In the
first part, I talked about how we began to dive into the subject of liquid immersion cooling. The time has passed and I want to talk about the depths reached.
Having finished 2016 on a very successful note, the project was completed, the customer is satisfied, we expected, if not a wave of orders, then at least continuation of work. Reality has cooled our plans. The customer said that the following funding under the program can be obtained as early as 2020. We decided to occupy ourselves with some other business, while waiting for time to pass quickly. Attempts to immerse microclusters and supercomputers in the container failed - there were no exits to interested people.
And then our eyes turned on the miner cryptocurrency miners. This topic was completely new and unexplored for us. All we knew was that three small boards with chips consume almost 2 kW of electricity. We had experience with FPGA (FPGA) and we boldly rushed forward, in the sense of depth.
Unlike the previous project, where we had a very reasonable customer and TK, there was neither one nor the other here. The number of miners and their diversity plunged into depression. Indeed, for semi-industrial equipment, an important condition is the uniformity of its constituent elements. It is much easier to make a container for one type of board than for a whole bunch. The option to buy a miner and make a container for it was also considered by us. This option, among other things, promised a quick payback of the project as a whole. However, referring to the ignorance of the subject area and the rational consideration that during the gold rush it is best to sell shovels, we refused it.
Acquainted with many mining enthusiasts, we made the following conclusion for ourselves. It does not claim to be the title of truth - this is solely our opinion. The world of mining is divided into two parts - ASIKi and video cards. Among ASIKs, the lion's share is made up of products with board sizes of 15 to 25 cm, plus minus a centimeter. In the video map section we didn’t find ourselves due to the very large looseness of the farms. Even with the fans and radiators removed, the layout is very cumbersome.
We started developing the device, starting from our understanding of our target audience and the maximum electrical power available to the common man. This power is about 10 kW.
We decided to rely on our successful 5 kW box and scale it to 10 kW. The matter seemed to us to be absolutely simple and not risky, and therefore we decided to do without producing a prototype and order a small series at once. How wrong we were))) Having broken the old rule to make no more than one change at a time, we decided to make a lot of changes right away. What was done.
First of all. We decided to give a chance to the Russian manufacturer and replace the domestic German copper heat exchanger with a domestic one. They found the manufacturer and the prototype, painted in a beautiful black color, looked exactly like its foreign predecessor.
Secondly, we increased the size of the container so that it could hold 24 boards of the selected form factor 15 * 25 cm. The total power of these boards just fit into the selected range of 10-12 kW. The perimeter of the flange connecting the lid and the container itself increased accordingly.
Thirdly. We do not immerse the power supplies in liquid and leave them outside, therefore, there is a need to bring about 800-1000 amps into the container. This decision was due in particular to the fact that the power supplies are also very loose and they need a lot of liquid to cool them. In addition, the power supply miners are different for everyone and it is very difficult to make a universal container for any blocks.
I must say that all three technical solutions were failures. Unfortunately, we could find out about this only after the first batch was made. I'll tell you more.
1. The internal heat exchanger was not completely copper, the fins were aluminum. Since we are "smart" and laid a certain margin on the power of the heat exchanger, we could survive this. But it turned out that the slats are not soldered, but simply put on the copper pipe. When turned on, these lamellae quickly warmed up to a temperature of 50 ° C and stopped taking heat away.
2. The stiffness of the flange was insufficient. When heated, the metal was deformed and it became difficult to achieve tightness of the container mega.
3. Power input was done through brass bolts. The currents are not childish. At first it turned out that there was no copper fastener. There is copperplated. These are the same steel studs and nuts, covered with a thin-thin layer of copper. For us, this option was unacceptable. We found a Russian manufacturer of copper fasteners. The volume of our order caused him apparently Homeric laughter, but the order was fulfilled. The cost and timing of its implementation have added a few gray hair to our hair))) The trouble came later, when the containers began to work. Repeated switching on and off, warming up to the working temperature and cooling resulted in the fact that the fastening of the bolts ceased to be tight, the vapors of the heat dissipating fluid siphon through the transitional power panel.
I want to say thanks to our customers who dared to buy our containers. Together, we have identified all the weaknesses. According to the test results, the “work on mistakes” was performed. We have developed a new container.
