If everyone is moving towards 28nm and lower in not too distant future, then many will realize very quickly that physical limits of W/m2K heat transfer in air and single phase liquid can't be overcome.
Another example of lack of understanding of physics and intentionally confusing analysis. The secondary loop in Allied Control's system is single phase. Obviously this poster is neither semi-literate nor semi-numerate. As of now I see two possibilities: 1) dishonest analyst (a.k.a. shill) 2) honest analyst that confuses single-stage and two-stage liquid cooled systems with air-cooled systems. Time will show, so I'll quote the whole message against possible future deletion or editing. Compared to this Rube Goldberg contraption, the costs would be insignificant. Further, to mount proprietary gear in these "Data Tanks" also requires NRE. Unless you think that the installation is simply a question of backing up a dump truck full of PC boards to a 40-footer and tilting the bed.
https://docs.google.com/uc?export=download&id=0ByWHHc0u_thNdzB3c2hvVzJkcTQHave you read the design guideline document? The installation will be just plugging in mining hardware via edge connector (like a GPU card), which takes a few seconds at most. The same edge connector could be used for air-cooling as well, so that no NRE required. It is an attempt to introduce a similar standard like PCIe to the mining hardware industry with several key manufacturers very interested to participate. Compare that to installing fans and heatsinks with multiple screws, taking several minutes at least. It's like desperately holding on to installing a GPU via manufacturer-specific interface instead of via standardized PCIe. Setting up a few is no problem, setting up many thousands is a different story. Rube Goldberg contraption are subject to context and perspective. You may see that as over-engineered if you don't compare correctly. But with multiple hardware generations, every time mining hardware manufacturers have to design a new case, new cooling infrastructure, cooling and performance tests, assembly and manufacturing of everything surrounding the boards, this sounds to me much more as over-engineered. Considering that air is inherently inefficient to transport heat away (insulator in every Starbucks double-walled plastic mug, double-paned windows, etc.), a lot of engineering has to be employed to make it work effectively. Did you know that Tencent, Baidu and other big data centers in China have an army of technicians who do nothing else but exchanging broken fans? Did you know that Intel has reliability studies of electronics rusting away within a few months in India and China because of much higher sulfur content in the air/humidity/rain? I know for a fact that many Chinese mining operations need to exchange a lot of broken fans as well and have to deal with tons of heat issues - if not mining hardware, then network switches, PSUs, etc. Those open air chicken farm facilities will probably make it impossible to reuse most of the components like PSUs if they corrode away. http://www.intelfreepress.com/news/corrosion-testing-procedures-adapt-to-rising-air-pollution/6763http://www.oregonlive.com/silicon-forest/index.ssf/2013/10/intel_finds_asian_pollution_ma.htmlSure, some of today's mining hardware at 40nm+ may just barely get away with so-called 'free cooling'. But that comes at the expense of spreading out in low density (e.g. huge 'chicken farm' / entire building vs 1 container). If everyone is moving towards 28nm and lower in not too distant future, then many will realize very quickly that physical limits of W/m 2K heat transfer in air and single phase liquid can't be overcome. To cool down hotter mining hardware, it either requires relatively cool air (chiller in hot climate, or cold climate but then often paired with higher electricity/logistics/labor costs and taxes) and/or a LOT of air volume. I've heard of China installations having to be shut down in the summer and/or having to deal with sub-par performance because of temperature induced down-throttling. What a surprise, since it all still worked so nicely in the winter, as many have set up air flow only for that scenario... 2-phase immersion cooling has a much higher heat transfer, so that it doesn't require many pumps inside the tanks at all to transport heat away. Less moving parts = less maintenance. So how is this different from "576 power supplies [that] cost money"? How is this an example of savings?
