Up-Cycling Dell Server Power Supplies

Cryptocurrencies, in general, are a controversial topic. All decentralised “cryptos” use some sort of algorithm to generate signatures that ensure the integrity and immutability* of their respective blockchain/ledger. The predominant family of cryptocurrency algorithms are known as PoW or Proof-of-Work.

PoW is extremely controversial for a variety of reasons, but the biggest issue with PoW is that PoW requires machines to do… work and “work”, by definition, requires energy to be expended. The “work” is generally carried out by a processor (in some recent algorithms novel approaches like “Proof of Capacity” are being trialed).

So “work” needs power and power comes from power supplies. Power supplies are expensive and so any way of making the acqusition of power supplies for mining cheaper improves profitability.

Enter the Dell Poweredge 1950 server, which includes two very beefy power supply units capable of ~49A on 12V apiece… very nice. These servers were released to the market circa 2006, which makes them 12 years old. In the computing world, that’s ancient. They’re extremely loud and noisy, big, awkward, use loads of electricity and can’t really do that much. Unfortunately, they’ve reached their expiry date. It happens. I’ve collected several of them off auction sales and some have even been given to me.

A Dell PowerEdge 1950 Server with cap off.
Dell D670P-00 670W Power Supply – each server has two of these.

Innovative peeps have built interface cards for larger variants of these power supplies, specifically for mining operations, but unfortunately, these power supplies are a bit too small to make that kind of time investment viable. A more, er, “DIY” solution is required.

These power supplies are designed to be cooled using ram-air provided by a set of jet-like fans in the main server chassis. The servers are old and hopelessly inefficient, so I won’t be using them in future. I therefore removed the fans and drilled holes through the power supply enclosures to provide cooling.

Dell PowerEdge 1950 fans – powering them took a bit of guesswork…
Ghetto as hell, but wood-gluing the fans to the power supplies is cheap and effective.
Fans are “installed” with the airflow directed upwards to take advantage of convection and keep dust accumulation on the outside of the housing.

Next, we need some cables to attach to our power supplies. I removed one set of ends while the glue was drying in preparation for soldering.

PCIE connectors stripped and ready to be soldered.

This is all pretty high-current stuff (made worse by it all being 12V), as such it’s important to make sure solder joints are really solid (more so than what’s demonstrated in this picture).

A better set of soldered joints.
Two sets of pins need to be shorted to make the power supply run.

And there you have it! A “proper” AND efficient 670W 12V supply. These can be used to comfortably drive 3x overclocked AMD RX580 GPUs. It could probably power 4x GPUs, but as always, it’s better to er on the side of caution.

Powering the fans took a bit of guesswork as well as trial and error, but essentially fans have a positive, negative and sensor line. One set of fans use black (GND) and red (+12V). Yellow is the sensor line and for this application remain unused.

Reverse Engineering the Ubbey Box

Some time ago I came across an ICO (distressed choir wails in the background as if welcoming lucifer to a dark cathedral) and said ICO was aiming to build a decentralised storage system. “Cool”, I thought, eager to experiment with a small amount of Ethereum on a diverse set of ICOs. I gave them some ETH… but before doing so I discovered that decentralised storage isn’t a new thing. Several projects have tried (and somewhat succeeded) in building dencentralised storage networks already; big examples being BitTorrent and IPFS being (neither offer in-band remuneration). In addition, cryptocurrency-supported examples include SiaCoin and StorJ – both of which have serious issues. After contributing to the Ubbey/Universal Labs ICO I decided to give Sia hosting a try… but that’s another story.

Ubbey’s 2000 Ethereum ICO was ultimately successful and I got some tokens. Shortly thereafter Ubbey allowed ICO participants to use some of their tokens to order their “Ubbey Box” at discounted rates. By this stage I had acquired additional Ubbey tokens at a fraction of the ICO price on decentralised exchanges (I love DEXs). Yesterday, the Ubbey box I ordered arrived. I only recently got a chance to play with it.

As with most boxes I don’t control, the Ubbey box has gone onto a DMZ network, which means it can’t access my home security cameras, storage machines, geyser controller, irrigation controllers… you get the idea.

First things first, unpacking :

Very nice Ubbey… the packaging looks good, even after being bashed around by DHL.

An Ubbey Box, by Universal Labs

This box is very obviously a general purpose media player that’s been rebranded to fit Ubbey’s needs. There’s nothing wrong with that, it’s sensible. They inadvertently hint at it being a general purpose media box with their specs, which quote the device as having an ARM CPU in it with 2 GBs of RAM, a VERY common configuration for ARM dev boards from around 2 years ago. This is begging to be disassembled…

So, onto the DMZ it goes. I installed the Ubbey app from the Google Play Store and soon enough it found the box (through what looked like an IP scan of the local subnet). The scan was followed by a login prompt and then a firmware update (forced) :

Unfortunately, it’s been about 2 hours now and it’s still updating…

While we wait for it to update, let’s see what’s inside :

This is how the device looks inside with the lower cover removed (it’s upside down in this shot).
The Ubbey logo is illuminated by an over-spec’ed 7-segment LED display.
The top-side of the main board of the Ubbey Box.

The top side of the main board of the Ubbey Box is -very- interesting. The device is equipped with a dual-chain/MIMO wifi module and the box has antennas for it installed. The top cover of the box has a large metal mass, which is squeezed against the top of that shielded cage/case with a thermal pad sandwiched inbetween. The board is clearly designed to have multiple display outputs with one set of traces unpopulated. The FORESEE module is probably flash “ROM” storage for the OS, which is OpenWRT. Also present are traces/holes for a TTL UART, an SD Card module and two USB ports (one of which appears to be USB 3).

It should be noted that the Ubbey Box runs an SSH daemon by default – attempts to log in to it using root/root, root/password, root/etc. proved fruitless.

The top side of the Ubbey Box PCB with the CPU/RAM shield/case and thermal pad removed.

A quick Google search for SEA Beelink SEA I found me this Amazon page :

Wireshark packet sniffing indicates that the device spends a lot of time communicating with 47.100.119.151, which seems to reverse-resolve to api.yqtc.co. Visiting the server with https indicates that the server’s HTTP certificate expired in May… so I thought the client device may be less than strict with it’s certificate acceptance. Some DNAT rules on my router and mitmproxy later and I got this :

mitmproxy is very cool software. Unfortunately the Ubbey Box’s Go client, although permissive, isn’t that permission 😉

It would appear that the best way of gaining access to this device is via the TTL UART.

Update 22 September 2018 : Serial Debug Console

Not my neatest work, but I now have access to the device’s debug console…
Part of the Ubbey Box’s boot process and the OpenWrt failsafe mode entry point.
The Ubbey Box’s labeled partitions.
The contents of description.xml inside what looks like an overlay filesystem called “nasetec”

More to follow…