High Wattage CPUs: Engineering Plausibility?

[Arqane]

Not sure if this would best fit here or under Computing and Technology, but since it has more to do with the engineering plausibility I'm putting it here for now.

I have a project which would benefit from CPUs that use a very high amount of wattage. Of course, this is the opposite of what the goal usually is for CPUs, but for my project I am looking for a few main things: high wattage in a relatively small package, relatively low cost, and then as much processing power as I can get with the combination of those two things. Inefficiency isn't a big problem, and heat is being taken care of for anything outside of the chip itself (internal problems with heat could still be an issue, but heat distribution/dispersion around the chip would be excellent).

So, theoretically, I'm not sure whether this makes sense or not. As an example, I'll use the i9 processor's, which has a max power consumption of 241W. If I could make a processor where the max power consumption was double (482W), and the processing power of that CPU was 1.5x a single i9 processor, but the size of the chip and motherboard housing was similar, then that would be a success. Especially if the cost stayed low, which I figure might be possible as the usual emphasis on power efficiency wouldn't be a factor.

So to sum up, from an engineering standpoint, would it be possible to make a CPU that is:

• High power consumption
• Decent computing power
• Compact
• Cheap

or would I run into engineering issues with making something like this?

 

[berkeman]

I have a project which would benefit from CPUs that use a very high amount of wattage.

Can you say more about this? TBH, it makes no sense on the face of it.

Of course, this is the opposite of what the goal usually is for CPUs, but for my project I am looking for a few main things: high wattage in a relatively small package, relatively low cost, and then as much processing power as I can get with the combination of those two things.

The power dissipation of a CPU IC or Module is the result of the power consumption of the IC(s), not something that is independent of the circuit design and package design. The power dissipation comes from a combination of all of the switching currents and parasitic leakage currents multiplied by the appropriate supply voltages.

So, theoretically, I'm not sure whether this makes sense or not. As an example, I'll use the i9 processor's, which has a max power consumption of 241W. If I could make a processor where the max power consumption was double (482W), and the processing power of that CPU was 1.5x a single i9 processor, but the size of the chip and motherboard housing was similar, then that would be a success. Especially if the cost stayed low, which I figure might be possible as the usual emphasis on power efficiency wouldn't be a factor.

No, sorry, so far it is not making sense to me. Are you planning on designing this million transistor new processor?

 

[phinds]

Are you planning on designing this million transistor new processor?

I think you meant billion. The i9 has about 4 billion.

 

[Arqane]

Can you say more about this? TBH, it makes no sense on the face of it.

I'm working on an application where I actually want heat output. I'd be using the heat output from the processors, as well as the processing power itself. I could use the most power-hungry CPUs currently available, but I also am looking at producing a few kW of heat in a fairly small space, and linking 10-20 i9s together with the requisite boards may be a bit too large (might be fine if I stacked them in a 3d setup rather than on the same plane, but that would likely cause its own problem).

The power dissipation of a CPU IC or Module is the result of the power consumption of the IC(s), not something that is independent of the circuit design and package design. The power dissipation comes from a combination of all of the switching currents and parasitic leakage currents multiplied by the appropriate supply voltages.

My concern with this part was just any internal problems with the CPU itself. With enough surface area, it would probably be fine, I'd assume. It would use liquid cooling, so my concern was just whether any section of a processor built that produced more wattage would overheat before it got to any medium to radiate the heat away.

No, sorry, so far it is not making sense to me. Are you planning on designing this million transistor new processor?

Not really, it's a theoretical question that could have potential real-world application. Usually the heat is a really big problem, and I figured that heat concerns are one limitation for most applications that CPUs are used in. But if you're using the heat waste, it could be a reason to build a separate type of CPU. I wouldn't expect it to be put into production unless it did work for an application where it would be mass-produced. And at that point, it would probably be best as a modified version of the processors already being made by the big manufacturers. Just with much less stringent restrictions on their heat output.

 

[berkeman]

I'm working on an application where I actually want heat output. I'd be using the heat output from the processors, as well as the processing power itself.

This is extremely misguided, IMO. If you want heat, just use nichrome wire. If you want processing, use efficient processors. You can package them semi-nearby each other, but heating up processors unnecessarily is very counterproductive and expensive.