If not having to replace PSUs, but being able to reuse them for many future hardware generations, you are not forced to buy new miners again with new PSUs. But the much more important point is that if miners are selling you a finished case with PSUs, it will add on to lead time on their supply chain. Meaning that if a new mining chip has been developed, which looks really good on screen at the prevailing difficulty that time, it might only come into your hands and will be ready for mining several weeks or months later: How fast can boards be assembled into cases if they have to be supplied from elsewhere, shipping back to a logistics center after assembly, etc.? And as mentioned above on the NRE: Customizing previous cooling solution for the new mining board, testing it, starting to order fans/heatsinks/water cooling blocks in bulk with inevitable delivery time, before assembly can even start. In comparison to immersion cooling, the mining board could be shipped out right away. That's technically not any kind of saving, but additional realized mining income. The chart in the prospectus explaining the effect of delayed deployment time and text describe that this crucial time difference of just 10 days or less getting online faster could already pay off the entire DTM container costs and still leave you a nice income on top, while others are still waiting for their hardware and still have to start setting up afterwards. The miners are lighter, but (the 40-footer + cooling tower + miners) ain't. All that stuff doesn't ship itself.
It's all in the prospectus: The containers will be set up first at a location chosen by DTM. Only shortly before completion of setup, mining hardware will be purchased at best efficiency and price at that point of time. So only mining hardware has to be shipped around. But if it's necessary to ship an entire case, then of course the costs are going to be higher. Some cloud miners have revealed that they have to resort to split up mining hardware into boards in one shipment and cases plus heatsinks, etc. in another shipment for custom duties reasons, because those are much higher on finished products than on components. Hobby miners might not care if one or a few cases slip through customs. But if it's about an entire farm, the story looks completely different. And then again, we're not talking about one generation only (although those savings are already significant, despite shipping the container first time) - for every subsequent mining hardware generation, there are again much higher shipping costs. A 2U case is about the same volume as probably 5-8 mining boards, while already set up DataTank containers don't need to be shipped any more and will be ready for new generations. When a board (blade) goes bad in the "DataTank," replacing it isn't as trivial as replacing an air/indirect water cooled board. A whole cluster needs to be powered down and cooled before being serviced.
See above about replacing time. Consider furthermore, that most cooling solutions are already at today's maximum. If the TDP is higher, then they will need a new solution. So it's very likely not just reusing the old cooling solution for new mining hardware. You can be pretty sure that a cooling solution for 40nm or even 28nm won't work for 20nm or 14nm. Who is deliberately shipping oversized heatsinks and fans with current generation of mining hardware? Even if it would work, do you really think that mining hardware designers already consider to have the same PCB mounting holes for heatsinks, fan connectors, etc. at the same place for the next generation? With 2-phase immersion cooling all this doesn't matter, since the fluid automatically surrounds the new mining hardware, no matter which shape and format. If you extrapolate the 4kW simulation in 200cc fluid in 1L space to an entire server rack, you would be at a theoretical 3-4MW per rack - try imagining how many future mining generations that would be if today's air cooling capacity maxes out at maybe 35-45kW per rack with LOTS of effort. It's not only cheap for the first generation already, but the costs can be truly split over multiple hardware generations because of it's incredible excess thermal capacity. And how do you know that it's needed to power down a whole cluster? A single slot of many inside a tank needs to be powered down, since the PSUs are connected to up to 8 boards within that slot. But even that may not be the end of the road yet. Judging by the vids, hundreds of boards are sitting in the same Novec tank. This is a closed-loop system, with the vapors condensing and returning to the tank. Opening that tank with the miners powered up and boiling Novec would do 2 things: Vent the vapor into the environment (thus losing exorbitantly expensive Novec, if nothing else), and suffocating you (unless you think that Novec turns into air in vapor phase).
Again, how do you know? Anitrack has posted a picture of invited press, 3M people and himself standing around an open, bubbling tank. Antirack is obviously still alive and posting... furthermore, it's on the company website that it's Allied Control's expertise to minimize fluid and vapor losses. Having a way to hot-swap is one of the technologies already developed. Let me quote myself from another post: On toxicity and evaporation and all other arguments on that it's not practical - one of the many articles on the web: http://www.pcworld.idg.com.au/article/542462/intel_sgi_test_full-immersion_cooling_servers/Does it mean that Intel, SGI, U.S. Naval Research Laboratory, Lawrence Berkeley National Laboratory, Schneider Electric, etc. are all wrong, suicidal and have no idea what they're doing? Interesting... maybe you could make much more money by teaching all their PhD's, etc. a lesson in physics and chemistry and prove they're all wrong... It's also used as fire extinguishing agent in the Library of Congress and as well the military for extremely confined spaces with troops inside as a much healthier and better alternative due to its much lower no effects limits than other solutions - it is intended for fogging an entire room with people inside with high concentrations of Novec, not causing any suffocation: http://solutions.3m.com/wps/portal/3M/en_US/3MNovec/Home/News/PressReleases/?PC_Z7_RJH9U523007AE0IUQAESTF39O3000000_univid=1273685423873http://www.youtube.com/watch?v=KNSHsUWcploIt's perfectly fine to have another opinion and I will always respect that. But spreading wrong claims based on no source at all is not really strengthening any position. |
|
|
|
So basically you are saying immersion cooling is too expensive but you have no idea what it costs to build a 1MW farm?
