Topic: What would be the perfect ASIC Bitcoin miner? (Read 510 times)

This is a great idea! I liked the S2 rack case design back in the days.  

With the exception of liquid cooling, that sounds a lot like the specs I was thinking of for a rack miner around the middle of 2015. I've been slowly hammering on a project to make new boards that fit on an S1/S3/S5 chassis (practical limit of about 250W per board air-cooled) and was thinking about a rackable case design that could integrate about 8 of them, possibly with external PSUs. I'm looking at USB interconnectivity, which removes the need for a lot of proprietary control hardware.

Of course, with S1 formfactor boards that opens up C1 waterblocks for a liquid-cooling solution where you could fit about 16 of them in a case without a lot of custom work.

Okay so current miner looks like this.

1) Interchangeable hashcards allowing for ASIC upgrades and possible addition of other type of mining hash-cards (X11, Scrypt, AI)
2) 4U custom case allowing for better hash-board to controller ratio saving on cost
3) Liquid Cooling with or without radiator (centralized farm liquid cooling)
4) Decent GUI connected to a semi-future proof SoC (Zynq Ultrascale+?) with sufficient FPGA logic
5) Ability to have redundant power supplies (rear wall of power supplies in the 4U case)
6) 7nm ASICs designed for 90 degree F ambient temperature with on-ASIC temp control logic with ability to optimize for efficiency [J/(GH/s)] or for max operating temp (speed)
7) What else?

.. not sure about what would be the perfect ASIC Bitcoin miner, but it should only have network and power connections.  Not some oddball USB/RasPI 'controller' to futz with.  A good GUI is a must, CLI/API probably a must also.  The chassis should allow use of other than Bitcoin hashboards too.

A datacenter-type rackmount case has to be more expensive that what would be needed for a table-top or 'milk-crate' setup.  Might be nice to to offer either case option.

In my humble opinion the perfect miner is scalable and affordable. I've posted in another thread about my small group's desire for a smaller "cube" style mini miner type product that's 4-5 TH perhaps, but done affordably so it's easy to scale... and also manufacturable in quantities that can get them out to the masses. It'd be nice to grow a farm 5 TH at a time at around $500-$700 a box. Even if their not the most efficient unit on the block, being able to expand your hashing power when you want vs. being beholden to the big dogs or the eBay market would be liberating for both small time and home miners, and those who want to scale something a little bigger.

WiFi option, even cooler. And get it as quiet as you can. A smaller/quiet box... around 500 watts each... home miners can stick one in each room... or 3-4 in the garage, etc. A chicken for every pot, so to speak.

For me, it depends on the market you are trying to target.

If you really want datacenter-type gear, I'm a big fan of blade systems.  Even from the old days of the IBM HS20/21 blades all the way up to the modern Cisco B-series stuff, it's super convenient.

House all the management and intelligence in the chassis, and all the power/connectivity.

Blades would just be additional hash modules.


Probably not feasible, and I'm sure expensive, so it won't happen.

As for individual rackmount systems, 2U or 3U is fine.  No idea where these 44U racks are, every datacenter I've been in has been 42U racks.  Since systems/network are what I do for a living (as a consultant), I've been in hundreds of datacenters in just the past few years.  Can't recall seeing a single 44U rack, though I've seen plenty of weird 21U floor standing and 11U & 7U wall mount nonsense.  That said, with the exception of SANs (which are all sorts of different sizes) and large core switches, damn near all gear these days is 1U or 2U.

Redundant hot-swap power, and if you're going to have a CRAPLOAD of hash per box, redundant network as well.  It would make me sick to think that I've got 100TH in 3U and a bad NIC took it out.

That's right.  The 3U is a specialty form and when I went to source those there are very few off the shelf.  4U seems to be more popular.

The S9 shoebox is a good design as it crams the components together so there is a no pathway for the air to flow other than through the ASIC heatsinks.  Given this, some sort of dummy cards may be needed if the hash-board base in a 3U or 4U chassis isn't filled out with cards to make sure that the air flows over the cards with ASICs.

