|
The first 2 I got from Bitmain have 12 PCIe connectors. Also no provision for DC on/off unless there is a header inside. Now they all come with 10 PCIe and a DC on/off pigtail plug w/jumper on it.
|
|
|
|
|
Then it becomes a write-off. Hell, stuffing the DC with 28nm chips could be part of their R&D budget...
|
|
|
|
Why would you spend money on filling a DC with older gen chips especially when you will receive the newer chips with MUCH better efficiency beats me. You either have money to burn or you will simply build a new DC for the newer chips or you simply pay so little to do this that you can afford to replace everything in 1-2 months. I can understand filling just a part of it for tests, but a whole DC...something doesn't adds up. Just my take on it but: Because one has to start *somewhere* and get the ball rolling. The older chips were on-hand and they have working boards for them. As a bonus they get to test the DC at full load. Newer more efficient chips/boards are in the pipeline but not on-hand so you either have a DC sitting there empty/partly used or ya plug in what you have right now and upgrade when the new designs get into full production. |
|
|
|
I was thinking more along the line of TDk's PFE500S-48 http://www.us.tdk-lambda.com/ftp/specs/pfe.pdf good for 48vdc @ 504w each. Given multiple power feeds to the strings, use as many as needed per board. They'd love being in Novec next to the hash boards... Power bricks like that have been used for telco and such for decades (can you say easy and near-instant battery backup protection?) so there should be a shitload of them available as pulls not to mention new. It sure would work, but it is an overkill. The ultimate mining power supply is essentially a welding rectifier with some more ripple filtering and arc-start disabled. The normal welding rectifier have some very coarse regulation operating at line frequency (50Hz or 60Hz), not in the kHz or MHz ranges of precision voltage regulators used for non-power electronics. With such power supplies your at-the-wall GH/J will be the same as at-the-chip GH/J within the normal measurement and process tolerances. If any home experimenter reads this message: I'll reiterate the necessity of disabling the arc-start circuitry of the normal welding rectifiers. They intentionally produce higher voltage when the output current is near zero, then rapidly drop it to the nominal output voltage once the electric arc starts. Definitely ja. For home users battery chargers not so great either on single phase. 3-ph to DC and things get pretty rosey real fast with ripple dropping like a stone depending on the secondary side configuration and very stable. A delta-wye bridge on 3-phase 60hz give 720Hz ripple at ~ 0.5% pk-pk with no filtering whatsoever. This has not gone unnoticed by data center power providers pushing the high-DC feed to racks solutions. |
|
|
|
I like the 48vdc string feed idea. Considering that is very common telco power it makes for using fairly low(ish) cost but utterly reliable POL bricks to drop higher voltage ac/dc feed lines to board usage levels
Yeah 48dc is smart idea. Plenty of good efficient 48 watt psu from mean well for instance I think you are both missing the point of the 48V DC supply. Somewhere between 25V and 50V there's a point where power no longer needs to be regulated. It is sufficient to just rectify and ripple filter it. With lower voltages you cannot risk it because accidental overvoltage will permanently destroy the oxide layer on the chip. With series/string implementation backed by some sort active voltage divide balancing you get enough oxide layers in series to be no longer afraid of surges. There's enough margin between the normal operating point and the breakdown voltage. Nobody cares for under-voltages or sags. They just cause momentary increase of erroneous results. The only thing that requires regulated power supply (and uninterruptible power supply) is the mining controller. But it has negligible power requirements compared with the hashing engines. Remember that coin mining equipment isn't really a computing equipment or telecommunication equipment. You can reset it as often as you like and you never store any information for more than milliseconds. No I got that part of the equation. Look into the LED strings inside of any high-power lighting LED chip (clear lens, not ones with phosphors, can't see through it) from Cree or Phillips and you see my work there which hit the factory floor in 2007 giving the ability for fit (currently) just over a dozen emitter dies pushing over 25W total into a chip package >2mm 2. That in turn with the already known to begin with advantages of any series circuit and the major power LED makers were in heaven to start pushing the lighting biz to new limits. Got am inexpensive LED bulb from Cree, Philips/LumiLEDs, Osram and no doubt others or one of the great blindingly bright LED flashlights boasting a 1-5w Cree emitter in it? You're welcome.  