 

[Arqane]

This is extremely misguided, IMO. You want heat, just use nichrome wire. You want processing, use efficient processors. You can package them semi-nearby each other, but heating up processors unnecessarily is very counterproductive and expensive.

I understand why you'd say that, but it's not much different than closing a door in a small room with a computer, so that you don't have to use your home's heater at all if you're the only one in the house. There are certain times where you can use the fact that a watt used for some other purpose also heats the area around by the same watt, and use it for something quite beneficial (EDIT: instead of using double the energy for the same thing).

 

[phinds]

This is extremely misguided, IMO. You want heat, just use nichrome wire. You want processing, use efficient processors. You can package them semi-nearby each other, but heating up processors unnecessarily is very counterproductive and expensive.

 

[nsaspook]

I understand why you'd say that, but it's not much different than closing a door in a small room with a computer, so that you don't have to use your home's heater at all if you're the only one in the house. There are certain times where you can use the fact that a watt used for some other purpose also heats the area around by the same watt, and use it for something quite beneficial (EDIT: instead of using double the energy for the same thing).

No, it's very different. Heat is a harmful byproduct of inefficient solid-state electronics that can harm not only the processor die but affect the life span of most major components in the device.

It's like saying you want a heater in an ice-cream freezer so the freezer compressor heat can warm the room because it's running all the time.

 

[Arqane]

No, it's very different. Heat is a harmful byproduct of inefficient solid-state electronics that can harm not only the processor die but affect the life span of most major components in the device.

It's like saying you want a heater in an ice-cream freezer so the freezer compressor heat can warm the room because it's running all the time.

No, you're taking it one step too far. It's simply saying that because you're already using your freezer for something else, that you'll take advantage of the fact that it makes heat. Putting a heater in the freezer is ridiculous, inefficient, and isn't something that you already normally do.

Taking advantage of the fact that people already use computers, or taking advantage of the fact that people are already using electricity for heating, is just utilizing an already-spent resource (think regenerative braking). That's quite different than trying to set up a perpetual motion machine where every step incurs losses.

 

[Averagesupernova]

So are you saying it is better to design an inefficient processor, heat a room with it and NOT use nichrome for heating rather than design an efficient processor that has identical computational performance and use nichrome? This is nuts....

 

[Arqane]

So are you saying it is better to design an inefficient processor and heat a room with it rather than design an efficient processor that has identical computational performance and NOT use nichrome for heating? This is nuts....

Nah, it's actually doing it backwards. It's saying that we're already wasting electricity burning it off to make heat, like with nichrome, and instead getting processing power out of that wasted electricity and still getting the exact same amount of heat.

 

[Averagesupernova]

I don't even know where to start here....

 

[Baluncore]

I'm working on an application where I actually want heat output. I'd be using the heat output from the processors, as well as the processing power itself.

To get the heat out will require a temperature difference.
What temperature difference do you require to reuse the heat energy.
Is that temperature less than can be tolerated by the CPUs?

 

[Arqane]

No, it's very different. Heat is a harmful byproduct of inefficient solid-state electronics that can harm not only the processor die but affect the life span of most major components in the device.

If the heat is not dispersed, absolutely. But if you had a 240W processor in a computer that was dispersing heat very well, it's going to be less damaged by heat than a 120W processor in a computer with horrible dispersion. In the same vein, no matter what maximum wattage the processor uses, if you have a computer that is dispersing heat quite well and you warm the room it is in up by 10F/5C, it will have a pretty negligible effect on the components in the computer. It's a little taxing, but not much.

When I was asking about a CPU with more wattage, it's under the assumption (as I somewhat mentioned), that the thermal design power/dispersion would be able to handle it. At least any components around the CPU. I do know that internally it could eventually cause a problem, and that was the crux of the question. But cooling systems are already dealing with much higher wattage than they used to, and as this would be built around a liquid cooling system, it should fare about as well as possible at dispersing the heat from the CPU very quickly.

 

[Arqane]

To get the heat out will require a temperature difference.
What temperature difference do you require to reuse the heat energy.
Is that temperature less than can be tolerated by the CPUs?