The 3U is the minimum size to put GPU and may help make sure that the air flows through the ASIC hash-cards if we were to go with standard GPU sized cards.  If we went with a 4U then some metal spacers may be needed on the top to sit flush with the ASIC hash-cards guiding the air-flow that way.

Any chassis that we use will be most likely stamped special order because of the cooling requirements and the associated fans.  So the sourcing shouldn't matter too too much.  As long as the production run is in the 1000's manufacturers will usually work with you on the design aspect and then we would have to pay for tooling cost after design.

Another thing I can see being something that will need to be worked out is the heatsink itself.  The 16nm BM1387 heatsink on the S9 is already like 100x more volume than the packaging of the chip itself.  Assume that you kept the same die size for the a 7nm you are going to need twice the heatsink as the size decrease of transistor outweighs difference in power efficiency.  Increase the size of the heatsink or increase airflow or god-forbid, slow down the chip.  

Going to a much smaller die than the BM1387 is a challenge as it is already pretty small.  Because of that and the size of the associated heatsink needed for 7nm it may get more challenging to increase the density of hash-rate per volume of server.  Hopefully power efficiency will offset some of those effects.

Its also good to hear that people went much higher than 4kw.

The last Bitfury box I remember seeing was 6U and I "believe" had 4 power supplies in a "3+1 redundant" sort of setup, but it's been a while and I might be misremembering on the power supplies.

The Spondoolies SP50 had something like 11 power supplies in either a "n+1" or "n+2" configuration, totalling 15-20 KW of active power supplies - but that was in a custom 11U case.

3u is kind of an odd size for rack mount gear - I've seen some around, and the cases are fairly common for servers, they're even tall enough I *think* to mount standard-size GPUs, but they tend to not come out "even" in the fairly common "44u" racks a lot of data centers use though they'd be fine in 48u (less common) or 42u (common) racks.

This is a link to Google's ASIC TPU2 project.  Just seeing how other ASICs are being implemented.

https://www.nextplatform.com/2017/05/17/first-depth-look-googles-new-second-generation-tpu/

Thinking it through it may be possible to get higher density of hashrate with a traditional server chassis and shave some cost on controllers at the same time.

Probably need to upgrade the Xilinx Zynq SoC-FPGA to accomodate the extra hash boards, but that won't offset the saving of replacing 1 board with say 3.

You can still shelve the 3U chassis and it allows for that server rack option whereas shoebox not so much.  

The power is the issue that comes up.  Do I trust a box with ~4000watts running through it?

I originally looked into the 3U chassis because it allowed for interchangeable hash-boards making upgrades possible.

 Low rig density.

 Current ASIC miners pack 1200+ watts into a case that is less than a QUARTER of the volume of most standard 4U cases, and less than a THIRD of most "shorty" 4U cases.

 I'm pretty sure the last rack-mountable designs were something BitFury or possibly the Spondoolies SP50 - and folks were moving away from the rack-mountable form factor for a couple years before THAT.

 Miners are NOT designed or intended to run in traditional data centers any more, as those data centers don't have the INFRASTRUCTURE to support the power requirement of mining.

 There have been temp sensors built into the ASIC chip on some previous designs I think - pretty sure the Spondoolies "Rockerbox" chips had on-chip sensing.

I would say changing the voltage outside some parameters would void the warranty.

I love the idea of swappable hashboards and a 3U form factor. I don't anyone moved away from the server form, I think it was more efficient to pack more hashpower in a smaller form factor, both from the air cooling aspect as well as fitting more machines in smaller spaces.

While the common wisdom is that miners are made to run in data centers, I think the reality is that they're made to run in air cooled warehouses (and not temperature controlled data centers) moreso. The shoebox form factor fits that much better than data center rack concept.

