Topologies and regulations, what should I look for?

[LukeSavenije]

Introduction

This time I'm doing a short bit on a couple of topologies and methods of regulation, and their up- and downsides. A short conclusion will be written below, so you don't have to read the whole thing to make your PSU choice.

 

Topologies

- Double Forward

- Active Clamp Reset Forward (ACRF)

- LLC Resonant

 

Double Forward

Double forward or two switch forward is a single forward configuration with 2, rather than 1 MOSFET to keep the core from running into saturation

 

(Vishay)

 

D= Diode

Q= MOSFET

T= Transformer

Cin= Voltage in

 

Upsides:

- cheap

 

Downsides:

- only scales up to 750w

- not meant for high efficiency PSUs, as it generally only goes up to 80+ bronze

- due to hard switching more likely to whine

 

Active Clamp Reset Forward (ACRF)

ACRF is a topology close to Double forward, but unlike Double forward is able to continue switching without load being applied, making it more efficient, but still use hard switching. mostly produced by FSP

 

Upsides:

- relatively cheap

- shown to be scalable up to 1000w

 

Downsides:

- more expensive than Double forward

- efficient enough only to meet 80+ gold

- due to hard switching likely to whine, but less than Double forward

- mediocre design cause worse transient response

 

LLC Resonant Converter

LLC stands for L (inductor), L (transformer primary which is an inductor, too) and C (capacitor). There are two inductors (LL) and a capacitor (C) used which form a resonant circuit . It's made out of 5 parts, in case of a Half-bridge (two switching FETs, transformer, inductor and capacitor).

 

(Texas Instruments)

 

Vin=Voltage in

Q=MOSFET

Vsq= unipolar square-wave voltage

Cr= resonant capacitor

Lm= inductor

D= Diode

 

Upsides:

- efficient enough to (generally) meet up to 80+ Titanium

- low chance of whining

- high scaling in wattage

 

Downsides:

- most expensive

 

Regulations

- Group regulation

- Double Mag Amp

- DC-DC

 

Group Regulation

I went into group regulation and why it's a problem before here. It uses two coils, a big and a smaller one. The big one will regulate 12v and 5v, while the smaller one will regulate 3.3v. Thus, because the controller tracks both 12v and 5v rails as a whole, in crossload situations (if the load on one of them is high, while the other is low) voltages can go out of nominal (5% tolerance by ATX specifications). Specifically, This is common situation with modern PCs that, first, support C6/C7 sleep states, in which 5V rail get almost no load while 12V rail still loads relatively high, and second, modern PCs generally don’t load 5V rail much even when not in standby, because the only hardware that still uses it are HDDs and SATA SSDs, while 12V rail can be loaded very high, especially with high-end GPUs. This is especially troublesome with fast peaks of modern GPUs, which switch between 50 and 450 Watt multiple times per second. If the output capacitors can not buffer that (particularly in older units), the main regulator has to follow those peaks - altering also the 5 Volt output voltage with it. This leads to strong 5 Volt fluctuations even if there is little load on the rail.

 

(Jonnyguru)

 

Upsides:

- cheap to produce

 

Downsides:

- voltage can easily get out of spec due to regulating 12v and 5v together

- generally doesn't meet c6/c7 sleep states or can’t keep voltages in specs in crossload situations associated with them

- not recommended for anything beyond an APU system

 

Double Mag Amp

Double mag amp is one of the two ways of an "independent" regulation, in this case regulated from the secondary winding, using an inductor to step down the current to either 5v or 3.3v. This is relatively uncommon with the introduction of DC-DC, since this is less efficient. An example where this is still used would be Seasonic's S12iii. Also, it can not work with an unloaded output (luckily a situation that doesn’t occur in a normal PC). The picture below marks the 3 coils compared to two on group regulation, by which you can see it's a double mag amp (in this case the s12 based corsair TX 80+)

 

(Anandtech)

 

Upsides:

- relatively cheap

- individually regulated

 

Downsides:

- needs more load to work, hence generally not coming higher than 80+ bronze

- low efficiency compared to DC-DC

 