The liquid would max out/shut down at about 150F/65C range. It is not a great temperature difference from safe conditions of a CPU at that maximum, but the flow rate over the CPU would be high and would be replenished with 80-100F/27-37C liquid most of the time.

 

[russ_watters]

Yeah, I don't get it either. The math isn't difficult here. You execute a billion instructions a second and you dissipate 500W. Ok. It doesn't matter if 250W are dissipated by the processor and 250W by an electric heater or if all 500W are dissipated by the processor. The result is exactly the same.

Can you explain, specifically and in similar numerical terms what you are trying to do that wouldn't be equivalent to what @berkeman suggested? How would this not meet the requirements you specified in the OP? All I'm seeing here is vague handwaving about wanting the heat from the processor. Why?

But whatever - have you googled this question? How does an i486 compare with an i9 in terms of efficiency, for example...?

There are certain times where you can use the fact that a watt used for some other purpose also heats the area around by the same watt, and use it for something quite beneficial

Such as?

 

[Averagesupernova]

I'll choose my efficient processor and burn the nichrome. Then when the weather warms up, you can run your air conditioner and I'll just turn the nichrome off.

 

[Arqane]

Yeah, I don't get it either. The math isn't difficult here. You execute a billion instructions a second and you dissipate 500W. Ok. It doesn't matter if 250W are dissipated by the processor and 250W by an electric heater or if all 500W are dissipated by the processor. The result is exactly the same.

Can you explain, specifically and in similar numerical terms what you are trying to do that wouldn't be equivalent to what @berkeman suggested? How would this not meet the requirements you specified in the OP? All I'm seeing here is vague handwaving about wanting the heat from the processor. Why?

But whatever - have you googled this question? How does an i486 compare with an i9 in terms of efficiency, for example...?

With the i9 I mentioned, if you want to output the same heat as an average space heater (1.5kW), you could run about 6 processors at max capacity (241W each) to get that heat output. If you had (currently imaginary) 750W processors, you'd only need two to produce the same heat, and as long as all the processors I just mentioned distribute the heat well, they shouldn't have heating issues. It looks like the 486s were around 5W, so you'd need 300 of them running to make the same heat. But the sizes of those three options would generally be vastly different in order to perform any useful computing along with creating the heat you want. Since I'm looking for a compact solution, I was wondering about the viability of a processor using a greater amount of watts in about the same size as most processors run.

They have dual CPU motherboards, which is getting at the idea I was talking about. But as they mention with the dual motherboards, they tend to be less efficient than motherboards with a single, more powerful CPU. So I'm just trying to find out the viability of something in the 750W range for a processor, as it would probably satisfy the size requirements, heat output and processing power that I'm looking for.

 

[russ_watters]

But the sizes of those three options would generally be vastly different in order to perform any useful computing along with creating the heat you want. Since I'm looking for a compact solution, I was wondering about the viability of a processor using a greater amount of watts in about the same size as most processors run.

How "compact" and why does "compact" matter?

They have dual CPU motherboards, which is getting at the idea I was talking about. But as they mention with the dual motherboards, they tend to be less efficient than motherboards with a single, more powerful CPU. So I'm just trying to find out the viability of something in the 750W range for a processor, as it would probably satisfy the size requirements, heat output and processing power that I'm looking for.

Well this is great! A nichrome wire is much smaller than a motherboard. You can do what you want with a processor and a nichrome wire and it would be more compact than multiple processors on multiple motherboards. Have I solved your problem?

 

[Arqane]

How "compact" and why does "compact" matter?

Because ultimately I'm looking for closer to 5kW of heat in somewhere around 1 cubic foot of space. And fitting 20 of the current highest power CPUs in that space is difficult.

Well this is great! A nichrome wire is much smaller than a motherboard. You can do what you want with a processor and a nichrome wire and it would be more compact than multiple processors on multiple motherboards. Have I solved your problem?

It would work for the heat portion, but every watt from the nichrome is not being used for any computing. The project is both for energy and cost savings. Using each watt for both computing and heat is much more efficient than just using it for heating alone, or for computing alone (where you usually want to spend extra energy cooling it).