Do you'll know why they moved away from the standard server chassis?

Whose decision was that?  It seems to limit ability to upgrade the miners and may have been a business decision.  Another reason is the wind tunnel with the two fans, but that could be replicated with walls of fans in the 3U design as shown in original picture.

Temperature sensors and temp control logic may need to be built into the ASICs themselves in 7nm because of the increased power density.  It might be helpful, if it doesn't effect the performance, to have the ability to retrieve that information and then post that to the GUI.

As far as clock speeds and voltages, how would warranty work then?  An unlocked product without warranty?  
That could be a possibility if you could easily swap out hash-boards.  Basically there would be a warranty on the controller and so on but not the unlocked hash-board?

Configurable voltages! The ability to adjust the frequency of a board chips and it's timings.

If I want to feed my miner 1500w, even though it only needs 1200w, it should be able to be clocked higher.

Better thermodynamics - the 100ish chips per board w/heat sinks and a high static pressure fan works in the current shoebox form factor, but why not other form factors?

Coming from a server background, and because this is a wish list I'd like the fans and power supplies to be redundant and hot swap-able. It'd also be nice to have mirrored internal SD cards to run the controller OS. Some sort of out of band mgmt. port (like an iDRAC or RIB) would be nice as well. Along with sensors not only on the chips but various areas of the chassis and an intrusion notification switch. Those latter items would be really nice for colo or remote locations. Front panel status LCD would be nice, but I am sure my list is turning in to big $$$ by now. Anyway for mounting I would want rails rather than a shelf mount.

That's all I can think of for now.

Bitcoin datacenters do not use traditional server racks, putting them in a server case would serve little purpose since infrastructure would have to be changed within the mines to support the change.

There used to be this style (Antminer S4 and Cointerra come to mind) but the industry moved away from the design for a reason.

We are US based and will have assembly/support from the States.

As suggested I edited all of the bounty/token stuff on Bitcointalk except the mentions in the actual whitepaper posts.  We already turned down the VCs, and are committed to decentralizing the Product Development of ASIC Miners.  The development is a serious undertaking involving companies that have the best track records.  Its the chicken or egg right?  You want to see a miner and hold it and test it, but right now there are multiple groups privately planning and making 7nm ASIC Bitcoin miners for personal use and they will own the mining market if this doesn't move forward.  The market capilization of bitcoin has increased to the point that it makes sense for the bigger mines to produce their own 7nm ASICs leaving no chance for smaller independents to exist.

If you'll have any input with regards to form factor, layout, interchangeable ASIC hash-cards we would like to know.

I wouldn't worry about the form factor, but start focusing on your community presense.
I mean, if i see a newbie account posting about a token sale, claiming to be able to produce 7nm chips out of the blue, i get a bit suspicious...

If i were you, i'd drop the token sale, it makes you look fishy... Then design your ASIC, give a couple demo units to high ranking community members so they can test the hell out of them, then open an official company in the EU or the US, where the buyers are protected by the law...

If you do this, i'm pretty sure you can put your hashboards with 7nm chips in an old shoebox, and you'd still be able to sell them... But if you're just here (as a newbie) to get your token sale going, i seriously hope nobody will fall in the potential trap.

Just my 2 satoshi's tough...

The perfect miner would be from a perfect company that actually cares about its customers and would actually send you a miner when you bought it.

What would be the perfect miner?

I am working with a group build ASIC miners based on 7nm architecture.  At the current moment our thoughts are to implement a more traditional server design.

Proposed form factor would be a 3U or 4U chassis.  These  sizes allow for interchangeable hash-boards similar to the GPUs on a PCI board.

Some benefits for 3U ASIC Server with interchangeable hash-boards:

1) Server chassis and main controller could be re-used
2) Potentially multiple alogorithms hashing within the same chassis
3) Allows for standardized server racks to be used

Any feedback or alternative designs.  A simple mock-up of potential layout is below:



Or is this just too much?