DC-DC

DC-DC uses a similar, yet quite different technique to double mag amp. it does share that it uses independent regulation, but does it in a different way. It uses buck step-down converters to lower the voltage directly from 12v to 5v or 3.3v. This is more efficient, and needs less load to function. LLC PSUs can even work properly without any hardware attached on a rail, if necessary. On the picture below i marked a dc-dc converter, in this case on a Seasonic Focus PX

 

(Relaxedtech)

 

Upsides:

- individually regulated

- very efficient, since it can function with less load

- most common in modern PSUs

 

Downsides:

- most expensive

 

Verdict:

In the most ideal situation you get a DC-DC unit with an LLC Resonant Converter, but due to budget this might not always be possible.

APU system: preferably DC-DC, any topology

Low-end gaming system: DC-DC, ACRF or LLC

midrange-high end gaming system: DC-DC with LLC

 

Sources:

https://www.ti.com/seclit/ml/slup263/slup263.pdf

https://www.vishay.com/docs/91616/twoswitch.pdf

https://www.tomshardware.com/reviews/power-supplies-101,4193-14.html

https://www.techpowerup.com/articles/overclocking/psu/160/5

https://www.anandtech.com/show/2450/3

https://www.relaxedtech.com/reviews/seasonic/focus-plus-ssr-850px/1

http://www.ti.com/lit/ml/slup129/slup129.pdf

 

License:

Spoiler

These works by Luke Savenije et al. are licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

 

Credit:

Moritz Plattner - Tech-review.de

@Juular

[Dedayog]

Fantastic for folks like you just don't have the energy to wrap their head around this stuff.  Thanks!

[LukeSavenije]

Just now, jstudrawa said:

Fantastic for folks like you just don't have the energy to wrap their head around this stuff.  Thanks!

downside: i had to

[Dedayog]

1 minute ago, LukeSavenije said:

downside: i had to

Forced good works are still good works

 

I know what ya mean tho, sucks for the do-er.

[Bartholomew]

22 minutes ago, LukeSavenije said:

Double Forward Downsides: not meant for high efficiency PSUs, as it generally only goes up to 80+ bronze

Active Clamp Reset Forward (ACRF) Downsides efficient enough only to meet 80+ gold

LLC Resonant Converter; Upsides: efficient enough to (generally) meet up to 80+ Titanium, low chance of whining

 

Group Regulation Downsides: generally no c6/c7 sleep state

Double Mag Amp Downsides: needs more load to work, hence generally not coming higher than 80+ bronze

DC-DC Upsides: very efficient, since it can function with less load

 

Verdict:

In the most ideal situation you get a DC-DC unit with an LLC Resonant Converter

Summarized like this, is it possible to translate the above to:

 

"Generaly speaking, if wanting a excellent unit with least probability of coil whine, opt for something rated higher than gold, prefferably titanium; with c6c7 sleep state support mentioned in the spec sheet.".

 

This is disregarding the budget constraint aspect of course; but it simplifies initial search a bit when looking for a good psu and limits to stuff actually (generally) listed by the manufacturer/brand info (unlike the topology / regulation).

 

Granted, topology and regulation can likely be found in reviews; which is another thing to look for

 

Asides:

1. Thank you for the interresting post!

2. If the above (by me) is deemed "plain wrong" or way off the mark etc. please be blunt/clear; im more than happy to edit the post such that at least others cant mistake it for "absolute truth" should it be deemed very misleading.

[LukeSavenije]

1 minute ago, Bartholomew said:

"Generaly speaking, if wanting a excellent unit with least probability of coil whine, opt for something rated higher than gold, prefferably titanium; with c6c7 sleep state support mentioned in the spec sheet.".

there are still platinum rated units that can have problems, like failed fan controllers on aerocool p7, high overload ripple on seasonic focus px 2017, forgetting otp on the raidmax RX-700AT and so on.

 

but purely looking at regulation and topology, an llc resonant dc-dc psu is the best

[Vejnemojnen]

would be nice to add this thread to the sticky ones.

