Design suggestions needed for new bitcoin miner. What NOT to do. What to avoid.

[MarketNeutral]

I want some input from experienced, long-time bitcoin miners: Considering all the ASIC miners you've mined with or experimented with…

What design flaws have irritated you the most?
What would you change about this or that particular miner?
Which miner made you think, "What were they thinking? Why did they do this?" or "I could have made a better miner than this!"
Issues that should have been non-issues?
Problems that could have easily been prevented?
Etc.

I'm not asking about poor communication from the mining companies or delays in shipping. I'm asking about hardware.

In other words, what mistakes should one avoid when developing a bitcoin miner?

I have purposefully kept the overarching question open-ended. Your answers could be in regard to any annoyance or frustration, from a misplaced heat sink or shoddy USB, or to more low-level components such as the chips themselves. Anything, just tell me your grievances. I want to learn. I don't want to make the same mistakes.


Quick background:
Over the past year, I've been researching bitcoin mining hardware. I found some investors, started a new company, and decided to cross the Rubicon and create a new bitcoin miner.

Except that our miner is not quite the same as the others. Although our company is a for-profit company, and we are contractually obligated to provide working units very soon, our mining research and subsequent design have aligned us much closer to the experimental, proof-of-concept side of mining. We're not aiming to compete directly with the traditional ASIC companies. We don't intend to sell a zillion units and overtake the ASIC market. This is an experiment to create a new class of bitcoin miners. It's more akin to the brief FPGA days of bitcoin mining than anything on the market today. We intend to change the way miners mine and the way they perceive mining in general. More on all this later. When the sales pitch is ready and we launch, I'll start a new thread.

For now, we've spent nearly all our time and energy on optimizing the source code/microcode of the miner. It rocks. It's amazing. It packs some serious hashing power. And it's coming to market. We still have a ways to go, but we're getting much closer to the production stage. Thus I request suggestions about the final design. Of course, form follows function and absent technical specifications, I realize there's only so much one can suggest. Nevertheless, I would sincerely appreciate any input from any long-time established miners.

Thank you.

[torepia]

For the home-miner, its all about dB. I wouldnt mind a oversized cabinet/miner, if it helped on the dB levels.
A silent blade-server style bitcoin miner rack would be awezome. Lets say a 10-20U rack, half filled with current date hashing blades. And it would be compatible with future gen hashing boards.

Make my day!

[sidehack]

Quiet cooling
No chips over about 20W - nothing requiring multiphase VRM per chip, exotic cooling or extreme die temperatures
No BGA
Adjustable core voltages, preferably software-based
Accurate temperature sensing at the chip, rather than at the cold side of the PCB
Modular design - controller and hashing units separate, with non-proprietary interconnect (good for upgrading and piecewise testing)
If PSUs are integrated, use server-grade for efficiency and reliability
If PSUs are external, make sure to have a reasonable number of power jacks (for counterexample, see KNC Neptune)

[QuintLeo]

usb connections. Bloody unreliable long-term, nothing more than an IRRITATION.

Excessive power levels per unit.
The 600W ballpark the S5 was was fairly nice, gave lots of options for powering them.
The 1200+ of the S7 is way too high for a "home" miner.

Modular design actually tends to INCREASE the cost overall, though the flexability can be nice *IF* you can get the hashing boards seperately at a resonable fraction of the cost of an entire unit.

*IF* you are going to use an integrated PSU, use something with a STANDARD size, not something propriatary with very few options to replace it when it dies.
Better yet, do NOT use an integrated PSU, gives the end user more options.

Rack mount case sizing is ... not really a good idea for a home miner, unless you don't mind wasting LOTS of space on at least 3U form factor.
Cooling in 2U is a pain, cooling in 1U is a MAJOR pain, and the fans get LOUD for decent airflow at anything less than 3U (you can fit 120mm fans in a 3U form factor case).

Massive single chip miners are a pain, and tend to be very inflexable on usage of the chips.

 Don't even get me started on the very poorly designed power control circuitry on the Gridseed stuff - if you want examples of HOW NOT TO DO THINGS just look at everything Gridseed did (except chip reliability, that was pretty good). BAD board design to the n'th degree, the whole stupid "mine 2 different algorythms on one chip" idiocy that they have continued into the SFARDS iteration....

[MarketNeutral]

This is great feedback.

Keep it coming!

[sidehack]

If it helps, I agree with pretty much everything QuintLeo said. External PSU would, for the most part, be the best way to go because it allows the end user the most options - which includes using PSUs already acquired for retired hardware. Modular design with the provision for parts (especially hashing boards) purchased separately would be great, especially if successive generations used the same form-factor. If that, then make sure the interconnects between board and controller are the same as well so the controller can be reused with updates to software only. Rackmount anything should be at least 3U.

[AJRGale]

im with the irritation of USB, most development skimp on ESD protection from the USB ports, so the one odd day you plug your miner in, *pop* goes the controller. simple little 0.30c protection diode would do the trick.. iirc the U3 did have that on the data lines.

[toptek]

if your going to sell to the home user which is still a very good market to get into .


noise , weight , small , make it only use what Electricity is actually needed while it runs.


beadle to under clock and over clock and volt the same way.


fully customize able  fans in the GUI

use any controller even if it has own built in one



and test test even after it is released all the common sense stuff it should do .


be upgarde able use those old parts that are usable.

[Finksy]

Hashing density by optimizing design without creating cooling problems is a plus.  I always liked the AM Tube design, it could be cooled reasonably well in a push-pull configuration without too much noise, shipping was relatively cheap per GH/s for the time due to the heatsinks and frame being flat-packed and requiring assembly @ home, and because it used 4 boards/heatsinks arranged radially, you got some cooling efficiencies from having a sealed wind tunnel in comparison to S1/S5 designs.  I feel it was on the high side for power consumption @ ~800-900W, but all things considered I feel it is a good starting point for design (major improvement would include a shield to protect the PCB's and components on it.  Lost more than a few capacitors when the legs slipped through a wire rack...)