Actually let me write a short summary: Datatank advocates two-stage Novec+water cooling system that is two-phase immersive in its first stage. The successfully deployed pilot installations used either chips unsuitably packaged for high-power application (Xilinx Spartan) or used under-engineered ASIC chips&boards from a vendor who already admitted to not being able to hire suitable engineering talent. All Datatank systems already/will contain a full power water cooling system (either open-loop or closed-loop) that works as a second stage of their design. In my estimation a large scale dedicated Bitcoin miner will realize savings in both capex and opex by utilizing a single-stage liquid cooling system that is essentially equivalent to just a second stage of the Datatank system. Those savings are contingent on being able to hire suitable engineering talent with expertise in high-power and mixed-signal installations. In my estimation the only way where investing in Datatank securities and/or equipment makes sense is when the Bitcoin mining is just a temporary stage in a long term strategy to compete in the compact/nomadic data center market currently served by e.g. SGI. |
|
|
|
You understood both those abbreviations perfectly fine so I would say I am not misusing them. And a bitcoin mine is a datacenter. Not sure what is even debatable about that.. Regardless of the semantics you completely avoided the question.
Air conditioning is not required? So you know something that everyone with a datacenter in warm climates doesn't?
It seems like A and C perfectly apply to you. As for B, I believe you understood the question.
Again how much do you think it would cost to build a 1MW "bitcoin mine"? (excluding the asics)
OK, so you decided on sophomoric trolling. However the last question is sensible (without the parenthetical qualifier). Again how much do you think it would cost to build a 1MW "bitcoin mine"? My answer will be to the general readers of this thread, not to jimmothy in particular. That question is interesting. I assume that 1MW was just a shorthand for "industrial size" sold in 1MW increments. With this type of investment you can afford to hire the services of two engineers: 1) ASIC engineer with knowledge of mixed-signal and/or analog IC design including the IC packaging. It doesn't take a rocket surgeon to understand that plastic QFNs and thousand pad BGAs aren't the proper solution. I've actually given this advice to friedcat long time ago, he replied and then quickly deleted his reply. https://bitcointalk.org/index.php?topic=91173.msg1062969#msg10629692) chemical process engineer with knowledge of thermodynamics-intensive processes and basic understanding of electromagnetic field theory or electrical engineering. He'll help you design the proper heat exchange system while paying attention to the electrical power&signal requirements of the chips. Just let the two of them work for couple of days/weeks and they will come with a truly competitive Bitcoin mining installation. |
|
|
|
2112, How much money do you think it would cost to build a cheap as possible air cooled 1MW DC? (including the cost of the warehouse, heatsinks, fans, AC, psus, basically everything but the asics themselves)
There are multiple problems with your question. It is seriously slanted. 1) DC - I believe you meant datacenter. When I write DC it always means "Direct Current" and I always spell out the whole word datacenter. To repeat myself from just previous page: Bitcoin mine is neither a datacenter nor a supercomputer. It is a mass quantity of unusual chips with very rare property: they are all about gazing at their own navel and thumb twiddling, they do not communicate with each other and barely communicate with the external environment. 2) AC - I believe you meant air conditioning. When I write AC it always means "Alternating Current" and I would always spell that out. Again I repeat myself: air conditioning is not required. Bitcoin mine is basically a giant distributed semiconductor resistor, unusual with the requirement that the voltage it dissipates has to be less than 1V with somewhat specific tolerances. For such a resistor air conditioning is not required, there are many other ways to evacuate the waste heat of computation/thumb twiddling. So I'm basically left with the following possibilities: A) you are trying to troll me B) you haven't read and understood what I wrote C) you are trying to use a strawman argument Which one of A,B,C applies here? |
|
|
|