Yes other tech advances were involved with gazillions of Other Peoples Money over the years leading to them but my bit in the chip package mfg'ing equation opened the door for it all to come together relatively inexpensively.. Anywho, back to the topic at hand, and question for me just brings up, how they protect against chip failures? High current Zeners in the chip package (not on die)? Since as you've said, the actual miner logic needs are pretty damn small, that leaves the door open for more to be happening in the chip package but off the (now 16nm node) ASIC die real estate per-se. Say just using the die real estate to also hold a safe operation fault logic controller for the off-die fault switches. Perhaps active internal power MOSFET switching to take chips out of the string or other failure protection mechanism? Ease of very low tech battery backup until the diesels kick in is just icing on the cake for this app. Very easy availability of the power bricks is another. |
|
|
|
I like the 48vdc string feed idea. Considering that is very common telco power it makes for using fairly low(ish) cost but utterly reliable POL bricks to drop higher voltage ac/dc feed lines to board usage levels
Yeah 48dc is smart idea. Plenty of good efficient 48 watt psu from mean well for instance I was thinking more along the line of TDk's PFE500S-48 http://www.us.tdk-lambda.com/ftp/specs/pfe.pdf good for 48vdc @ 504w each. Given multiple power feeds to the strings, use as many as needed per board. They'd love being in Novec next to the hash boards... Power bricks like that have been used for telco and such for decades (can you say easy and near-instant battery backup protection?) so there should be a shitload of them available as pulls not to mention new. |
|
|
|
Fair point on Cheap hyrdo power in China eg the 3 Gorges Dam but frankly, that vast majority of overall power production production is spoken for. For most heavily industrialized countries, for just one segment - aluminum production - takes one helluva chunk. ref: http://wordpress.mrreid.org/2011/07/15/electricity-consumption-in-the-production-of-aluminium/"According to Alcoa, the world’s largest producer of aluminium, the best smelters use about 13 kilowatt hours (46.8 megajoules) of electrical energy to produce one kilogram of aluminium; the worldwide average is closer to 15 kWh/kg (54 MJ/kg). Worldwide production of aluminium in 2010 was 41.4 million tonnes. Using the figures above this means that 621 billion kilowatt hours of electrical energy were used in the production of aluminium. To put that in perspective, the total world production of electrical energy was 20261 billion kilowatt hours, meaning that more than 3% of the world’s entire electrical supply went to extraction of aluminium." Toss in the other more pressing power needs of China and you get my point. Ja certainly a lot of low cost power will be available to local (ish) peta-farms but not as much as you think. Outside of being right next to the power plant/dam (the now closed Alcoa smelter site in Washington state comes to mind) the infra over there just does not support someone suddenly deciding to run 25-50MW much less 100MW or more without a helluva lotta prep and time/cash for it to be done. |
|
|
|
|
I like the 48vdc string feed idea. Considering that is very common telco power it makes for using fairly low(ish) cost but utterly reliable POL bricks to drop higher voltage ac/dc feed lines to board usage levels
|
|
|
|
Not sure what we are betting on but I will definitely take the bet-
So I will order 50Th-100Th the first day the miners are announced, if they arrive to me before April 1 I will pay you 1 BTC but if they do not then you send me 1 BTC.
How does that sound?
