Bringing back the Gigahertz?

Champion Of Sparta

Reputable
Dec 3, 2014
299
0
4,790
Hello all, I just had a strange thought, back during the pentium 4 era everything was about gigahertz and how high you can overclock it.....well I just had a idea to bring that back to life. Since the pentium 4 was a 120nm cpu and today the cpu's are at 14nm why dont we make them 120nm again? Granted im not computer expert but since we have the ability to make 14nm processors wouldnt it be possible to keep it in the same form but 120nm. Instead of making hundres of thousands of tiny pieces work in a really inclosed space have them spread out, distributing the heat. Heck if it will work wouldnt it be possible to run processors without heatsinks or watercoolers entirely at 4ghz? I would seriously like your guy's input on this, and if I wasnt clear on anything please explain.

I also had a second idea though, why dont we make the cpu larger. Keep the nm the same but inrease the size of its surface area, to reduce heat and increase overclocking.
 
Do you hear your phone ringing? That's Intel - wanting more information. The knock on your door? That's the AMD guys calling. The email - well, that's Nvidia. The big helicopter? That's IBM coming to license your idea. Apple, Google and a plethora of other companies aren't far behind
 
Pentium 4 had 42 million transistors and was 217mm² in size, and consumed around 100w of power. A modern Intel CPU has close to 2 billion transistors, and consumes <65w. If we designed Skylake on the 120nm process, it would be 10cm x 10cm across, or roughly the size of a CD/DVD, and would draw around 5,000 watts, which means you'd have to have a special power outlet installed, and it would produce almost as much heat as as 4 large electric space heaters... in the area of a CD or DVD.

As transistors get smaller, they also consume less power, for reasons I don't feel up to getting into right now.
 
Pretty sure the transistors need to be close together. Spreading them out would require an increase in power which in turn would increase heat and defeat the purpose of said scale change.
From my understanding the reason 14nm is faster and uses less power is because the components are closer together (and also smaller).

It would have been interesting if they kept the same footprint and just stuck more into it. Though I think the power and heat would have been overwhelming for any cooler to handle.
 
The way to reduce power consumption and cost of a CPU is actually to make the transistors smaller and closer together, so they produce less heat and are cheaper to manufacture. Once you do, you can build larger and more complex designs for the same price and power usage, or the same designs for less cost and less power usage.

A Pentium 4 on 14nm would probably draw less than 1 watt, and be pennies to manufacture.
 

Quick question then, couldnt we just stuff the crap out of the 120nm size with the size of transistors there are today? We might be able to include trillions of transistors where the temps wouldnt go past 30c without a heatsink 😀
 

But lets say we make a single core'd version, how well would that hold up. Instead of all those transistors trying to disipate the heat of 6 cores it will only try to disipate the heat of one core, less cores + less heat = $$$$OVERCLOCKING? 10ghz cpu lol.
 
Heat and core count aren't what prevent us from scaling up frequency, generally. Or, perhaps, it's a lot more complicated than that; sometimes heat is a limiting factor, but more generally it's that you have to pump dangerous amounts of voltage in to get past a certain frequency, and you risk damaging the transistors.

A single Haswell-E core would likely not clock any higher than it already does, and Intel has not yet been able to design a 2 billion transistor single-core CPU.

I would assume it's possible for Intel to design a transistor that's capable of tolerating more voltage and/or capable of higher frequency, but they have deemed it economically unfeasible, or they'd be building CPUs out of them.
 


Thank you for the well educated answer, guess my idea got show down lol...and not with handguns.....were talking cannons 😀
 
Here is an 80386 I came across today in my pile o crap. Mid 80's.
https://en.wikipedia.org/wiki/Transistor_count
2YBgXVN.jpg


1,500 nm process vs 14nm process with current CPUs.
275,000 transistors, vs 1,400,000,000 with a current i7.

Basically the same size chip.
If we were to try to put i7 capabilities into this die size, it would be larger than your desk.