 

[russ_watters]

Because ultimately I'm looking for closer to 5kW of heat in somewhere around 1 cubic foot of space. And fitting 20 of the current highest power CPUs in that space is difficult.

K....that's why a heater will provide what you say you want.

It would work for the heat portion, but every watt from the nichrome is not being used for any computing.

So what? The two parameters are independent of each other. You've specified the problem that way. You want X processing and Y heat. Any combination of processors and heaters that adds up to the desired values will give you what you say you want. Why is is so important to you that the processors provide the heat?

The project is both for energy and cost savings. Using each watt for both computing and heat is much more efficient than just using it for heating alone, or for computing alone (where you usually want to spend extra energy cooling it).

No it isn't. The math here is straightforward, as I showed in the prior post. But please: show us your math. What are you doing with the processing? What are you doing with the heat? What are the instructions per watt?
[edit]
Your OP talked about 1.5x the processing power of an i9 and 2x the electrical/thermal power. That's a fairly small difference. You could under-clock two i9's by 25% and add a small heater to make up the difference and get exactly what you said you want. What's the problem with that?

 

[Borek]

This is extremely misguided, IMO. You want heat, just use nichrome wire. You want processing, use efficient processors. You can package them semi-nearby each other, but heating up processors unnecessarily is very counterproductive and expensive.

Mining bitcoins and reusing heat output for house heating is a win/win.

Not that I plan to use it, and still, I would prefer power efficient processors for that.

 

[russ_watters]

Mining bitcoins and reusing heat output for house heating is a win/win.

1. If you can actually use the heat, which is difficult because it is low grade heat. I don't do a lot of data centers, but I haven't seen one that does heat recovery. Perhaps OP is looking into a way, but testing that wouldn't seem to require a contrived/downgraded efficiency. Depending on the method I'd still be inclined to test it with an electric heater first to demonstrate that it works before potentially risking computer hardware on it(for example if it is water cooled). Heat recovery can use low grade heat though, and if it's insufficient you just suppliment/boost it.

2. Op thinks the processors he's chosing from are too efficient, which means less of a "win" on the processing side after you downgrade them.

 

[Borek]

1. If you can actually use the heat, which is difficult because it is low grade heat.

Several years ago I have read about a guy in PL who attempted to do that, as you can guess it sparked nonsensical comments from people who slept on physics lessons in school. No idea how well it worked.

Depending on the method I'd still be inclined to test it with an electric heater first to demonstrate that it works before potentially risking computer hardware on it(for example if it is water cooled).

There are commercially available liquid coolers for high power/gaming computers, so it is probably not that risky (or at least you don't have to reinvent the wheel, you can buy of the shelf components)

But in general I agree with what you are saying, all I am trying to add is that the idea of reusing heat from mining adds another perspective to the discussion.

 

[Arqane]

Yes, the original idea came from bit mining, though that isn't exactly the end goal. But it was thought of after a guy burned off his extra generation from his solar panels with bit mining (generally more lucrative than selling it back at wholesale rates). And it already has been done using an electric heater of about 5kW. That is why I was aiming for that goal. And the main reason I was looking for inefficiency was the size issue, though it was also a potential question about the cost, as less efficient processors *might* be able to be made with a cheaper process. For this purpose, I could lose some efficiency if it fit the size requirements, and especially if it had an effect on the cost.

 

[nsaspook]

Today, intentionally less efficient processors will cost more (due to market factors limiting production) because nobody on the planet designs that way because the raw production costs are the same for the most efficient processor in the same FAB on the same process line.

 

[Arqane]

Today, intentionally less efficient processors will cost more (due to market factors limiting production) because nobody on the planet designs that way because the raw production costs are the same for the most efficient processor in the same FAB on the same process line.

Yes, I counted on that, though I was still curious about the viability of the idea in the first place. If it was promising, then it's an application that could use about 1 million of these printed pretty quickly. With that scale, it makes more sense, especially if the design was mostly a copy of the current one, just essentially doubled or tripled up. If a change of materials in the same basic casting process would make it cheaper, though less efficient then it would make sense. As I said, twice the power output for 1.5x the computational power would definitely be fine if the cost was lower. But that seems less likely as the power output was directly related to the processes performed in general. That was useful information.

