Tech Teardown: Amazon $33 1KW Modular Cryptomining PSU - How bad can it be?

[mnementh]

Part 1: How Bad Can It Be?  Will I Get a Power Supply or a Box With a Brick in the Bottom?
                                    

Excelvan 1KW Modular PSU -$40 $33 on Amazon

   So last week, I'm migrating my old gamer rig (VERY OLD) to a new case for room to use a new video card I got for STEAM compatibility; I find my stalwart but aging OCZ Modular PSU's MB cables are a wee bit too short. In keeping with the new case, I start looking at extension cable sets, thinking "Guess I might as well dress that up too". Lots of spiffy individually braided kits in the $25-35 range, but I just can't seem to make myself pull the trigger.

Then at the bottom of page 2, Amazon plonks a "suggestion" for this:

"Computer Modular Power Supply/PSU for PC/Desktop/Gaming Computer, 1000 Watt 80+ Bronze Certified PSU with Silent 140mm Fan, 3-Year Warranty: $39.99"

( Now $32.99... I just bought another. No, I don't know why.  )

So I'm thinking to myself "Okay... no way this thing is really good for 1KW... and no "Certification" visible anywhere, aside from the FCC and CE stamp on the rating plate which are usually meaningless for China-direct product such as this. But 1KW AND Modular? There has to be some error. I'll bet it's mis-marked and these aren't really modular."

I don't care about "Bronze" efficiency; my computer is asleep most of the time anyways. I DO care about modular cables; I long ago exorcised the tentacle monster from my PC builds, and I DO care about it being at least reasonably quiet. 140mm fan promises that, and I think there's a chance I might actually get an error in my favor for a change. Maybe somebody just needs to liquidate a containerload of GPU Cryptomining gear quick; now that the bottom fell out of the market a few months ago, even out-of-touch old curmudgeons like me know the bubble has burst for GPU mining.

(Got me a XXX Edition Radeon 580 for $125 in absolutely mint condition last month. )

So I put it and some other "add-on" items to my cart and wait for it to come. I'm dreading opening the box... I just KNOW it's going to be something... ANYTHING other than what I bought it thinking it was.

So take it out of the bag plastic bag (Yup... so ugly they hid the box inside a mailing baggie inside a box) and...

Holy crap. This thing is heavy. like 3-4 Kg. 

It comes in a big black iBox.   

It has one word on it: Excelvan. Oh crap... weren't those the guys who made all that "White Van Special" audio gear back in the day...? 

Open it up and... it's a big honking PSU. With a big honking bundle of MODULAR power cables. A kilo of them!!! Okay... cheap, cheesy shit ones, right?

Eeeehhh... yeah, cheap. Every one of the modular cables is made from 18ga Pliovic unimolded flat-ribbon. So close to the cheapest you can get, but not quite. But plenty heavy enough, and definitely will sleeve up nicely to match the rest of my case modding. At least the Pliovic sheathing is flexible... I loathe stiff Vinyl-clad wiring with a passion.     And look... a 16ga IEC cord! 

Really, the overall feel of the thing is quite satisfyingly hefty, and rigid. A short drop onto the bench yields a very satisfying "WHUMMMP!" 

Cable list is:

QTY    DESC

1        24-Pin to 20+4-Pin ATX MB Cable, 490mm long
1        8-Pin to 4+4-Pin CPU 12V Cable, 495mm long
6        8-Pin PCIe to 6+2-Pin PCIe Cable, 500mm long, with 2nd 6+2-Pin PCIe connector on 50mm pigtail
4        5-Pin to Daisy-Chain 2 MOLEX & 2 SATA, 500mm to 1st SATA, 150mm to next SATA, 150mm to next MOLEX, 150mm to last MOLEX (No 3.3V)
1        5-Pin to Daisy-Chain 2 MOLEX & 1 SATA, 500mm to 1st MOLEX, 150mm to next MOLEX, 150mm to last SATA (No 3.3V)
1        16ga AW/G IEC Power Cable, 1.5 meter long

Hmmm... things are starting to look better. Aside from the cheesy documentation. And included warranty is 1 year, not 3 years as advertised. And nowhere is there any kind of seal for the alleged "Bronze Certification"; that thing on the corner is just "80% Efficiency", NOT any of the accredited "80 Plus" logos.  And when I look on the Excelvan website, all I find is small appliances and cheap personal electronics. Hoo boy...

And then, while I'm handling fondling it... just about the point I'm thinking to myself "Wow... would you look at those big fat inductors!!!"... out shakes The Obligatory Bit of Loose Solder.   

So, now comes time for the "Teardown" part of the Tech Teardown; of course I'm not going to plug this thing into my house in until I'm reasonably sure its SOME measure of safe to do so without releasing the pent-up fury of a thousand angry pixies. 

END PART 1 FOR NOW - Will Return WITH MORE ASAP.

mnem
*Juicy*

[mnementh]

Part 2: Hell Hath No Fury Like 1,000 angry pixies!

                           

Okay... so when last we looked in on our brave TinkerDwagon, he was about to... oh, would you look at that. He already has it apart. AND taking pictures. 

