Does wattage have a limit?

[Protomesh]

So the question is, can I keep pushing more and more power through my CPU if I have enough thermal and power headroom? 

 

Will my CPU spontaneously combust if I push 500W through it even if it's operating within safe temperatures?

 

I guess the issues lie in voltage. Could you raise wattage without raising voltage? I know you can remove the motherboard limiter, but it won't make the CPU use more power alone.

 

Could I just keep raising the clockspeeds without raising voltage? Or is that just not possible?

 

If I can just keep increasing CPU wattage, will I eventually hit a limit because my motherboard won't/can't allow more? Power rail limit?

[emosun]

firstly , what task are you trying to do that has you questioning the physical limits of a cpu

[DreamCat04]

9 minutes ago, Protomesh said:

Will my CPU spontaneously combust if I push 500W through it even if it's operating within safe temperatures?

 

No. LTT made a video where they overclocked a 13900KS as far as they could on their 5000W chiller and they got to the limit of the silicon at around 480W. Past that, they didn't get more performance or clock speed. They could have pushed it further by cooling it with liquid nitrogen, but that would actually have decreased the power draw of that chip due to conductors lowering their resistance at lower temperatures

11 minutes ago, Protomesh said:

Could I just keep raising the clockspeeds without raising voltage? Or is that just not possible?

Not possible. There is a certain limit that's different for each CPU of how high you can clock it at a certain voltage

 

13 minutes ago, Protomesh said:

If I can just keep increasing CPU wattage, will I eventually hit a limit because my motherboard won't/can't allow more? Power rail limit?

That depends on the board, but most set their current- (and therfore power-) limit way above anything a CPU will ever draw. Eventually you'll reach the current limit of your power supply depending on its rated wattage and it shuts off if you exceed it

[Protomesh]

11 minutes ago, emosun said:

firstly , what task are you trying to do that has you questioning the physical limits of a cpu

Trying to maximize the performance of my CPU for an old unoptimized game.

 

Binning the 3 best cores on 14900K and doing the highest specific-core overclock a on those 3, setting affinity so the game only runs on those cores and perhaps disabling most of the remaining cores if necessary except core 0 which is mainly utilized by the OS.

 

I guess I would be voltage limited first

I want more power on those specific cores without actually running into stability issues.

[Protomesh]

11 minutes ago, DreamCat04 said:

Eventually you'll reach the current limit of your power supply depending on its rated wattage and it shuts off if you exceed it

I mean I could technically dedicate a 2050W PSU for the CPU alone, another 2050W for the motherboard alone, and then a 3rd PSU for the rest of the system if I was insane enough

[Spotty]

13 minutes ago, Protomesh said:

Trying to maximize the performance of my CPU for an old unoptimized game.

If it's a software issue then trying to brute force it with faster CPU probably won't help that much. What game is it? What CPU do you have?

[Mojo-Jojo]

The motherboard traces, the substrate traces, pins and bonding wires will all have their limits. The thermal interface material will have some thermal resistance as well.

 

So yes, there are many limits.

[Protomesh]

Just now, Spotty said:

If it's a software issue then trying to brute force it with faster CPU probably won't help that much. What game is it? What CPU do you have?

It's more just for shits and giggles. I already get 712fps under ideal conditions. Far from stable. I can do 400 stable. I possibly want to upgrade to a 500Hz monitor so getting 500+ stable would be a goal. Also hitting that magic 1000 peak.

[Protomesh]

5 minutes ago, Mojo-Jojo said:

The motherboard traces, the substrate traces, pins and bonding wires will all have their limits. The thermal interface material will have some thermal resistance as well.

 

So yes, there are many limits.

Direct die cooling with the Go Chiller graphene coolant is always a step closer but yeah. Thermal headroom is not really the issue especially when liquid cooling the motherboard and RAM is a possibility as well. 

 

It's Intel's and AMD's focus on higher core counts rather than faster individual cores that's kind of the problem. I understand why they do it though. Very few games actually benefit from 20 cores. But every game benefits from 20% faster single core/thread performance. Sadly we really only get 3-5% each generation.

[manikyath]

just a detail i want to add; wattage is a result. you cant 'push more wattage', more wattage comes from either the cpu pulling more current or the VRM supplying higher voltage.