I simply give up. You are not a friend of this technology and you believe I am misleading people. I accept your point of view of course. Can you please confirm or deny that a business plan with a giant heat sink (and apparently the invention of a complete new power distribution system) has been interrupted? In other words, do you believe what you are saying because you are fascinated by technology, or is it because you are trying to create your own efficient and competing solution to what DataTank is? The latter would explain a lot. DC Power: I am all for new solutions that are cheaper (evidently by the use of immersion), but I need to build mines with existing technology. Highly efficient 12V power supplies at less than $0.10/W are part of this. DataTank also doesn't mind using 48V or even 480VDC. Anything fits in them. But currently not too economical, without reinventing the wheel, but trust me when I say there are tests being down and there is progress in this direction  Will it be ready in a few months for deployment? Probably not... there are also some ASIC/board manufacturers looking into this direction and a whole industry is trying to create more efficient DC power systems. Space requirement: 1 container can hold the equivalent of 576 SP30. It has the foot print of exactly 1 container. With everything. I do honestly not understand what you are trying to say and why you think it would require the spac3e of 6-9 container units etc. You are free to think about me what you like. I know who I am, and that I'm not affiliated with any pro-Bitcoin or anti-Bitcoin organization. On the other hand from you I get a strong Amway/Quixtar vibe. Except that in your case you are a 3M-captive sales-person or sales-engineer. Until you started posting in various threads "3M Novec will let you get rid of heathsinks, fans, water, cables, etc..." I thought that Novec is a solvent. I've seen Novec name multiple times on the empty solvent containers, next to trichloroethylene, tetrachloroethylene, isopropyl alcohol, etc. in a nearby trash storage area for a medical facility. There were all kinds of generic names there, but two proprietary brands caught my eye: 3M Novec and Solvay Galden. Before writing this post I did both Google and local search on this forum for Galden. This mention is going to be the first. I've met many honest consulting engineers and honest investment advisors. All of them make a strong ethical point for doing a proper competitive analysis of the field before they give an advice. How come you as a "immersion cooling specialist" never mentioned a leading competitor to 3M? Because you are strongly encumbered by your affiliation. And this is a serious red flag, when you never mention potential competitors or risks. What else are you hiding in your marketing materials? It is difficult to get a man to understand something, when his salary depends upon his not understanding it!
|
|
|
|
Belatedly, but you finally admitted that "heathinks and fans are required": and of course a radiator is required. Physics are physics. I am not hiding anything. Why would I want to mislead anyone about these facts and why you are so hostile  Have you been working on a business plan that was interrupted by DTM? While you've made numerous posts that they aren't required, last just couple of hours ago. - Fans, fans, fans. More fans. Even in Iceland. And dust. - And so on. There is no doubt in my mind. Heatsink miners will soon stay behind. And go deaf too I really have hard time to understand your motivation, why you've spend so much time on this board to confuse prospective buyers/investors with word games. Don't tell me you aren't confusing, the most recent example of a confused reader is Lincoln6Echo. You keep giving intentionally slanted numbers. When making comparison with SP30 you've used only the volume of "payload" 2TEU cointainer, whereas your whole installation requires somewhere around 6TEU or 9TEU of space for operation and 4TEU to 6TEU for shipping/storage. Why aren't you making that clear? I'll let other prospects to do a more detailed analysis of your numbers, I just wanted to point another obvious omission: power supplies. Bitcoin mine is not a datacenter or supercomputer. It does not require buying and immersion cooling large quantity of modular precision power supplies that output +12V,+5,+3.3,-5,-12V. It will work with non-precision regulated ones, e.g. DC rectifying welders (note for home experimenters: defeat the "arc start" circuitry in them). Using high power DC supplies will also allow you to directly tap the utility power at the kilovolt voltages and skip the costs and loses in the intermediate step-down transformers (which once again antirack omitted, conveniently for him). The side benefit of researching cheap single-DC-voltage power supplies will be that the interested reader will see the proper power electrical engineering at the megawatt scale and see the properly engineered cooling and isolation systems using liquids (water, oil) and gases (air, SF 6). |
|
|
|
Immersion cooling is still in experimental phase, b/c it is not yet clear what secondary benefits can arise. That's the gamble investors take.
