Question Has much research been done into the real-world effectiveness of ADL E-cores?

[mikeymikec]

I was just reading about the Ryzen 7000 series and noticed the increased power usage for the 9-series, and it struck me how the industry seems to be pushing performance at the expense of energy usage and waste heat energy, which made me think of a flip-side question being Intel's ADL e-cores and whether they're actually useful, e.g. assuming that it's possible for say basic office apps, web browsers and Windows maintenance procedures to use E-cores exclusively then does it result in reasonably decent performance with a much reduced energy footprint? If that were the case then the P cores could just be used for 'big gun' processes like gaming and high-end productivity.

For me personally, my i5-4690k is absolutely fine for everything I do minus Handbrake, compression and gaming. If the E cores could be leveraged in the way I describe while bringing down total system power usage to a fraction of what my system currently uses, that would certainly be an interesting proposition to me.

 

[VirtualLarry]

My understanding is, that there is a significant limitation on the e-cores, that they share a voltage plane with the P-cores, and thus, don't run at their optimal fmax/voltage curve sweet-spot, in fact, the P-core may actually be more efficient, they just take up 4x the die area as e-cores. So the e-cores seem more marketing-related, a way to boost advertised core-count, without having to spend a commensurate cost on actual silicon. Eg. They're actually NOT "more efficient" in terms of operation. Only in die-size area. I mean, they're basically Skylake-equvalent, so it's not like their horribly inefficient or slow or anything, but they're not as good as a P-core.

 

[TheELF]

For me personally, my i5-4690k is absolutely fine for everything I do minus Handbrake, compression and gaming.

Look at your clocks/Vcore/watt on your CPU while doing easy stuff...
CPUs downclock like crazy and use next to nothing on easy tasks, and that doesn't change by having a huge maximum unless you set up a huge locked overclock and don't allow it to downclock ever.

P-cores will do most things using less energy than e-cores because they will downclock and still have more IPC and finish the task faster.

 

[Kocicak]

They are more efficient because they run at lower frequencies than P cores. And also having smaller transistor count for smaller tasks means less energy wasted. I have been watching them (in a 12600K) a bit and they seem to do something sometimes - this alone increases energy efficiency of the procesor.

But when I run antivirus scan, P cores do it, I have no idea why.

 

[TheELF]

My understanding is, that there is a significant limitation on the e-cores, that they share a voltage plane with the P-cores, and thus, don't run at their optimal fmax/voltage curve sweet-spot, in fact, the P-core may actually be more efficient, they just take up 4x the die area as e-cores.

That is probably true if they run individually, but in productivity if you already run 8 cores at full power the 9th and 10th P-core would get so little power that they would run way below their efficiency while e-cores should be specifically chosen to have that amount of power as their best efficiency point.

 

[zir_blazer]

I have benchmarked Cinebench R23 for a totally different purpose but the data points can be useful here.

-----------------------------------------

All had consistent 6 P-Cores 3.7 GHz, 4 E-Cores 2.8 GHz and Cache 3.3 GHz clock speeds, difference being usually Voltage and power consumption.

MSI BIOS 1.70 --- AC_LL/DC_LL 170 - Turbo No

12 Threads P-Cores only --- MT 11446 @ Package TDP 68W / ST 1440 @ Package TDP 16W
4 Threads E-Cores only --- MT 2758 @ Package TDP 18W / ST 768 @ Package TDP 10W

1 P Core (2 Threads) = 1907 MT points @ Package TDP 11,33W @ 168,3 MT points per Watt
1 E Core (1 Thread) = 689,5 MT points @ Package TDP 4,5W @ 153,22 MT points per Watt

Note than Core TDP could be a better metric than Package TDP, which would improve the E Core MT per Watt score. It should be on the Intel XTU logs, but too lazy to check.
Yes, the E Cores seems actually LESS power efficient than a P Core @ 2 Threads, but they have better density and can get around 45% more MT performance on the same area.
Note than clock speeds were fixed, and that there is no voltage/frequency curve optimizations of any kind.

 

[mikeymikec]

They are more efficient because they run at lower frequencies than P cores. And also having smaller transistor count for smaller tasks means less energy wasted. I have been watching them (in a 12600K) a bit and they seem to do something sometimes - this alone increases energy efficiency of the procesor.

But when I run antivirus scan, P cores do it, I have no idea why.

As a default setting I understand that - an automated scan ought to be having a background level priority, but if the user requests it then it supposedly must be for a good reason, so it's higher priority.

 

[dullard]

Hey Zir_blazer can you explain please how voltage affects power consumption? My professor says that the higher the voltage, the higher the power consumption for a pure resistive load, but that doesn't explain much to me.

There are several issues to consider with CPU power. There is dynamic power (this is the power that you were mentioning where the higher the voltage, the higher the power consumption). But there are also short circuit power and leakage power (no CPU is made with perfectly insulating materials, so there is always some power loss). https://en.wikipedia.org/wiki/Processor_power_dissipation#Sources

For a well-built CPU (CPUs with pretty good insulation, etc), the CPU power is dominated by the dynamic power term.

• Pdynamic = C * V^2 * f where C is the capacitance, V is the voltage, and f is the frequency of the CPU.

That equation above is similar to a resistive load where Power = V^2 / R.

However, an important factor is that in general the frequency of the CPU can be set higher with higher voltage. The frequency to voltage formula is non-linear at the extremes (either high voltage or low voltage). But most CPUs are operated roughly in the linear region where f is proportional to V. Thus, in many cases:

• Pdynamic ∝ V^3

You could bump up the voltage without bumping up the frequency, but that is just a waste of power. You could bump up the frequency without changing the voltage, but if you do it too much you will crash the CPU. Thus, there is some wiggle room in the power proportional to voltage cubed formula. A highly binned CPU will do better than a poorly binned CPU (They can be tested for the lowest power consumption and are sometimes sold that way). But, in general, the CPU power is proportional to voltage cubed. The exceptions are when you really push the voltage to either extreme.

 

[zir_blazer]

Hey Zir_blazer can you explain please how voltage affects power consumption? My professor says that the higher the voltage, the higher the power consumption for a pure resistive load, but that doesn't explain much to me.

I'm not the guy that you want to ask about purely theorical questions - I know than higher Voltage means higher power consumption all other things being the same, but I don't care why. I suggest you to read these two Threads by legendary Idontcare, and checking certain graphs that he posted (Albeit several are broken), specially this one, which will be more fun.

Power-Consumption Scaling with Clockspeed and Vcc for the i7-2600K

Since my asus maximus iv extreme-z has this nice convenient bank of voltage monitoring points (ASUS calls it "ProbeIt"), I thought I would have some geeky "sciency" fun armed with my i7-2600K, a volt-meter, a kill-a-watt power meter, and a garden variety rudimentary understanding of the...

forums.anandtech.com

Effect of Temperature on Power-Consumption with the i7-2600K

Building on a previous thread that was directed more towards power-consumption scaling with CPU clockspeed, of the many things uncovered in that thread the one thing that just never sat right with me, or with many other members, was the impact of the operating temperature on power-consumption...

forums.anandtech.com

 

[IntelUser2000]

Hey Zir_blazer can you explain please how voltage affects power consumption? My professor says that the higher the voltage, the higher the power consumption for a pure resistive load, but that doesn't explain much to me.

Recall that in a resistor, when you increase the voltage the current flowing through it increases as determined by the formula V=IR. Resistance or R is constant so when you increase voltage the current increases.

Power formula is P=IV, or when you rejig the terms you get I squared R(P=I2R). So when the current increases the power increases greatly.