Taking a look at the left side of the PCB, we can see an even bigger inductor than those we spied from the outside; one would expect this as the 12V rail is supposedly capable of delivering 73A/876Watts! And filtering that juice is... 8800uF of the cheapest Asian-manufactured electrolytic caps imaginable.    I mean, "Yay! 8800uF of Low-ESR 105°Caps!!!" but "Eeeehhhh...." on their choice of brand/product line.

Just as an aside: I know about this because I'm in the trade, but "Cheap as F***" electrolytics are EVERYBODY'S dirty little secret when it comes to computer PSUs; even your prestige-brand gaming gear that comes in black chrome and aluminum extrusion has... cheap-ass caps. Some will argue it's simple economics of cost per 100K units, some will argue that it's deliberate to force turnover when they die. But they're just a fact of life.  (Not to be confused with cheap ass-caps, which is just an unfortunate byproduct of the corporate mentality)

But right next to those cheap caps; what the hell is that... a copper heat-sink? NO! It is the freaking 12V Rail to the front panel!!!

Holy BLEEP!!!ing BLEEP!!!CRACKERS, MAN!

I can actually believe in that carrying 80 amps without breaking a sweat. Without getting warm, even. 

And would you look at that... 12ga bus wires for the 3.3V & 5V rails. Damn... not too shabby. Not at all. I'm beginning to think this thing really was made to power a mining rig. I mean, there is some serious industrial-duty design in here.

Looking at the other side, we can see... NOT the cheapest possible electrolytics on the line-voltage side! I guess if you're going to cheap out on your caps, better to NOT do it on the side that has an open channel to all the angry pixies. Next another massive inductor, some rectifiers on the ample heat-sinks, and OH, MY GLOB!!! Is that a RELAY?!? SERIOUSLY? I am REALLY beginning to like this PSU! 

Next we get our first glimpse at the belly of the beast; lots of SMD here, but still lots of big heavy THS components to handle big current with ease... not the least of which is the other leg of that big honking 12V rail. Man, they are SERIOUS about getting that juice to the far side of the front panel unmolested.

You can also see the huge solder pads where the +12V & GND rails solder to the mainboard... just awesome. And there's a big old slab of silicone thermal pad there under the switching elements, to augment the heatsinks and wick heat away to that huge, thick metal shell. 

Next up are overviews of the main PCB... here you can see the general layout, and grok in fullness all the delicious inductor pr0n.  For as busy as it is, and how much space that 140mm fan displaces inside the cabinet, things are actually pretty tidy. And Praise Ifni! Instead of hot-snot, they used silicone RTV when they needed to glue stuff down. 

Next we can see a close-up of the huge chunk of copper they used to make the GND bridge to the front panel; not only is it a soldered pad almost 25mm square, it also has 3 screws through it! I think we found the source of that loose bit of suspiciously disc-shaped solder! 

And finally... the business end of the thing. Here you can see all the breakout of all the sockets. All of the 5-pin sockets are wired and complete circuits for 3.3V, 5v, 12V and dual GND. All the 8-Pin sockets are 4+12V/4 GND. The 24-Pin socket for ATX power appears to be a proprietary pinout; the cable is not 1:1 from end to end. 

If you look at the board, you'll see some spots populated with 100uF/16V electrolytics. There is 400uF worth on each of the 3.3V, 5V and 12V rails. C26 and C27 unpopulated are across the 5V and 12V rails; all other unpopulated spots are for the 12V rail. That's 1300uF left out, plus whatever they didn't put in C27!!!    

I mean... yeah, of course it's not mission-critical. But it was good engineering to design in a "bank of filter caps" right at the end of the heavily-loaded 12V bus, with lots of small value caps to massively drop ESR right where it was needed most.  But either as a cost-cutting measure or due to shortfalls in supply chain, that clever filter bank never made it out the door. 

I really feel a need to replace those missing caps; but they're a little bit of a bastard child and will need to be ordered. I have lots of caps that will fit in there at ~8mm diameter, but they need to be 11mm high or shorter to fit behind the panel, and every damn one I have of any useful value are 12.5mm high. 

I do have oodles of these 470uF/16V ones left over from another project I think I can sneak in from the back side however... and maybe I'll touch up the soldering on the 12V rail.

But THAT is a story for another day.

mnem
 

[mnementh]

Part 3

mnem
Yes, I plan to use this much room. 

[mnementh]

Part 4

In Contusion:

mnem
Maybe even this much. 

[bitseeker]

Not bad for $30!

[oPossum]

Is IC504 a TL494/KA7500 ?

What is the PFC chip and buck regulator controller chips?

[mnementh]

Not sure it does PFC; it doesn't even have minimum current failsafe on the 5V rail. I can turn it on with nothing but a 1K resistor across PSON/GND. This thing seems to be primarily brute force; lots and lots of 12V watts for cheap.