Secondary benefits include:
- rapid deployment
- less/no impact from ambient temperatures
- operating temperature may be adjustable
- higher clock rates may be possible (e.g. 10% increase reduces the capital cost requirements for hashing power, adding to the ROI of the DataTank)
However, to be fair investors need to determine for themselves whether spending bitcoins is worth the gamble. BTC strikes me as undervalued right now, which may confirm what Spondoolies is asserting. But then, by the same logic any investment into bitcoin mining is questionable at the moment, if you don't have access to equipment at cost.
I mostly agree with you. I just wanted to emphasize that you didn't make it clear that you comparing direct air cooled system with two-stage Novec+water cooling system. Maybe some benefits of liquid cooling can be realized in a single-stage water cooling system? |
|
|
|
DataTank isnīt ommitting anything, as the neccesarry cooling equipment is included in the 0.6-0.7$/W cost of the whole setup. Before calling anyone illiterrate, maybe first open your own eyes and try to understand the technologies at work here, afterwards everyone will be glad to answer any remaining questions. Oviously the 3M Novec is only used for heat transfer, and doesnīt magically make it dissipate into "nothing", chillers on the top of the Containers are used to get rid of the heat. However, due to the quicker heat transport away from the chips, you basically only need a giant fan on the chiller to get rid of excess heat, leading to a very good PUE. You do realize that a facility already exists where asicminer has deployed 0.5MW of equipment? Please, 2112, go ahead and tell them that their existing facility isnīt supposed to work and ask them, where they are hiding their giant chilling towers and water pumps  Actually antirack was intentionally omitting the costs and dimensions of the secondary loop. Allied Control was helping him in their production and editing of the marketing video they did for the FPGA farm. For the Novec tanks they used wide-angle lens whereas for the water tower they used a brief telephoto shot. I admit that I don't fully understand what is the relation between antirack and Allied Control. I'm primarily against antirack's deceptive marketing. To me he seems like small-time huckster who sold a quantity of small laboratory-size open-loop systems. By "open loop" I mean takes cold water from the faucet and dumps hot water into the drain. Those weren't the industrial-size facilities with different economies of scale. One can still operate "open loop" at the industrial scale if the Bitcoin mine is located near the waterfall or another source of a free cold water. As to "where are they hiding chilling tower and water pumps"? In the plain sight, on the front of their prospectus. I see a three level pyramid. The fully enclosed "payload" containers are on the bottom. On top of it is a layer of semi-open containers, probably pumps and fans. Then on the top there is a third layer of open frame containers, probably the tops of the water towers. So you have two "overhead" containers per one "payload" container. Do you have any better, more accurate data, to actually compare the three systems: (0) direct air cooling, (1) single-loop water cooling (2) two-loop series Novec+water cooling? |
|
|
|
Correct. It's not "heatsink vs no heatsink". A fair comparison is 5000 heatsinks/fans vs 1 giant heatsink/pump/fans. The end result is quite clear from the HK DC with a PUE of less than 1.01.
If 2112 did a bit of reading instead of calling everyone illiterate he would know that you only need a very small amount of novec fluid per KW. In fact as little as $2.5 per KW (200ml @ 4kw) or $2,500 per MW.
OK, by why a sales represenative of Allied Controls can't state this clearly? Why suddenly independent people jump to the defense of Allied Control? Again, my argument is not only about cost of the Novec fluid. It is about a sum of costs of Novec and the whole secondary loop. Care to quote actual numbers for the cost of the secondary loop? Don't tell me "is included". Unbundle all the costs the same as antirack did unbundling costs of enclosures and small fans in the air-cooled system. |
|
|
|
well, the ''magic'' of two phase immersion cooling is that the change from liquid to vapor state consumes a lot of heat energy! That's thermodynamics, too.
So you need a lot less cooling equipment than with air cooling or normal water cooling.
This phase-change occurs only in the "primary loop". I use "loop" somewhat inaccurately because it is a natural convective boiler-condenser setup, not a pumped loop. The Novec is condensed using a normal single-phase water loop with normal radiator and fans at the cold end. They work in series. Novec immersion cooling is simply a replacement for lack of proper engineering of the mining boards. Bitcoin mining for sure doesn't require large flat boards. All the heat-dissipating components can be directly mounted onto the secondary cooling loop using larger quantity of smaller PCBs. This whole Novec tank is just a distraction here. As stated before a huge amount of heat energy gets absorbed for changing phase (liquid to vapor/gas) of 3M Novec. That's the advantage to one phase watercooling.
