This Sketchy 2000W Power Supply

[AlexTheGreatish]

We got a cheap 2000W power supply designed for mining... could it possibly live up to it's output? Also seriously don't do anything we did in this video, playing with power supplies is a very bad unless your name is Mehdi.

 

 

 

[da na]

Ah yes the classic AliExpress PSUs... good rule is divide by 10 and that's what wattage of hardware may possibly be safe to run it with.

 

[HenrySalayne]

Small correction: the plural of "Warnung" is not "Warnungs" but "Warnungen". And kudos to the manufacturer for including multilingual warning labels.

[Somerandomtechyboi]

Only experience i had with chinese psus is just an unbraded one that blew up trying to power a basic 775 rig that prob isnt gonna draw much more than 100w while being a "600w" psu

 

Guess not all chinese psus are bad though id still just buy a used psu instead, theyre going for dirt cheap

[Mandom]

I had a sketchy 650W power supply(brand was DEER) that weighted nothing! I then got a corsair cx450m(what a downgrade right?) that weighted 2 or 3 times the sketchy one. It was quite shocking how much more quality the corsair felt just by touching it and how much it weighted.

[Espresso48]

First time posting just to share some thoughts on y'alls comments in the video:

• The designs between a variable V input power supply that supports 120V and 240V is not substantially different from a dedicated 240V supply. The biggest differences will only be in component selection and control scheme. There isn't "way more circuitry" to support both input voltages. The difference in circuitry is likely connected to additional filtering, power stages and/or phases. 
• The difference in components; especially the difference in capacitor and transformer count is likely due to the push in the industry to adopt designs that use less of the components. For example, moving from active bridge boost PFC to totem-pole PFC. 
• The reason they may have "rated" the power supply for 2kW is that it likely the end-customer would only use it between 50-60% load. Due to the load-efficiency curve, power supplies are often "rated" for wattage based on their efficiency point instead of true lifetime reliability. Is this a good practice? No. However, this is just a common logic in the bitcoin / server power supply world.
• Large bulk caps are not really typically designed around voltage droop (though it is a concern) but instead voltage ripple requirements. Or at least this is the system requirement I typically see people design from. 

I am an engineer at a major semiconductor company in their GaN power device team. 

 

[Nystemy]

To be fair. I have seen worse PSUs before. But lets break these two units down a bit. _{(and partly some of the BS of the video itself...)}

 

First of, the statement about needing way more components for supporting a universal input voltage is BS to be fair. Adding more transformers isn't required in the slightest for handling a universal input voltage.

 

Depending on the type of switching technique used, there is a few ways to achieve a universal input voltage. For fly-back inverters, then universal input voltage is more or less a given as long as one's power stages can handle the higher voltage, while for resonant mode converters a more "fixed" input voltage is more preferable. However, almost all PSUs these days uses power factor correction before the main switching converter, and PFC is effectively just a boost converter, and this pairs very well with a resonant mode converter since the PFC more or less gives it a fixed input voltage to work with. _{(the PCF will just use an inductor, typically looks like a transformer though. And one of the transformers of the "quality supply" is just an inductor in disguise.)}

 

And the NTC is an inrush limiting resistor, it effectively is having a higher resistance when cold, and as current flows through it it slowly heats up and its resistance drops. This is to ensure that the two large capacitors it goes to doesn't just take a huge gulp of current when one plugs in the PSU to the wall. With the NTC the inrush is more gentle and won't as easily trip a breaker. _{(PSUs in the server/industrial scene uses a boring standard resistor instead, and then a relay that shorts out the resistor when our caps have charged up, this is also normal on some audio gear and why some receivers makes a "click" 0.1-0.3 seconds after one presses the power button.)}

 

Also, more capacitors do not make voltage drop less severe. It only makes it easier for the power supply to handle power transients more effectively. _{(ie, short bursts of added power demand do not require the switching convert to react as instantly. Same for a short term load release where the capacitors more or less absorbs the energy put out by the switching converter until it throttles back down again. In short, output capacitance is a nice thing in PSUs.)}

 

The "quality" supply shown though seems to have 2 PFC phases. _{(I can't work out what else the other inductor there could be used for.)}

And it also seems to divide up its output into more than one power phase as well. Other than the typical additional standby supply.

