Why don't we have 3D CPUs?

3D PIG DISGUSTING

Are you trying to make a computer or a bomb?

Attached: not_enough_exploding_vrms.jpg (480x360, 43K)

Physical distance there doesn't mean shit, you can route CPU to motherboard connections around your house and that'll be better latency than normal CPU to RAM delay.

ok but what if i want the same power of multiple motherboard connections in one motherboard?

Slap an Nvidia sticker on that bad boy and sell it on ebay.

Heat. If you can dissipate the heat from a several layer thick CPU effectively, you can have your 3D CPUs.

OK, what's a good method of dissipating heat from a 3D CPU?

even with ordinary non-stacked chips like we're using now it's a pain to suck current levels of heat out of chips. Even if you build in a tiny liquid-cooling channel like said, that heat still then has to make it to the ambient air somehow. Stacking gives you more transistors but those are useless without an increased thermal budget to use them.

what if you use Magnetic refrigeration?

There is none, every stack concentrates an insane amount of heat to the surrounding chips. Every chip attempted this way essentially melted no matter the cooling solution.

You're legit better off raiding area 51 and getting a hold of asvanced optical cube processors that run on 1W of power desu senpai.

damn. what if the processors were made of memristors? don't those produce less heat?

The lowest CPU would be extremely heat constrained and chips going down the stack would have to run at lower and lower clocks. HBM was a pretty good example, the lowest HBM slice basically runs constantly near the Tjunction limit and this forces a hard wall on the speed.

There are some ideas like micro water channels but right now it would be extremely expensive. TSVs simply cannot handle the heat density of a 4Ghz+ part.

Corsair AIO

so the fact is that there's just too much heat? what if the architecture is changed so that there is a central channel in each core for heat dissipation? like a well for a microscopic fan to pull the heat upwards towards heat piping or whatever? even amd chiplet designs could support that right? since you could use the infinity fabric inbetween cores but just have a central heat exchange void.

fpbp

Im retarded but why cant you just space the "stack" out more? Just have a bigger case

You just almost invented multi-socket systems.

>why don't we have 3D CPUs?
you are the 3d CPUs

Attached: 1532999219399.png (1119x1080, 1.14M)

t.3D CPU

This. Mainstream CPUs are already constrained by the power we can feed them and the heat we can remove from them. Therefore right now, a stacked CPU doesn't gain you any extra transistors yet.

You can get heat out of the chip but you've still got to dump it into the air.

what if you route it to a big ass fan? like a gigafan the size of the house?

Thats what I was thinking at first(just put more sockets on the board) but if the idea of a 3D cpu works off 1 socket, as thats how I understand the concept, why does that necessitate the gap between dies being smaller than the length of a gnats dick?

doesn't the whole benefit of stacking CPUs vertically come from performance? like you have them interconnected, so i figured having a gap would make that nonexistent.

>slave die
I have reported you to HR for hate speech and discrimination. Your future communications will be monitored.

tunnels for liquid cooling are the subject of research papers, not actually implementable at a scale appreciable to manufacturers of the world's most complex mechanical mechanism.

It will more than double the complexity of their chips, resulting in decreased yields per wafer (they have to throw away anything that has a defect in it)

and now you will have stupid manchildren who think that they can just increase the power on their pumps to cool the CPU faster when it'll just cause the entire die to explode from the pressure.

It also makes each layer a heatsink for the ones above and below it.

Liquid technology is not at that point. This is one of the many reasons Theranos failed, they couldn't move blood through pipes that small.

Microtunnels through the chip could literally be done with etching trenches first, filling with something easily dissolvable, covering the whole smash with another layer of silicon, then building on the layer.
If that's not doable, then flip the wafer before/after processing and do the etching and covering then.
If that shit still isnt possible, then cover a completed chip's surface with a layer of glass or other electrically insulating material, stack the chips, then use the gaps between the chips themselves as the cooling passages.

Blood is different because it's like living tissue and has all kinds of dirt in it that can clog pipes.

It's not that the pipes don't work, but that it, like gallium nitride, is not a battle-proven technology and does not have an industry built around supporting it. You would need third party vendors beyond the chip fabs to provide pumps and what amounts to aftermarket support.
If we're going to see these things come out from anywhere, it's probably going to be something like radio/radar transceivers for the military. The semiconductor industry is perhaps unsurprisingly conservative considering the amount of money they spend on developing their processes.

The communication time between cores and cache become smaller, and in turn increases performance. Essentially, the speed of light is the limitation here, which is why photonic interconnects are such a big research field.

Attached: 1563555555586.gif (951x534, 706K)

Slave die and Master die?

Attached: sl54.png (374x340, 262K)

>It's not that the pipes don't work, but that it, like gallium nitride, is not a battle-proven technology and does not have an industry built around supporting it. You would need third party vendors beyond the chip fabs to provide pumps and what amounts to aftermarket support.
>If we're going to see these things come out from anywhere, it's probably going to be something like radio/radar transceivers for the military. The semiconductor industry is perhaps unsurprisingly conservative considering the amount of money they spend on developing their processes.

I simply figured that for that just make it so that the CPU IHS has some coolant-path routing like a waterblock so the coolant is directed to where it needs to go, then connect the IHS directly to the external cooling system. So long as the chips are protected properly and the coolant used isn't corrosive, it should work just fine.
And if need be, they can skip the initial microchannel step first and just seal the chips with a gap that coolant can flow through between them and start that way. We already know that chip stacking is possible and somewhat commercially viable, turning it into prototype flow-through tech should be a relatively easy step from there.

Is the speed of light really much of a limitation in practice? Its the fucking speed of light

Attached: 1492368271454.gif (192x197, 87K)

Yeah, that's cool. Now pay for the R&D, then pay for additional time to invent the additional things you need to do that at scale, then refit your factory to do it, then train some staff to understand what's going on, then design more architecture for the motherboard, then try to actually justify all that capital expense when you could put way less funding into just making a slightly better CPU with existing technology and making plenty of money.

The speed of light was always the limitation, even back when grandma hopper was building computers for the navy.

funimation.com/shows/serial-experiments-lain/kids
jump to like 21:00

Yeah but then divide by the clock speed and we're down to like 7cm per cycle now. This makes some sequences of work take an extra cycle because they had to travel too far back and forth across the chip on indirect paths.
The same core, if it could be fully designed in 3D, could have shorter-enough paths to change the timings and make higher clock speeds useful.
You can also have lower latency memory access with HBM stacks right next to the CPU die in the same package.

Attached: NEC-SX-10-03.jpg (887x768, 233K)

underrated

Attached: 1438714549462.png (520x533, 283K)

Lets see you make it, if its so obvious

It's moving that way.

>slave die
Now, this is problematic.

>what is viscosity
tunnels that narrow will suffer from that.

temps?

Terrible thermals.

Liquids don't work the way you think when scaled down to that size. You'd need an industrial pump for it to work at all.

Well, that's easy to solve. Now all they need to do is use superfluid liquid helium as the coolant. Don't thank me.

why not use an industrial pump then?

Why not just switch to the materials that work at 200+C or something? High temps don't matter if the TDP is constant.

that's even harder if not impossible

Solder and plastic melt before 200C and paper can ignite at like 220C.
Housefire would no longer be just a meme.

presumably op just means building the same effective chip but more like a small cube than a larger plane
that is, same tdp, but smaller maximum distance between two points, potentially allowing for higher clock rates

fpbp