Don't forget to add, "In good working order and according to specifications." Should be a given but also add, "On promised delivery date at time of order verification (eg. yer wallet says the Tx went through and has a couple verifications or check/other fiat xfr was approved) subject to strictly defined definition of "Late/delayed" and resolution process if late or delayed" This comes to mind on both points: http://web.archive.org/web/20131003021440/http://advancedminers.com/bitcoin-mining-hardware/520-ghs-bitcoin-miner from Oct. 3rd 2013. Ordered mine Feb 28th 2014 when it was not pre-order but instead "delivery in 4-wweks or less"... We all know how the AMT/Bitmine.ch A1 saga is progressing for most folks...  |
|
|
|
Ja. Looks like fun for testing. My only concern for long term use is the bearings in the fan. 'Prolly dunna like running in what is essentially a vapor degreaser  I assume that was to keep tank pressure at a minimum by pushing as much vapor as possible through the small radiator in it. I take it that's why the data centers setup looks to just use the natural and I dare say rather violent vapor flow in the system for circulation along with heap big vapor-to-water heat exchangers in the tanks. Nice. No moving parts in the data tanks. |
|
|
|
! Thats exactly what I want to make! But.. 3M won't sell me the fluid! Can you hook me up with Novec 7000?  How are your temps with the 7100? It boils at 61C, so I guess they are high?  Wouldn't 7000 be better for high powered ASIC chips, as it boils around 32-34C, if I remember correctly? Won't sell you any? Then I assume you are not a registered company. Most industrial use fluids/chemicals/etc. will not be sold to the general public unless the manufacturer specifically allows it. Stuff like this is not only expensive but in many ways a hazardous substance so they need some assurance that if something happens they are not the 1st target in a cleanup cost liability suit. |
|
|
|
" http://www.theguardian.com/technology/2015/jan/16/elon-musk-falcon-9-rapid-unscheduled-disassembly" heh heh heh... I like that... Frankly, I like that BitFury seems to have take a proper approach to marketing this chip vs the AMT/BFL/Hashfast days.... They know the market is there with the unclean masses slathering at the gates to get their hands on machines. For now they are tossing out tidbits but still keeping the gates locked. Don't forget that making the chips is only half the battle, now they have to be characterized and a reference board design made. Note the emphasis on reference. As in here is an example of how it should work, not 'copy-this-x-times' like Bitmine.ch tried with Inno's ref design. Only once that is done can integrator's begin work on a full hashboard design. It takes time if done right. Sure the market screams we want it yesterday! Tough. I like this better than weeks/months of folks bitching about promised product that is weeks/months late or never to be made at all. |
|
|
|
Another good hope is a group gets together, works out a large chip order and gets a board designed around 'em that can be manufactured as soon as chips arrive.
and at least BitFury or their integrators haven't taken the pre-order and give us oodles of cash for our fabulous miners NOW! route with the general public. At least none that I know of... |
|
|
|
<snip>
Bitcoin mining is full of chips designed by barely competent CAD monkeys. The access to capital is not sufficient to produce a worthwhile chip. It is the access to the required knowledge that is the differentiator.
<snip>
I've been there and done it. Many ASICs are designed and manufactured just for the purpose of copy protection and design security: to make coming to market with competitive product too expensive by eschewing anything off-the-shelf or available from a second source. From what I understand such nearly-empty chips are nowadays even more popular to hide the patent infringement amongst the completely bogus chip-space-filling circuits.
Agreed. In my dealings long ago with the now infamous Joshua Zipkin of the AMT/Bitmine.ch mess, he mentioned that at the time there were only 2 people that led the designs of chips from ASICMINER, BFL and HashFast. He dinna name names but did say they collaborated and mainly used the different chips to test out their ideas on How To Do It and sold said companies on producing them. We know how well that went... BTC miner maker Cabal conspiracy theory? Don't know if it was true but ya gotta admit, all 3 companies by and large had the same issues in trying to get miners for the general public out the door . Considering chip performance they seemed more concerned on as fast as possible getting a chip that worked at all vs one properly tweaked to work well and to the pre-order specs. re:IP protection, FTDI and their USB-RS232 converter chips comes to mind.One tweak they did to combat the flood of generic microcontroller (and cheaper) based converters was add an ID code and got MS to look for it when you run Win7 and higher. If not properly signed Winbloze refuse to use it. Grant you many of the counterfeit usb-rs232 adapters were crap and FTDI's *do* work better but still.  |
|
|
|
I don't think we've communicated. I wasn't talking about a separate, discrete chip. I was talking about using a little free space (and maybe pins) on some larger chip with completely different purpose and market.<snip>
I agree on that point. The hard part would be to get the chip customer interested in bundling a mining circuit into said other use chips. I'd LOVE to see folks making chips for IoT jump on this. Last I heard the market will be several billion IoT chips this year. Shades of the BitFury light bulb or Bitmains AntRouter... Cree and Phillips are already working on their LED bulbs acting as a light-based wireless link along with a WiFi circuit in them for lighting control. One step further is their using POE to not only communicate with the luminaries but also to power them. Gets around needing electricians to install the wiring and code restrictions on how/where the wiring is ran. To me that screams to put a miner in there! Of course, as a single bulb/miner it would hardly be worth the time to even setup a wallet for the 'income' but if you are talking about an entire building full of lights... Hmm. |