 

[berkeman]

Using each watt for both computing and heat is much more efficient than just using it for heating alone, or for computing alone (where you usually want to spend extra energy cooling it).

Sorry if I missed it, but what are you going to use this low grade heat for? You are not going to generate electricity from it. You just want to heat your soup?

If you can actually use the heat, which is difficult because it is low grade heat.

 

[nsaspook]

There are lots things we could do if we wanted things to be intentionally less efficient for some reason.

 

[Arqane]

Sorry if I missed it, but what are you going to use this low grade heat for? You are not going to generate electricity from it. You just want to heat your soup?

I want to heat a water heater, which are generally heated with 4.5kW heating elements

 

[Arqane]

I guess I will just explain the whole concept instead of dodging it, as it should help explain things much more clearly. I am looking to build a system to put processors into water heaters and use the output heat to do the job a normal heating element would. Water heaters use about 20% of the total electricity usage per month in homes. I am putting together a sustainable apartment business, and even with a single 400 unit apartment complex, that is a *lot* of computational power coming from 8,000 processors using the electricity usually just burned off every month.

Why do I want "inefficient" processors? Because nobody wants their water heater heating up too slowly. 4.5kW is the goal, but with current processors that's hard to fit in a 40-gallon tank effectively. Some people use lower power heating elements, but that would be a pretty big negative for most people.

Why is cost a factor? Because the initial cost of this system is by far the largest part (along with some maintenance). Other than that initial cost, it's basically a data center with $0 cost for electricity and $0 cost for cooling. The electricity is already being used for the water heater's original purpose and the water cycling through takes care of the cooling.

 

[berkeman]

Why do I want "inefficient" processors?

I'm skeptical about your overall concept, but one tip would be to think of using extra background processes on the CPUs (that aren't needed) to increase the power consumption, instead of thinking of finding "inefficient" processors. You can run them at full speed when it's not needed and execute loops that use all of the math processor, etc. when it's not really needed.

 

[russ_watters]

The overall concept is pretty sound (not without complications though), but the idea of using inefficient processors on purpose still isn't.

...but one tip would be to think of using extra background processes on the CPUs (that aren't needed) to increase the power consumption, instead of thinking of finding "inefficient" processors. You can run them at full speed when it's not needed and execute loops that use all of the math processor, etc. when it's not really needed.

Yes, I see no fundamental problems with that. From an economic standpoint though, the duty cycle of a water heater is terrible for this. You'd need to install your 5kW of processors and run them at an average of probably 200W or so. It's a lot of money to spend on processors that most of the time aren't running.

The OP gets 2/3 of the equation right: you want high wattage and cheap. But there's no need to specify low efficiency. You just take what you get for where cost and wattage intersect.

 

[Arqane]

The overall concept is pretty sound (not without complications though), but the idea of using inefficient processors on purpose still isn't.

Well the thought process is that I simply wanted higher power processors in the same space. 20 i9s, or especially 30+ i5s or below (to make it cheap) just doesn't fit in the space. They don't have to be inefficient. But at the same time, if I could find a cheap generic company that skipped corners and used substandard materials with a bit more resistance, but could still pull off a decent amount of efficiency compared to current processors, then that would be fine for this purpose.

I mean, it would be nice if these were the same size and just as replaceable as heating elements are in water heaters currently. But I was looking if a middle ground was possible: a bunch of cheap processors put in there as the heating element that could produce enough heat to work just as well as a water heater. It doesn't make much sense as a one-off for personal use in your own home. But when you have thousands of apartment units being powered by solar and you could network them together and use it for applications like bit mining or running physics/weather calculations, it makes a lot more sense. In fact, at the current rate that data is being rented out, it would actually have a very sizeable impact on revenue even compared to rent itself. Besides also being very environmentally friendly.

 

[russ_watters]

Well the thought process is that I simply wanted higher power processors in the same space. 20 i9s, or especially 30+ i5s or below (to make it cheap) just doesn't fit in the space. They don't have to be inefficient.

Yep. Looking back at your OP, most of what you described makes sense, it's just the scenario with the i9's that went sideways.