What is this chip measurement called?

[IsraelPainter47]

Forgive me if this question is just wrong, I'm just curious and was hoping someone could help explain.
With the performance of so many devices (smartphones and tablets) largely being dictated/bottlenecked by its thermal capacity (the larger the device, the more heat it can handle), manufacturers want to go with a chip that has the best "performance per heat-output" ratio, so they can get the most performance out of their device design. This measurement is the true measurement of performance for these chips, but what is it actually called? (I highly doubt it's called "performance per heat-output ratio" hahaha). Also, obviously the manufacturing node of the chip plays a big part in this measurement but doesn't the microarchitecture also play a big roll? Ideally chip companies would design with this measurement in mind because it's the true dictator of performance.
I wish there was an industry standard for this measurement and chip companies would advertise it when they announce new chips

In my opinion, it's just as important and relevant as performance per watt or performance per dollar

 

[Tatty_Two]

Many things that produce heat as a by product come with cooling and often therefore negate the heat.

 

[DeathtoGnomes]

I've never seen that ratio in any review for a device, is that something new?

 

[Vya Domus]

There isn't a metric for that because there isn't just one way of testing it.

What you are referring to is called sustained performance and you can't have a standard metric for it. Well, you can but it wouldn't tell you much because you would need to have a fixed type of load, cooling, and time over which you measure performance and all of these things can obviously vary.

 

[Frick]

Many things that produce heat as a by product come with cooling and often therefore negate the heat.

Yeah this.

Electronic components have specs about what kind of performance you can expect at various temperatures (... sort of), but for complicated chips it's sorta meaningless as cooling is intimately related to all electronics.

 

[Bill_Bright]

manufacturers want to go with a chip that has the best "performance per heat-output" ratio, so they can get the most performance out of their device design.

No! Sorry. That is not how it works.

As yourself this, why is there heat in the first place? These devices are not ovens or space heaters. They process digital data. Why does a power supply produce heat? It is not heater. It just converts an AC voltage into DC voltages. Why does an incandescent light bulp produce heat? It is only supposed to produce light. Where is all that heat coming from?

The heat is being produced by wasted energy! Because humans are - thus far - incapable of making the perfect, zero resistance device, there is always some resistance causing energy to be loss in the form of heat. This resistance is due to friction. As electrons flow through a conductor, even the best super-conductors, billions of electrons are banging into each other and to the walls of and impurities within the conductors. That creates friction, which in turn, creates heat. And it wastes energy.

So it is NOT about performance per heat output! Manufacturers are striving to create a device that has the best efficiency. That is, the best "input per output" ratio. Ideally, that would be 1:1.

what is it actually called? (I highly doubt it's called "performance per heat-output ratio" hahaha).

It is called "efficiency". And it is generally specified as a specific value based on a specific load, or it is specified as an efficiency range over a variety of expected loads.

(the larger the device, the more heat it can handle)

Not always true either. In many cases, the larger the device, the more heat it generates and thus the greater the cooling requirements.

 

[MrGenius]

ASIC Quality Percentage. Though nobody seems to know exactly what that means. Or, rather, can agree on what it means. It's supposed to mean an IC:

1. Has more or less leakage current.
2. Requires higher or lower operating voltage(s).
3. Has hotter or colder operating temperatures as a result of 1 and 2.
4. Is capable of operating relatively faster, or can only operate relatively slower, as a result of 1, 2, and 3.
5. Is more or less well suited for, and/or is more or less efficient at performing in, the specific application it's intended to be used for as a result 1, 2, 3, and 4.