[LukeSavenije]

1 hour ago, Vejnemojnen said:

would be nice to add this thread to the sticky ones.

you can try to see if any of the mod team is interested in doing so, but i know they're trying to limit the amount of pins

[PianoPlayer88Key]

How about if the mods make one sticky and link the other important posts in there? (They could also leave a few of the most important ones stickied as well.)

 

 

One issue I have with looking for PSUs is ... sometimes (or often) my ideas of use cases involve loading a PSU, possibly to close to its max advertised rating, using components OTHER than video cards or high-power CPUs.

 

For a few "less-normal" examples,

one could be a system with a 1600W PSU, about 1580W worth of hard drives connected, and running a heavily underclocked (0.8 GHz, 1 core with SMT off) 35W (nerfed to like 2W) APU.

Another (currently way over my budget) might be a board like the X10QBI paired with heavily underclocked E7-4809 v4's (what might their power consumption be at 0.8 GHz, 1 core, HT off?), and 96 DIMMs which is the max that board supports.  (Or is there another configuration that could run like 1.5 kW of RAM with like 15 W or so of underclocked CPU?)

A third scenario could involve a ton of usb devices.

 

 

Basically ... I want to be able to load a PSU with ANY combination of components (storage, RAM, usb devices, etc) or rails, up to the big advertised wattage number, without having to use GPUs, or mostly load the 12V rail.  It seems a Corsair HX750 or a Rosewill Tachyon 1200W PSU would be a waste if all I was powering was 16+ HDDs and an underclocked APU. (Looked at a couple other PSUs, looks like they max out at 6 jacks for SATA/peripheral cables - at 4 devices per cable that's 24 devices. Even so I still don't think that's enough HDDs to max out the wattage.)

 

 

 

 

I ran a test earlier today with a few drives (mentioned in the spoiler with other parts used) powered by my Corsair AX760 in my desktop system, hooked up to a Kill-A-Watt meter in a power strip. (Couldn't reach the wall outlet, also I know @jonnyGURU / others have said it's pretty inaccurate, but it's all we had laying around.)

 

Spoiler

 

• 1x 256 GB 2.5" Crucial M550 SATA SSD (boot)
• 1x 240 GB 2.5" Crucial BX300 SATA SSD
• 2x 1050 GB 2.5" Crucial MX300 SATA SSD
• 2x 1 TB 2.5" HGST 0J22413 5400 RPM SATA HDD
• 1x 1 TB 2.5" Toshiba MQ01ABD100 5400 RPM SATA HDD
• 1x 2 TB 2.5" Samsung Spinpoint M9T (Seagate ST2000LM003) 5400 RPM SATA HDD
• 2x 2 TB 3.5" WD Green (WD20EADS & WD20EZRX) 5400 RPM SATA HDD
• 2x 256 GB 2.5" Transcend TS256GSSD25S-M SATA SSD (power only, ran out of SATA data cables)
• 2x 20.4 GB 3.5" Maxtor DiamondMax VL 40 5400 RPM PATA HDDs
• 1x 8.4 GB 3.5" IBM Deskstar DTTA-350840 5400 RPM PATA HDDs

That's a total of 12 SATA and 3 PATA drives connected.  I now realize I should have maybe also connected power to 5 more PATA drives - i have a functional 250 GB, 80 GB, 4x 40 GB, and 2x 20 GB to choose from, but they're all 7200 RPM drives. I also had 2 M.2 SSDs available (SATA & NVMe) but I needed all my SATA ports.

 

Other parts included:

• ASRock Z97 Extreme6 motherboard
• Intel Core i7-4790K CPU (nerfed in BIOS to 0.8 GHz, 1 core, HT off, turbo disabled)
• Cooler Master Hyper 212 Evo
• 32 GB G.Skill Ares RAM (DDR3-1600, but I turned off XMP.)
• Stock fans in Fractal Design Define R5 (but I wonder if I could have run the system fanless considering the massive underclock I had on the CPU, and was using the iGPU only - my 3GB 1060 was not in the system.)

 

 

Peak wattage reading I saw with the KAW were:

• 116 Watts, when powering up the system (after underclocking ; was 129 before)
• 79.2 Watts using HDTune benchmark on 12 drives.
• 74.0 Watts using HDTune scan on 12 drives.