While the "sketch" supply just goes with the standard single transformer, but seems to still have PFC. _{(The cheap supply does have some input filtering, but the picture bellow has it hidden behind the metal mesh, and the picture is cluttered enough with text as is.)}

Here is my guess to what we are looking at:

 

Also, to test a PSU, it is generally smarter to have an electronic load, or a suitably heft resistor bank. Blowing up actual PC parts is an unnecessary risk for seeing if it works. _{(however not an unnecessary risk for actually checking if it does a good job. And a good electronic load can be set up such that it simulates a real system's power behavior and can even be set up as to stress the PSU beyond what any real computer could achieve.)}

[stipuledfatcat]

58 minutes ago, Espresso48 said:

First time posting just to share some thoughts on y'alls comments in the video:

• The designs between a variable V input power supply that supports 120V and 240V is not substantially different from a dedicated 240V supply. The biggest differences will only be in component selection and control scheme. There isn't "way more circuitry" to support both input voltages. The difference in circuitry is likely connected to additional filtering, power stages and/or phases. 
• The difference in components; especially the difference in capacitor and transformer count is likely due to the push in the industry to adopt designs that use less of the components. For example, moving from active bridge boost PFC to totem-pole PFC. 
• The reason they may have "rated" the power supply for 2kW is that it likely the end-customer would only use it between 50-60% load. Due to the load-efficiency curve, power supplies are often "rated" for wattage based on their efficiency point instead of true lifetime reliability. Is this a good practice? No. However, this is just a common logic in the bitcoin / server power supply world.
• Large bulk caps are not really typically designed around voltage droop (though it is a concern) but instead voltage ripple requirements. Or at least this is the system requirement I typically see people design from. 

I am an engineer at a major semiconductor company in their GaN power device team. 

 

Majority of the viewer base is probably not going to notice or care about stuff like that but it would have been nice to see this power supply featured once LTT Labs is set up so that we could accurate understanding of the design and capabilities instead of just a let's see if we can blow this thing up for entertainment. But then again most of the viewer base for LTT videos aren't electronics engineers so saying more than more caps and bigger caps = good would probably go over the intended viewer bases heads.

[Spotty]

2 minutes ago, Nystemy said:

(PSUs in the server/industrial scene uses a boring standard resistor instead, and then a relay that shorts out the resistor when our caps have charged up, this is also normal on some audio gear and why some receivers makes a "click" 0.1-0.3 seconds after one presses the power button.)

It's also not uncommon in desktop PSUs. Most modern high quality ATX power supplies feature a NTC thermistor with a bypass relay, it helps improve efficiency slightly by bypassing the NTC thermistor after startup.

 

4 minutes ago, Nystemy said:

Also, to test a PSU, it is generally smarter to have an electronic load, or a suitably heft resistor bank. Blowing up actual PC parts is an unnecessary risk for seeing if it works. _{(however not an unnecessary risk for actually checking if it does a good job. And a good electronic load can be set up such that it simulates a real system's power behavior and can even be set up as to stress the PSU beyond what any real computer could achieve.)}

It has been a long time since I have seen a PSU test use PC components to stress test a PSU.

Also weighing a PSU? I get the point they were trying to make but I haven't seen anybody do that since the days of passive PFC and dodgy PSUs putting blocks of concrete in their PSUs to make them weigh more, thankfully those days are long over.

 

27 minutes ago, Nystemy said:

Here is my guess to what we are looking at:

The cheap PSU is correct.

For the silverstone the board you have labelled as PSU controller is the 5VSB board. What you have labelled as 5VSB is (I believe) an inductor part of the LLC resonant converter. The controllers/supervisor ICs are possibly mounted on the rear of the main PCB.

 

 

 

Overall I'm pretty disappointed in the video. I don't have much expectations for a PSU test from LTT given they simply don't have the appropriate equipment and it wouldn't be worth buying for a one-off review of an aliexpress PSU so I can completely understand not doing in-depth and accurate testing or having access to oscilloscopes or load generators, but I know they at least have a digital multimeter, killawatt, and FLIR imaging camera lying around they could have used. Even one of those $10 power supply testers from amazon that you plug the motherboard cable in to and gives you a read out of voltages and power good timing would have been a huge improvement. This PSU test was no more insightful than the average Amazon power supply review. "PSU worked with RTX3090ti. Power supply feels lightweight, fan is noisy. 4 stars"

[dogwitch]

am surprise anthony let linus near this...

[Bitter]

59 minutes ago, dogwitch said:

am surprise anthony let linus near this...

Anthony and Alex must be next in line of succession once Linux drops himself.

[Nystemy]

2 hours ago, Spotty said:

For the silverstone the board you have labelled as PSU controller is the 5VSB board. What you have labelled as 5VSB is (I believe) an inductor part of the LLC resonant converter. The controllers/supervisor ICs are possibly mounted on the rear of the main PCB.

Yes, I do think you might be right in that.

Going off a single image is not usually how one reverse engineer a PSU, but yes, the small vertical board is likely standby. Since 5V standby is rarely a lot of current to speak off.

And it is indeed more logical for the other black transformer looking block to be some inductor for the resonance mode part of the design.

 

When making the sketch I did try to see if I could find another inductor, since for a proper two phase design each transformer would need its own. But one can run the "two" phases as just one phase. After all, resonance mode is more or less just a high frequency inverter running a linear transformer, so one can add an almost arbitrary amount of output transformers. _{(adding more will need to be considered in the design, so one can't trivially add more after the fact, since the transformer typically adds some inductance to the loop and therefor affects the resonance frequency.)}

 

Running them as one phase but with two transformers in parallel is likely done for efficiency reasons, carrying magnetic flux long distances is usually a good way to reduce efficiency, so two transformers keeps the distances smaller.