|
|
|
You thought correctly. I don't do chip design, I'm purely a hardware guy. However, most of the companies my company supports *are* in the semi industry ergo my knowledge of where the chip mfg processes comes from. Our end of things is tightly integrated to what size nodes the Foundries can produce. That said, I can tell you it is a fact that you are looking at several million $ to develop a chip from scratch. Even when using foundry or design house supplied pre-made IP blocks, TSMC will not even talk to you about 16nm chips unless the chip quantities are over 100k/month... Push back to 20-22nm and higher and they are much more approachable but that is also outside our target.  Here's my take on the cost: The hashing engine is so small that one could get it developed and samples manufactured for free in the small corner of some bigger project. The mega-dollar capital outlay would be required only in the final stage of mass-production. The semiconductor foundries routinely "waste" much more chip real estate on various testing structures. So why nobody did that? My limited experience shows that the prospective market is so filled out with liars, bullshit artists, con men, mythomaniacs and associated naïfs (that are honest, but completely indistinguishable for the former classes) that it is completely not worth the effort. The only way to do it is as an intellectual curiosity (in secret) or when from the start you plan on ripping off the fools. Please don't go blaming the messenger. This is the reality. I'm not going to claim that I know everyone in the Bitcoin milieu. But I've known enough to come up with some generalizations. Overall, a spot-on assessment. On 'spare' or wasted Si real estate, ja it's there and available for sample runs. If you have a design already laid out that is. Then the only problem becomes getting the foundries to be interested in running them. That simply will not happen unless you convince them that the sample runs will eventually translate into full wafer production. A good example is from one the customers we did R&D for regarding RFID tags for lab animals. The tag itself is a very common chip with nothing new except for size (needs to fit on the ear of 1week-old lab mice) and packaging material (encased in ceramic to resist chewing). Even with the chips using 22nm junctions they had a devil of a time getting a foundry and packaging house to make samples until they pointed out to them that the the market for the tags is well over 100million/year. THAT got their attention and said tags are now in production. |
|
|
|
ask sidehack or NotFuzzyWarm they both do design etc and can give some kind of ball park I think . yea we can't leave out the Home miner , i think even if i didn't mine my self at home, i would agree to.
I thought they may have only worked with chips other people engineered and designed. I didn't know they too can engineer and design chips themselves. I'll give them a shout to ask some questions. You thought correctly. I don't do chip design, I'm purely a hardware guy. However, most of the companies my company supports *are* in the semi industry ergo my knowledge of where the chip mfg processes comes from. Our end of things is tightly integrated to what size nodes the Foundries can produce. That said, I can tell you it is a fact that you are looking at several million $ to develop a chip from scratch. Even when using foundry or design house supplied pre-made IP blocks, TSMC will not even talk to you about 16nm chips unless the chip quantities are over 100k/month... Push back to 20-22nm and higher and they are much more approachable but that is also outside our target. |
|
|
|
Right, but if that heat could be used by something (like piping to neighbors in wintertime, or running greenhouses or something) you could increase the overall profitability of an operation.
Of course. I'm just saying it isn't that easy. I do use the miners I keep at home for heat during the winter, I haven't used my actual heater once this winter. I think that is the most common use of miner heat. But I agree, when possible the heat should be used, I just can't think of that many ways to use it, especially in places where the ambient temperature is already relatively high. Well, here's an idea I've mentioned before: Use water cooling for the miners with the heat going into large spa pools... |
|
|
|
Given the diff jumps and the cost to make the chips/boards I don't think we will ever see these chips in the wild. Bitfury will be loosing money on selling chips at a price that people will pay soon. They came in a little late with this Gen and I think their margins after R&D and all are really small now, if not non-existent. Oh the tragedy.
They planned everything very nice. They released the chip demo videos at the same time when they were already populating their DCs with these chips in order to have the false impression that they barely got the chip. Then after they managed to get all the pre-order contracts ready they started to fire up their new chips. Well played Bitfury! I applaud you! If so, at least it can be said that they didn't try an AMT/Bitmine.ch, BFL, HashFast et al approach of taking pre-orders from the general public. This way the pain of production ability/miner design timing falls on private investors. Looks like folks learned and if/when the chips become available to us a least we will know they work and hopefully have reliable reference designs for them released as well. Open-sourced driver code would be icing on the cake... |
|
|
|
|
Show Posts