The phase change occurs only in Novec. Then in series with this works a regular single-phase water cooling system. I'm getting flack for calling people semi-literate or semi-numerate. So what is the better word to describe a person like Lincoln6Echo who seems to have completely missed that the Allied Control cooling system is a series connection of two cooling systems? A mark? A dupe? I blame this partially on him and partially on deceptive marketing from antirack, and partially on Allied Control being unclear in their marketing materials. Edit: OK, lets use the word "confused". So we have confused prospecive investors confused be the confusing marketing materials from Allied Control with the confusion increased by antirack (personally) by omitting important necessary items in the cost projection. |
|
|
|
Ok, lets see. DataTank = Cheaper. We can build a few for you if you want, and I can almost guarantee you will love them and thank us later - Price of DataTank = ~$0.50/W = $500/kW.- Ready to mine machine, including power, network, servers, liquid, everything required (brand name parts). - Just add internet and a power feed from the grid - Or even easier, let DataTank Mining run it for you and we charge $18/kW per month (+ a modest 20% managing fee) - Reduces power consumption (lower leak currents, no fans, etc) - Lower price is the whole point, why would we build it otherwise? ASIC hardware side: - ASIC hardware is now reduced to boards and ASIC chips. - No more power supplies (ie. SP30 = 2x 1200W, $200-$300? so $100 to $150 per kW) - No more enclosure ($50 less??) - No more heat sinks + fans $25 less?) - No more assembly etc (maybe $50 less?) - No more heavy duty FedEx shipping charge (maybe $150 less? so $75/kW) - $300 to $350 saved on each SP30 so far (and days or weeks assembly and purchasing time?) For heat sink mining you still need a building, some CRACs or fans, and of course power: - $100/kW cheap hosting (or more) = $600/half year- Or build your own mine = ? few hundred dollars per kW China style to few thousand dollars per kW Washington State style - 19-Inch racks ($300-$500 in China?) - Switchgear, power distribution, cables and power strips? (DataTank has fully automated remote PDU) - How much space for 576 of your SP30? That's what fits in a DataTank. - 144 racks at 10kW/rack, or 72 racks at 20kW per rack (and 2,500 CFM / 1,180 L/s of air flow per rack) (20 to 30 racks in shipping container, so a couple of shipping containers on the way from China just for your empty server racks, another $100 to $200 just for shipping) - Roof, walls, building etc ? - Fans, fans, fans. More fans. Even in Iceland. And dust. - And so on. There is no doubt in my mind. Heatsink miners will soon stay behind. And go deaf too As usual antirack is omitting the cost of: - water tower - water pump - water circulation pipes - fans in the water tower His whole marketing campaign is aimed at illiterate/innumerate people, those who can be convinced that the law of thermodynamics can be somehow suspended after paying for an expensive patented 3M cooling fluid. The reality is obvious: you can't omit the heatsink, but it can be either local or remote. Interestingly, they actually built some of their water-cooled setups so they must have the exact costs available, they just don't disclose them, hoping to snare some semi-literate or semi-numerate investors who'll buy into the "no more heatsinks nor fans" spiel. |
|
|
|
The SoC is proven silicon. Including RAM, it accounts for less than 2% of the die, and improves functionality and performance.
Using the same voltage and clock rate for all areas of the chip is common in simpler ASIC designs, and could cause issues similar to those mentioned by 2112. Fortunately, our design isn't that simple, and our Lead ASIC Designer is very experienced.
