Some say that DDR5 consumes less power, but does that hold true in practice?

[lpwin7]

I posted this thread on reddit too (https://www.reddit.com/r/hardware/comments/111v30g/some_say_that_ddr5_consumes_less_power_but_does/ )

I wanted to hear the opinions from this forum as well, since many users here are highly knowledgeable about power consumption.

Apparently for the same given frequency (ex: DDR4 4000mhz vs DDR5 4000mhz) , DDR5 consumes less power as DDR5 Voltage is 1.1 while DDR4 voltage is 1.2 (XMP off) to 1.35 (XMP on). However new generations of ram have WORSE performance for the same given frequency, so basically DDR4 3600mhz = DDR5 5000mhz (or something close to that) performance-wise.

DDR5 ram runs hot. That could be because it takes more power (even though it's 1.1v default, it could take enough amperage to consume more power total than DDR4) or the heat could be just due to the fact that the power consumption is controlled on the ram itself in DDR5 rather than the motherboard. Here's a couple Debauer testing videos where he observes heat and voltage from 2021:

https://www.youtube.com/watch?v=fpVFeAxwtf0

YouTube

An dieser Stelle steht ein externer Inhalt von YouTube, der den Forumbeitrag ergänzt. Er kann mit einem Klick geladen und auch wieder ausgeblendet werden.

YouTube-Embeds laden

Ich bin damit einverstanden, dass YouTube-Embeds geladen werden. Dabei können personen­bezogene Daten an YouTube übermittelt werden. Mehr dazu in der Datenschutzerklärung.

YouTube-Embeds laden

Datenschutzerklärung

So my main question is- how likely do you think it is that the heat and/or higher frequency of DDR 5 (necessary to achieve the same performance of lower frequency DDR4) means that the DDR5 amperage is high enough cancel out the power consumption benefits of DDR5 running at a lower voltage ? For equivalent performance, which do you think takes more or less power, DDR4 or DDR5?

Thank you.

 

[gabbercopter]

to me personaly i do not have the feeling that DDR5 is the real deal at all. My expectation is that there will be something different released within the next 2-5 years wich is more performant then ddr5. because yet i dont have the feeling that the pricing , heat and power consumption realy comes up with the performance of ddr5

 

[Loopman]

War bei DDR4 (und 3 und 2) am Anfang auch so. Erst nach einiger Zeit konnten die neuen Techniken ihre Vorteile ausspielen.

 

[xxMuahdibxx]

Power = Volt * Ampere ...

Volt only says nothing.

 

[till69]

My DDR5 system takes little more power than DDR4 did before (same CPU&GPU). Both RAM at 1,35V.

 

[Qarrr³]

Eine ganze Industrie einigt sich auf einen Standard für Produkte, die von geringem Stromverbrauch profitieren. Armeen der besten Ingenieure sitzen am Design der Standards und Chips. Falls ddr5 schlechter wäre als ddr4 wäre das jemandem aufgefallen. Es gibt andere Herausforderungen als bei ddr4 und im Einzelfall kann ein Modul mehr verbrauchen. Gerade im Desktop, da dort nur die Performance zählt.

However new generations of ram have WORSE performance for the same given frequency, so basically DDR4 3600mhz = DDR5 5000mhz (or something close to that) performance-wise.

Bandbreite ist höher. Bei ddr5 stimmt die Aussage zum ersten Mal aber nicht Mal mehr für die reale Performance pauschal.

 

[Baal Netbeck]

I don't have a DDR5 System to test it.

It could well be, that DDR5 consumes more power when it is clocked high and timings are decent.

But it is hard to compare it with DDR4 1 on 1.
Do you look at bandwidth or latency?
Or when you compare performance in Apps, you can change the results by choosing different Apps.

Do you compare basic jedec modules that no one should use for gaming or XMP modules that vary very much?
Idle or in use?

I think DDR5 is a step in the right direction...we needed more bandwidth for dual channel systems and going from 1x64 Bit per channel to 2x32 Bit is a good idea when more and more CPU cores ask for data.

Right now we have worse latency and this is bad for most games but I'm hoping for some progress in terms of timings and price.

On the long run it will be the better technology....right now I'm still fine with DDR4.

 

[0ssi]

 

[lpwin7]

It was the same with DDR4 (and 3 and 2) in the beginning. It was only after some time that the new technologies were able to demonstrate their advantages.

Ok, but did DDR4 get more power efficient over time? Or are there reasons to expect DDR5 to become more power efficient over time?

 

[lpwin7]

An entire industry is agreeing on a standard for products that benefit from low power consumption. Armies of the best engineers design the standards and chips. If ddr5 was worse than ddr4, someone would have noticed. There are other challenges than with ddr4 and in individual cases a module can consume more. Especially on the desktop, because only performance counts there.

Bandwidth is higher. With ddr5, the statement is not true for the first time for the real performance.

"Worse" is subjective and nebulous terminology. DDR5 can be more powerful and also takes more energy, obviously. (Because of the power advantage nobody can really say it's generally "worse".) But my question is very very specific - it's a question of efficiency. The question is- for equivalent performance ddr4 3600 vs ddr5 5000 (or whatever frequency of ddr5 that you want to say is equivalent to ddr4 3600), how much power does each take? Because I'm wondering if DDR4 is more efficient. I think if companies had hard numbers to show DDR5 takes less power for roughly equivalent performance, they would definitely show consumers that in order to better market their product. They have not released any data to show that it takes less energy, which is what many articles trying to claim while pointing to .10 lower voltage, as if amperage isn't part of the equation.