I uploaded a few short videos / pics to a Google photos album.

 

 

SO ...

 

How would I max out the 760 watts that PSU should be able to do, using ONLY SSDs and/or HDDs? Or at least close enough so that one more SATA SSD would put it over.  And, no adding video cards, or running the CPU above the lowest possible settings (like I had it, although I didn't undervolt this time), etc.

 

And when the time comes to plan my next build (after DDR5 & PCIe 5 or 6 have been released & come down from the initial "early adopter tax), how would I ensure that I can power any combination of components up to the "X-Watt PSU" maximum?

[Juular]

On 2/26/2020 at 10:17 AM, PianoPlayer88Key said:

How would I max out the 760 watts that PSU should be able to do, using ONLY SSDs and/or HDDs? Or at least close enough so that one more SATA SSD would put it over.

Why ?

Corsair AX760 rated 25A on 5V rail, AX1600i tiny bit above that at 30A, 40A OCP tripping point. That's about 200W, ~10W per drive, 20 drives, you'll be limited by the amount of peripheral power cables first. You can theoretically get some 12V-to-5V DC-DC adapter and power more drives through it from PCI-e cable tho, in this case you can power a literal shitton of drives, vibrational interference between them would become your concern way before you could trip PSU's 12V rail.

[Somedude96]

How and where can I find a website that lists the full spec of a PS unit? JonnyGuru doesn't have every unit.

[LukeSavenije]

39 minutes ago, Somedude96 said:

How and where can I find a website that lists the full spec of a PS unit? JonnyGuru doesn't have every unit.

Aris (tomshardware/kitguru/techpowerup), Zardon (kitguru)

Mufasa (pceva)

Poiu (tweakpc)

The Mask (tweakers)

E. Fylladitakis (Anandtech)

 

and many more generally list this with the internal shots

[Somedude96]

16 hours ago, LukeSavenije said:

Aris (tomshardware/kitguru/techpowerup), Zardon (kitguru)

Mufasa (pceva)

Poiu (tweakpc)

The Mask (tweakers)

E. Fylladitakis (Anandtech)

 

and many more generally list this with the internal shots

thanks!

[CircleTech]

So when are we going to get the Corsair AX1000 @LukeSavenije edition?

[Gamuniter]

On 2/24/2020 at 8:40 PM, LukeSavenije said:

LLC Resonant Converter

LLC stands for L (inductor), L (transformer primary which is an inductor, too) and C (capacitor). There are two inductors (LL) and a capacitor (C) used which form a resonant circuit . It's made out of 5 parts, in case of a Half-bridge (two switching FETs, transformer, inductor and capacitor).

@Juular and LukeSavenije can you point LLC topology on a PSU for me?

[Juular]

20 hours ago, Gamuniter said:

@Juular and LukeSavenije can you point LLC topology on a PSU for me?

Usually there are an additional inductor near the main transformer, usually shielded and smaller than APFC inductor. Not to be confused with small 5VSB and gate driver transformers. Also Super Flower uses two-level transformer with integrated LLC inductor in it (they have a patent for that). Also there are no big inductor after secondary rectification stage with LLC resonant topology (not sure why).

So looking at some examples, this is LLC :

Spoiler

Seasonic - LLC inductor is shielded, to the top from the main transformer (with a sticker on it) :

CWT - LLC inductor is again, to the top from the main transformer.

HEC - LLC inductor is to the left from the main transformer.

Great Wall - LLC inductor is in the top-right edge, on Corsair TX-m based on similar platform it's covered by the putty to reduce coil whine so it's barely visible.

High Power - LLC inductor is again, in the top-right corner

Super Flower - LLC inductor is integrated in the main transformer, notice how it's somewhat bulkier than usually

Gospower - appears to be circumventing Super Flower's patent by integrating the LLC inductor in the main transformer horizontally

This is double forward :

Spoiler

Notice how in all cases there are big unshielded inductor on secondary side after the transformer and secondary rectifiers, near DC-DC modules in contrast with LLC where an inductor is usually shielded and located before the main transformer or near it.

Andyson :

CWT :

FSP (ACRF, not double-forward) :

HEC :