, x86 architecture limits it. VLIW architectures appeared for a reason.

Let's say, you cannot sum 2 and 2 faster if you have two calculators instead of one.