Another answer from a marketroid. Regardless of the design two most important things will be the substantially the same on the whole chip surface: temperature and electrical noise. The SoC would have to be super-overdesigned to survive both the temperature and the noise where the hasher cores still operate with tolerable fault rates. Such SoC controllers do exist (e.g. radiation hardened designs with lockstepping and other fault tolerance features) but they require additional manufacturing steps/masks. This would be a complete waste for the remaining 98% of the circuit and will increase both the NRE costs of the masks and the time to fabricate the wafers. Most likely also the software miners would have to be changed to actually take advantage of the SoC, when it isn't compatible with Linux (e.g. ARM Cortex-M instead of Cortex-A). I do not negate the possibility of the Lead ASIC Designer being very experienced. He's probably experienced in the low-power design for the battery-operated devices and didn't have time to fully understand the coin mining market requirements, which are more akin to the high-power, mixed-signal integrated circuits. And being very familiar with hammers he simply viewed mining as another nail to hammer in: just use SoC to bang two clocks. And the man who has two clocks never knows the correct time. Anyway, lets see if the "Lead ASIC Designer" will post here or if this is another Novello. |
|
|
|
Why are you wasting die area to SoC, and what SoC IP are you using?
It's not just waste of die area. It is a beginner design mistake: upon hash miscompare they won't know if the fail was in the hashing pieline or in the CPU. So essentially they won't be able to do overclocking/overvolting controlled by the hash failure rate. This is the same design mistake that the helveticoin did in their tested,but never deployed chip. |
|
|
|
Secondary cooling loop filled with what?
This is my key objection about your deceptive marketing: your magical fluid magically condenses on the lid of the container and drops back into the immersion bath. Either a miracle of thermodynamics or heat is tunneling into an alternate dimension. Rest assured that there is no commission from 3M. Find a better/cheaper liquid and we'll use it The commission structure is between you and your manager/vendor. Maybe somebody is ripping you off? The technical facts are simple: anything involved in Bitcoin mining that produces significant heat output can be rather easily redesigned to be wrapped around heat exchange cylinders. If anyone is building an industrial-scale facility they can easily hire the relevant engineers to design small PCBs. There is no technical requirement to become dependent on a single vendor of astronomically expensive patented heat-exchange component. Bitcoin mining is not limited by the speed-of-light or similar that constrained the computer designers at Cray, IBM, Intel and others. |
|
|
|
3M Novec surely could be cheaper, but what many people forget is that you only need a tiny little bit of it. You can cool 4kW of hardware with 200cc or 6.8oz of fluid if you want (that's what's tested, surely not the limit). That costs less than my usual double espresso per kW or less than a pizza for the complete 4kW. I would prefer reducing my hardware to a simple chips on boards any time over creating your a heat exchanger or my own buck regulators and put it in bulky boxes with cables. It's simple, really. Every facility you build has a price tag and its own set of pro and cons. Usually we compare cooling, and only on the device level, but there is much more than that to consider (power, labor, manufacturing etc). DataTank Immersion (3M Novec): <$0.65/W ( PUE 1.01) -> $650/kWMineral Oil Cooling: $3.14/W ( PUE 1.15) Economizer (evaporative): $7/W ( PUE 1.08) -> FB $750/kWWater Cooling: $10/W ( PUE 1.10) Cheap air cooling: $11/W (PUE 1.3-1.5? HK is PUE 2.2!) For you as hardware manufacturer, DataTank/Novec Immersion (<$0.65/W) has further advantages on the mining level: - Includes power supplies, networking, everything required to start mining - You don't need enclosures, heatsinks, assembly work, power supplies, all that - I estimate that will make a 20% to 50% difference on manufacturing costs of your mining hardware - You can ship your next hardware to the same customer in the same cheap form factor - One single person can upgrade a whole container farm within hours (ie. vs. how many palettes full of heavy 2U boxes that need to go into dozens and dozens of racks) Facebook prides itself for paying $7500/kW ($210M total) for their Prineville data center in Oregon (it's a 28MW facility). It took them years to build, they have created thousands of jobs, and they evaporate 7 liters of water per kilowatt per day. Ghetto Mining? Could we consider KNC as ghetto style considering it's scale? Rest assured it's probably not far from Facebook's pricing, after all the KNC site was built by the same people that built Facebook's place (The Node Pole). Here's some more info and an analysis of the KNC and other mines: https://www.googledrive.com/host/0ByWHHc0u_thNMWtQeDNiT2duU0E/Analysis_of_Large-Scale_Bitcoin_Mining_Operations.pdfNaturally, businesses need to make money, so we are all getting charged what it costs plus some more. Edit: if you are wondering, some of the above data is from here: http://www.grcooling.com/data-center-cost-savings/http://blog.schneider-electric.com/datacenter/2012/02/16/how-much-should-a-modular-data-center-cost/Dude, I'm not forgetting anything. All I'm saying is mount the Bitcoin mining chips (on small PCBs) directly onto your secondary cooling loop. There are no patents involved in that and exactly zero has to be paid to 3M or any other "engineered fluid" vendor. But it would also zero your commissions, so I fully understand why you keep pretending that you can't understand the difference. Bitcoin mining is not a "data center". It is an "embarrassingly parallelizable" problem where speed-of-light or any similar limitation do not apply. |
|
|
|
this is Scotland , after all  In Scotland probably at least some people have heated bathroom floors, where the warm water is circulated inside the floors all year round. Imagine that your Bitcoin miner instead of rather needlessly heating your garage could make your bathroom nice and warm to touch. |
|
|
|
aren't AC DT testing a few different chips? not only AM.