 

Though, I do wonder if a proper dual phase design wouldn't perform better in regards to output ripple and transient response. Dual phase designs usually do after all. But then one needs a second of those inductors and a second drive phase too, so that is an added cost for likely little benefit. _{(Since a PSU usually can solve all ripple and transient problems with tons and tons of relatively cheap output capacitors. Something the VRM for CPUs/GPUs can't reasonably do to the same extent, not to mention that transients have been smoothed out a bit by VRMs before they even reach the PSU. So a dual phase design is likely of little interest at these relatively low power densities.)}

[MultiGamerClub]

Reminds me when i saw 1750w power supplies..

 

edit: 15:09 "if your powersupply shuts down from your graphics card, thats is good"

 

Means my Corsair 750w Integra M protection worked.. Even tho it should have been enough? It wasnt.

 

RM1000x to the rescue.. only for HX1000 to be on sale the next day.. Sad.

Edited April 16, 2022 by MultiGamerClub

[starsmine]

1 hour ago, MultiGamerClub said:

Reminds me when i saw 1750w power supplies..

 

edit: 15:09 "if your powersupply shuts down from your graphics card, thats is good"

 

Means my Corsair 750w Integra M protection worked.. Even tho it should have been enough? It wasnt.

 

RM1000x to the rescue.. only for HX1000 to be on sale the next day.. Sad.

3080 TIs have massive transient spikes. so while a 750W PSU should be fine, Ampere has a bad habit of pulling twice the wattage it should for microseconds with load changes.
So while it should be enough, it saw the transient spikes that went out of the PSU's specs and noped out.

[Bitter]

2 minutes ago, starsmine said:

3080 TIs have massive transient spikes. so while a 750W PSU should be fine, Ampere has a bad habit of pulling twice the wattage it should for microseconds with load changes.
So while it should be enough, it saw the transient spikes that went out of the PSU's specs and noped out.

So what we need is a super cap in line with the GPU power cabling? A few Farad of capacitor should do it, right?

[Elisis]

22 minutes ago, Bitter said:

So what we need is a super cap in line with the GPU power cabling? A few Farad of capacitor should do it, right?

Supercaps are slow. Like, *really* slow. They're not appropriate for softening transients by any means. 

[maskmcgee]

Still waiting on that PC build guide that was scheduled to be released like 3 sundays ago

[Bitter]

1 hour ago, Elisis said:

Supercaps are slow. Like, *really* slow. They're not appropriate for softening transients by any means. 

Partly joking btw. I know they're slow to charge but I didn't know they were slow to discharge too, I thought they could discharge fast like caps do!

[Ultraforce]

2 hours ago, maskmcgee said:

Still waiting on that PC build guide that was scheduled to be released like 3 sundays ago

I wonder if they'll do a youtube premiere or if it will be posted like any other LTT video but just happen to be over 80 minutes long.

[shiveralex]

a KillAWatt watt meter is like $15 on Amazon…

love the BOOM mention..

[Required]

OMG 240V Sound very wrong to me!

I checked my UPS and we have thankfully still ~230V!

[Bitter]

12 hours ago, shiveralex said:

a KillAWatt watt meter is like $15 on Amazon…

love the BOOM mention..

They only measure input and tap out at 1500W peak.

[gamagama69]

On 4/16/2022 at 7:37 PM, MultiGamerClub said:

Reminds me when i saw 1750w power supplies..

 

edit: 15:09 "if your powersupply shuts down from your graphics card, thats is good"

 

Means my Corsair 750w Integra M protection worked.. Even tho it should have been enough? It wasnt.

 

RM1000x to the rescue.. only for HX1000 to be on sale the next day.. Sad.

I'm wondering if my 750w cheap rosewill psu not shutting down with a 3080 is a bad thing. Like I don't have enough over current protection to stop anything from happening

[long2905]

On 4/17/2022 at 12:04 AM, AlexTheGreatish said:

We got a cheap 2000W power supply designed for mining... could it possibly live up to it's output? Also seriously don't do anything we did in this video, playing with power supplies is a very bad unless your name is Mehdi.

 

 

 

doubt you will see this, but can you guys share the modded inf you guys used to get other cards recognized alongside 3090ti?

[Guest]

If only Jonny Guru could come here to compare this fake 2000w PSU to a real, though obviously a relic of the past, 2000w PSU. (edit, forgot to watch the video for a few seconds)

 

I don't really get ultra high watt PSUs these days. Multi GPUs are mostly obsolete, and processors already got more efficient and more thermally restrained than ever before, so most of the "metagame" PSUs today are just 1 to 1.5 kilowatt options with Titanium ratings.

Edited April 18, 2022 by Herrscher of Whatever