A higher or lower ASIC Quality % is supposed to tell you something about a given ICs amount of leakage current, operating voltage(s), operating temperatures, potential operation speed(relative to ICs of the same design), and/or how well/not well suited for, or more/less efficient at performing in, a specific application it is. Which might, or might not, have some bearing on real world operation of a given IC in a specific application. More or less(or not at all). Depending on who you ask. And what their definition of ASIC Quality % pertains to. Does a higher quality % represent more or less leakage current, higher or lower operating voltage(s), colder or hotter operating temperatures, capability for relatively faster operation or limitation to relatively slower operation, being more or less well suited for/efficient at performing in a specific application? Is higher quality really higher quality ? I mean higher quality % is better right? Then why do some people say higher quality % represents more leakage current? More leakage current means less electrical efficiency doesn't it? Does it not also equate to higher power consumption? Is less electrical efficiency and higher power consumption a sign of higher quality? Those same people then say higher leakage current also requires lower operating voltage(s). Wait...what? What kind of sense does that make? Why do some higher quality % ICs run relatively hotter? Is running hotter a sign of higher quality? Why do some lower quality % ICs clock relatively higher, and/or run relatively cooler? Are lower clocks a sign of higher quality? Where's the logic behind all of this?

I totally don't get it. But it's supposed to mean something along those lines.

 

[Voluman]

If i understood right, what do you want to ask than i guess, this:

So it is NOT about performance per heat output! Manufacturers are striving to create a device that has the best efficiency.
It is called "efficiency". And it is generally specified as a specific value based on a specific load, or it is specified as an efficiency range over a variety of expected loads.

With MrGenius thought too.

You mentioned tablets, mobiles: their goal is usually doing more operations with using 1 unit of energy (executing operandus consumes a few picoJoule energies)
Its simply physics: work, energy, heat. With handhelds you have given space, given accus (no real development in consumer market), given user demand (thanks to marketing too, small (?), thin, stylish gadget. You dont have better accus, dont have more space (thin stuff), but want more power (given by architecture, manufacturing, various tricks). With proper cooling, you can maintain higher performance longer than average stuff.
So performance per watt is probably the best thing to describe this to get it easily.

 

[TheoneandonlyMrK]

Forgive me if this question is just wrong, I'm just curious and was hoping someone could help explain.
With the performance of so many devices (smartphones and tablets) largely being dictated/bottlenecked by its thermal capacity (the larger the device, the more heat it can handle), manufacturers want to go with a chip that has the best "performance per heat-output" ratio, so they can get the most performance out of their device design. This measurement is the true measurement of performance for these chips, but what is it actually called? (I highly doubt it's called "performance per heat-output ratio" hahaha). Also, obviously the manufacturing node of the chip plays a big part in this measurement but doesn't the microarchitecture also play a big roll? Ideally chip companies would design with this measurement in mind because it's the true dictator of performance.
I wish there was an industry standard for this measurement and chip companies would advertise it when they announce new chips

In my opinion, it's just as important and relevant as performance per watt or performance per dollar

This is encapsulated in performance per watt , since a chips ability to do work verses the power used is proportional to it's heat output, the top limits of which for any chip would indeed require more cooling.
Every chip has to be designed to operate within a voltage curve and frequency curve , very specifically , and is not made ,then tested to see what it'll do blind, it should do at least what it was designed to.
So in short manufacturers want the specs and efficiency they want as dictated by the cash they want to spend , bigger chips dissipate more heat easier but conversely as chips progress generationally they get considerably denser and smaller limiting any top performance potential.

 

[qubit]

@IsraelPainter47 I think thermal design power is the metric closest to what you're looking for. As the others have said, there are too many variables for one simple ratio.

@Bill_Bright you're almost right. In a superconductor, electrons do flow without any resistance at all and hence power transmission is 100% efficient. In fact, the power supply to a ring superconductor can be disconnected and the current will keep on flowing!

Superconductors have various other strange and interesting properties too, such as excluding all external magnetic fields.

 

[Bill_Bright]

since a chips ability to do work verses the power used is proportional to it's heat output,

No its not. There are other factors involved - including voltages applied, ambient temperatures, distance between point on the die and more.

In a superconductor, electrons do flow without any resistance at all and hence power transmission is 100% efficient.