Allied Control will try anything, all possible chips, all possible ways of deceiving and all possible ways to bribe. The short of it is that immersion cooling is nearly pointless for Bitcoin mining. Just mount the mining chips on a small PCBs directly onto the "secondary cooling loop" from their design. Then the "secondary" becomes "only". The issue here is the sales commissions from 3M for flogging engineered fluids with astronomical profit margins. |
|
|
|
But please do describe to me how you will be competing with (possibly immersion cooled) equipment with the best W/GH ratio and lowest energy prices on the planet (letīs say 5ct/kwh) by the end of 2014.
The immersion cooling is just about nearly pointless with Bitcoin mining. The only thing that its proponents bring to table are workarounds for lack of proper design and engineering. Immersion cooling with 3M engineered fluids only makes sense if there are real constraints on the cooled equipment: speed of light limitation, unusual complexity, high density, etc. None of that really matter for Bitcoin mining. If your goal is low PUE then simply use typical evaporative cooling well known in chemical process engineering and refrigeration. Most of the required equipment is already available, all you'll need to do is bring a chemical/thermodynamic process engineer to the table with the PCB design engineer, who is already working for Spondoolies. Together they need to design a proper heat exchanger for the mining chips and the associated buck converters. I already asked Spondoolies if SP30 motherboard was designed for cutting into 15 smaller boards. Apparently not, probably because the PCB design engineer wasn't given this as a requirement. So they cut the heathsinks into 15 smaller pieces. They need to conceptually invert that design: single large liquid heat exchanger and multitude of small PCBs hosting the hot chips is the way to go long term. 3M's profit margins on their immersion cooling fluids are so high, that they require scientific notation to fit in the spreadsheets. Just use common chilling/refrigeration equipment instead where the profit margins are in expressed in two digits. |
|
|
|
It could actually be a group of people who all have the best of intentions and there is still little to no chance that it will succeed.
Highly unlikely that anyone there has "best of intentions". The way that the lead figurehead avoids, deflects and attempts to redirect any sort of technical questions is very telling. If you had combined 120 years of experience and even "moderately neutral intentions" someone from that team would be able to post some information how their product is unique, all without disclosing proprietary information. Especially something about "redundant engines" and why they improve mining performance. Remember the old maxim: "extraordinary claims require extraordinary evidence". Thus far you've received two references to college textbooks. So this is probably a "virtual corporation" (a.k.a. "front") where the key technical information is in the hands of subcontractor and the subcontractor has not been paid yet. |
|
|
|
Not sure I understand what 'fragmentary results' you would like to see?
It would be best for you if you've reviewed the posts from bitfury. It was his first ASIC and he went directly into full custom. See how he dealt with disbelievers. You can just read his English posts although his detractors were mostly posting in Russian and his responses were in Russian too. Basically, we want to see that there is a real technical depth in your organization. What we don't want to see is another Cointerra, where the "dream team" was really part-timers that quickly disassociated themselves from the CEO and the only technical person left was some German guy on a temporary visa in the USA. Your post, while quite good overall, essentially looks like a case study done by an MBA student. Edit: You can also review Spondoolies-Tech posts. The forum account is mostly manned by a their CEO, who isn't afraid to admit that he doesn't understand some technical issues, but is always capable of delegating the Q/A to an appropriate person within his organization. Edit2: some grammar fixes |
|
|
|
|
Show Posts