[Wh0_Am_1]

On 1/17/2024 at 3:55 AM, Protomesh said:

I mean I could technically dedicate a 2050W PSU for the CPU alone, another 2050W for the motherboard alone, and then a 3rd PSU for the rest of the system if I was insane enough

Okay to start you need to understand these two equations (just in case you don't) watts = volts x amps, and volts = resistance x amps. To understand these terms I will explain them in the way of water, as electricity is the flow of electrons (-) to protons (+) and the principles are generally the same (generally.... though there are notable differences): Watts is power or total flow water coming out of pipe, volts is pressure (how hard it wants to move) , amps is current (how wide the pipe is), and resistance is the friction slowing the water down or the waste. 

 

The more you increase the amperage the more the voltage is going to go up, the more voltage you push the more degradation you are going to get (the smaller the process the less voltage you can push through the processor before you degrade it, smaller process = smaller pipes, don't go over pressurizing your pipes.... it doesn't end well... other wise with that pressure, just like with a broken pipe the electricity may decide to overcome the silicon, or induce multiple other failure modes)

*(to be probs fixed) the more you increase the wattage the more heat you are going to produce from your resistance, and the more heat you generate from the resistance the more resistance you are going to have. 

You can only cool so much, as after a certain point it becomes very difficult to move any more energy away from something.

 

 

...I am going to stop my response here (This response is incomplete.... BUT I am publishing it ANYWAY! XD) and probably come finish it as I probs should head to bed after a long day work... and my train of thought derailed....

[porina]

Sounds like overclocking. The answers to these questions could be answered by looking at extreme overclocking but to get much more than air will be extremely difficult and expensive. 

 

With old school overclocking, there is a well known relationship. If you want more clock, you need more voltage. Power scales with voltage, actually voltage squared for a resistive load. I'm not sure how semiconductors fall into that. More power = more heat to get away from CPU. So you then throw better cooling at it. Eventually you will reach a physical limit where you can't cool it even if you could pump in more power.

 

In my personal experience, going from air to high end water gets you maybe 100 MHz, and going to chilled water maybe another 100 MHz more. You'd have to go phase change and well below freezing to go much further. Just not worth it outside of one off showcases, where liquid nitrogen is used.

[Protomesh]

12 minutes ago, porina said:

Sounds like overclocking. The answers to these questions could be answered by looking at extreme overclocking but to get much more than air will be extremely difficult and expensive. 

 

With old school overclocking, there is a well known relationship. If you want more clock, you need more voltage. Power scales with voltage, actually voltage squared for a resistive load. I'm not sure how semiconductors fall into that. More power = more heat to get away from CPU. So you then throw better cooling at it. Eventually you will reach a physical limit where you can't cool it even if you could pump in more power.

 

In my personal experience, going from air to high end water gets you maybe 100 MHz, and going to chilled water maybe another 100 MHz more. You'd have to go phase change and well below freezing to go much further. Just not worth it outside of one off showcases, where liquid nitrogen is used.

100Mhz doesn't really make sense to me. Direct die copper plate with graphene coolant with 4x D5 pumps retaining high flow rate through like 10x 83mm copper radiators with two large reservoirs with highest static pressure fans I can find, possibly industrial, on push/pull max RPM. 

 

All that compared to air. Not direct die. Possibly even inside a case with worse fans with less than 1/30th of heatsink area. 100Mhz? Cap.

[porina]

34 minutes ago, Protomesh said:

100Mhz doesn't really make sense to me. Direct die copper plate with graphene coolant with 4x D5 pumps retaining high flow rate through like 10x 83mm copper radiators with two large reservoirs with highest static pressure fans I can find, possibly industrial, on push/pull max RPM. 

 

All that compared to air. Not direct die. Possibly even inside a case with worse fans with less than 1/30th of heatsink area. 100Mhz? Cap.

100 MHz is an illustration based on my last silly OC attempt for fun. I don't recall if this was on an Intel 7800X or 7920X so a bit older CPU. Maybe newer CPUs can give a little more or even less but I don't have experience of them. Recent CPUs operate closer to their limits at stock than older CPUs ever did.

 

The curve above is well known to experienced manual overclockers. As you try to push progressively more power through the cores, you get less and less gain to clock. Hate to tell you this but the difference between high end water and high end air isn't that big when it comes to overclocking results. Do not dismiss air. A good air cooler is still very decent. While water cooling can give you a much bigger cooling surface, the limiting factor remains transfer from the CPU die itself. You might gain a few degrees under load but it doesn't translate to a lot more performance.

 

I love the look of condensation lit by RGB. This photo was taken during a run with chilled water, just above freezing. I ran it that way so that condensation would fall off. Like I said, this got me about +200 MHz over air cooling.