But there is no practical application of this in the real world of computing. It takes a tremendous amount of energy to bring those "superconductors" down to temperatures at which zero resistance can be achieved. And the equipment to do that, in the mean time, is wasting a tremendous amount of energy in the form of heat.

And I note even the purest of normal conductors (copper, silver, and gold for example) cannot reach 0Ω resistance, even at absolute zero temps. Plus, human are still incapable of making a conductor with zero impurities.

 

[FordGT90Concept]

...performance per watt...

That's basically what you're asking for:

Perfᵢₙ / Wattsᵤₛₑd = Perfₒᵤₜ / Wattsₘₐₓ

Wattsᵤₛₑd cannot exceed Wattsₘₐₓ.

Solving for Perfₒᵤₜ (your benchmark is throttled so what would it be if it weren't throttled):
Perfᵢₙ * Wattsₘₐₓ / Wattsᵤₛₑd = Perfₒᵤₜ

Solving for Perfᵢₙ (your benchmark isn't throttled so what would be it if it were):
Perfₒᵤₜ * Wattsᵤₛₑd / Wattsₘₐₓ = Perfᵢₙ

For example, 65w processor in a thermal design that can only handle 30 watts, a benchmark that usually gets you 1000 points would only get you 461 points.

 

[qubit]

But there is no practical application of this in the real world of computing. It takes a tremendous amount of energy to bring those "superconductors" down to temperatures at which zero resistance can be achieved. And the equipment to do that, in the mean time, is wasting a tremendous amount of energy in the form of heat.

And I note even the purest of normal conductors (copper, silver, and gold for example) cannot reach 0Ω resistance, even at absolute zero temps. Plus, human are still incapable of making a conductor with zero impurities.

None of that is relevant to the point.

 

[TheoneandonlyMrK]

That's basically what you're asking for:
View attachment 105490
Perfᵢₙ / Wattsᵤₛₑd = Perfₒᵤₜ / Wattsₘₐₓ

Wattsᵤₛₑd cannot exceed Wattsₘₐₓ.

Solving for Perfₒᵤₜ (your benchmark is throttled so what would it be if it weren't throttled):
Perfᵢₙ * Wattsₘₐₓ / Wattsᵤₛₑd = Perfₒᵤₜ

Solving for Perfᵢₙ (your benchmark isn't throttled so what would be it if it were):
Perfₒᵤₜ * Wattsᵤₛₑd / Wattsₘₐₓ = Perfᵢₙ

For example, 65w processor in a thermal design that can only handle 30 watts, a benchmark that usually gets you 1000 points would only get you 461 points.

Whats this performance in you speak relative ( in terminology) to performance out.

So performance per watt is equal to performance observed divided by tdp yet

Performance per watt and heat output are not intrinsically tied? Hmmn

 

[Bill_Bright]

None of that is relevant to the point.

LOL

 

[Totally]

Forgive me if this question is just wrong, I'm just curious and was hoping someone could help explain.
With the performance of so many devices (smartphones and tablets) largely being dictated/bottlenecked by its thermal capacity (the larger the device, the more heat it can handle), manufacturers want to go with a chip that has the best "performance per heat-output" ratio, so they can get the most performance out of their device design. This measurement is the true measurement of performance for these chips, but what is it actually called? (I highly doubt it's called "performance per heat-output ratio" hahaha). Also, obviously the manufacturing node of the chip plays a big part in this measurement but doesn't the microarchitecture also play a big roll? Ideally chip companies would design with this measurement in mind because it's the true dictator of performance.
I wish there was an industry standard for this measurement and chip companies would advertise it when they announce new chips

In my opinion, it's just as important and relevant as performance per watt or performance per dollar

That's not a realistic or practical metric to go by because of the nature of electricity as performance increases heat also increases, in addition that increase in heat is not linear and may vary wildly from chip to chip. On it's own it pretty useless as a metric.