My personal feelings that I do not have evidence for is that I think they do not want to show the truth, which is that DDR5 burns a lot of energy to get that power, when ddr4 is fine and will be fine for quite a few years. If ram companies wanted everyone to know the truth, they could pay to have another company make a motherboard that could use both ddr4 and ddr5 in order to compare them head to head. They just want to make money. I'm not saying ddr5 isn't a step in the right direction, I'm just saying it's unrefined and currently unnecessary. However if someone could show hard data that ddr5 takes even the same amount of power for equivalent performance vs ddr4, I would probably buy it for my new rig.

 

[lpwin7]

My DDR5 system takes little more power than DDR4 did before (same CPU&GPU). Both RAM at 1,35V.

Are the motherboards very similar? How much more wattage roughly does the DDR5 system take?

Thank you.

 

[Qarrr³]

Du träumst dir was zusammen. Ddr5 in der langsamsten Stufe ist normalerweise schneller als ddr4, sogar beim gaming. Latenz ist etwas langsamer, aber das ist eben nicht alles.
https://www.tomshardware.com/features/ddr5-vs-ddr4-is-it-time-to-upgrade-your-ram

Ddr5 ist im Normalfall auch effizienter, weil Spannungen gesenkt wurden, (weil der Regler auf den Modulen sitzt), die Leistung höher ist und größere Chips verbaut werden, d.h. weniger Chips bei der Spannung betrieben werden. Der Fokus lag aber nicht auf weniger Stromverbrauch, sondern höherer Leistung. Der geringere Stromverbrauch wird schlicht mitgenommen.
Der geringere Stromverbrauch wird nicht vermarktet, weil der Verbrauch des RAMs für Endkunden irrelevant ist. Lenovo erwähnt das sogar explizit:
https://lenovopress.lenovo.com/lp1618.pdf
Kannst auch bei micron von dem geringeren Verbrauch und höherer Performance lesen.
https://static.mr-wu.cn/doc/ddr5_more_than_a_generational_update_wp.pdf

 

[lpwin7]

How is 1.2v to 1.1v a 20% decrease? By my math that's an 8% decrease, so what am I missing?

 

[Baal Netbeck]

Power consumption is not linear with voltage.
It goes with V^2.

If I calculated it right, it should result in 16% less power....but this is just a rough estimate and it only works for the same chip...not in this case.

 

[xxMuahdibxx]

Power consumption is not linear with voltage.
It goes with V^2.

Realy... Without Ampere the Volt is nothing towards Power consumption.

 

[Baal Netbeck]

Ampere follows the votage.
If you lower the voltage at the same resistence, Ampere follows linear.

And in practise CPUs/GPUs and DRam under load react like a normal resistor.

Buildzoid made some videos about it and for DDR4 it holds true....not sure about DDR5 but I don't see why it would be different.

 

[Piktogramm]

New generations of memory are always more efficient, but efficiency quite often does not scale well with total energy consumption. If new technologies allowing a reduction in energy needed per bit transfered of 15% and the new tech also allows an increase in transfered data by 20%, the total power consumption is going to increase. Also, currently Sram does not scale well with ever decreasing node sizes. The gains in efficiency due to smaller nodes are way smaller than the gains one gets for logic circuits. One factor to this is, that the failure rate for Sram cells increases quite a lot with smaller nodes and newer DDR-Memory generations need to compensate for that by using on chip CRC error correction/checksumming. The logic needed for that eats quite a lot of the gains of smaller production nodes/new technology.

If you look at So-Dimm, driving high speed signals through sockets is a pain in the arse. It's quite possible, that current 1st/2nd gen DDR5-Modules are using big parts of their power budget just to drive the signals.


If you talk power, it's comlicated and memory does not behave as a static resistor at all[1].
Have a look at this little paper from micron in regards of DDR4 power consumption:
https://www.micron.com/-/media/clie...l-note/dram/tn4007_ddr4_power_calculation.pdf


[1]If one measures the memory modules including the voltage regulators incl. the smoothing caps with a slow multimeter, it may look like the behavior of an resistor, but no..

 

[Baal Netbeck]

And if you change the frequency, it changes the power consumption mostly linear....given that the voltage stays the same.

At least for CPUs/GPUs.

For the memory it might be less...buildzoid didn't test this very nice, because he left the subtimings on Auto and higher frequencys move less data per clock.

Here is the video:

YouTube

An dieser Stelle steht ein externer Inhalt von YouTube, der den Forumbeitrag ergänzt. Er kann mit einem Klick geladen und auch wieder ausgeblendet werden.

YouTube-Embeds laden

Ich bin damit einverstanden, dass YouTube-Embeds geladen werden. Dabei können personen­bezogene Daten an YouTube übermittelt werden. Mehr dazu in der Datenschutzerklärung.

YouTube-Embeds laden

Datenschutzerklärung

Voltage scaling starts at 11:50

Keep in mind, that he tests DDR4 and the power is running the DRam modules+the memory controller.

 

[Baal Netbeck]

Have a look at this little paper from micron in regards of DDR4 power consumption:

I did, and this is what I found:

On DRAM, power is typically related to the square of the voltage. This is because most of
the power is dissipated by capacitance, with P = CV2
f where C = internal capacitance, V
= supply voltage and f = frequency of the clock or command (see Frequency Scaling).
Thus, to scale power to a different supply voltage use Equations 24 and 24a.

 

[Piktogramm]

@Baal Netbeck
Well, reducing a 28page paper to some sentences of the paragraph in regards of "voltage scaling" alone is a tad extreme.
Also the micron paper already is toned down quite a bit (disclaimer: I don't fully understand the full scale academic papers).