That kind of circuit analysis dissection was supposed to be further down the line... I started down the ripple measurement rabbit-hole for a couple days and kindof got lost there for a while; need some time to get my wits about me again. 

mnem
 

[mnementh]

I looked through the pics I took while I had it apart; I've identified a Weltrend WT7502V PSU Supervisor (IC401). IC502/504 are a Fairchild FSBH0170A SenseFET driver.

The buck converter drivers appear to be on the little sub-boards; I'll have to look for those part numbers next time I have the cover off.

mnem
*Back to my cave*

[Deodand2014]

Are you sure the conductors are wound with copper, at that end of the price range it could be copper-coated aluminum.

[mnementh]

Huh? I never mentioned the wire on the inductors in any way. 

Remember that sale price in this context is unnaturally low; somebody is liquidating these at firesale prices because the cryptomining bubble has burst in general, and specifically the bottom has fallen out of GPU cryptomining, which is what this PSU was batch-manufactured for. Excelvan doesn't even acknowledge its existence on their website now.

The funny part is that these could just as easily power pretty much any Antminer, as well as a remote supervisor PC. Just nobody's shopping them for that application, just like nobody shopping them for multi-GPU gaming rigs.

mnem
No, I am not sure. Of anything. 

[oPossum]

I looked through the pics I took while I had it apart; I've identified a Weltrend WT7502V PSU Supervisor (IC401). IC502/504 are a Fairchild FSBH0170A SenseFET driver.

The buck converter drivers appear to be on the little sub-boards; I'll have to look for those part numbers next time I have the cover off.

Thanks for the info. My interest in this is possibly converting it to a 10V to 16V bench supply for testing car audio amps. If it has a simple SMPS controller it may be practical to do that. Server power supplies are too complicated to easily mod for wide voltage adjustment.

[NiHaoMike]

I see that cost cutting on the 5V has become a trend lately. I suppose it would be just fine for a dedicated mining rig (except massive clusters of smartphones common for Perk mining, but that's largely dead now), although for a more general use PC, 13A on the 5V doesn't leave much room for SSDs and USB devices.

Even some brand name PSUs aren't particularly spectacular in terms of quality.
https://www.overclockers.com/cooler-master-mwe-gold-750-power-supply-review/

[mnementh]

I looked through the pics I took while I had it apart; I've identified a Weltrend WT7502V PSU Supervisor (IC401). IC502/504 are a Fairchild FSBH0170A SenseFET driver.

The buck converter drivers appear to be on the little sub-boards; I'll have to look for those part numbers next time I have the cover off.

Thanks for the info. My interest in this is possibly converting it to a 10V to 16V bench supply for testing car audio amps. If it has a simple SMPS controller it may be practical to do that. Server power supplies are too complicated to easily mod for wide voltage adjustment.

You're welcome! Questions like yours are the reason I did this Tech Teardown. 

Why the desire for such a large voltage range? To test how well the amplifiers power supply compensates for decreasing battery power? Most of my high-current car audio amplifier testing is done with a DPS-1200W modded to 13.5V and two 1F caps across the outputs. But I'm not into dB drags anymore; that shit went out of my life back in the Tarantula days.

I'm still doing testing, but here's a teaser: This was taken from the 12V leg of SATA Plug 1 (It makes a lot of difference where you measure on this PSU; shocking with all those missing caps at the output header) while drawing 81.6A (942W) from a home-made wet resistor load. Voltage sagged from 12.27 at idle to 11.54V under approx 110% rated load; I think that's not too shabby.

Cap'n Kirk would definitely approve.

mnem
Scotty!!! We need! More. POWER!!!
Ah'm givin' ye all she's got Cap'n, but we're just goo-in' aroo-nd in sirrrcles; ahhnd ahht Warrp Tenn, that's gittin' noo-where mighty faaahhst..."

[amyk]

The OEM model is JM-BT1800, and this may be the second time that model is mentioned on the Internet,

Heatsinks look a little small for a 1kW PSU...

[oPossum]

Why the desire for such a large voltage range? To test how well the amplifiers power supply compensates for decreasing battery power?

To test to manufacturer specifications. Typically 14.4 volt for unregulated, and a range for regulated.
To check for instability over the possible operating voltage range. Cross conduction, core saturation,  etc...
To test undervoltage lockout and overvoltage shutdown.

[mnementh]

The OEM model is JM-BT1800, and this may be the second time that model is mentioned on the Internet,

Heatsinks look a little small for a 1kW PSU...

Thanks for the heads-up! 

Ehhh... FETS are getting lower and lower IR every few months, and PWM controllers are getting better and better at keeping them operating outside of linear mode so there's almost no heat to get rid of. Plus, this design leverages the large metal mass of the case itself to aid in heat dissipation.   

Here's the Walmeck-branded unit. Sure looks like even the same Revision 1.2, though several other models I just found make it obvious that it was released first as a "tentacle monster" version, which I could tell by the main PCB layout. There they claim PFC and 93% efficiency; yet again, no certification to back up that efficency claim. I think that is at least in part utter hooey; I really need to scrounge up a Kill-A-Watt for the next part of my testing.

mnem
"Weeth seex joo get egg loll."