There are never any bitcoin miners available to purchase and whenever Bitmain (the leading bitcoin mining manufacturer) releases a new batch of miners they all get gobbled up in seconds. For those who don't know Bitcoin mining requires the use of computers to solve complex math problems and the reward for solving the problem is bitcoin. Bitcoin mining evolved from CPU processors to ASIC chips and it seems only a few companies have been able to create their own custom ASIC chip. Is it really that cost prohibitive to manufacturer your own custom ASIC chips for bitcoin mining? The demand is much greater than the supply. How does one create a custom ASIC chip with the sole purpose of mining for bitcoin and how much would it cost? I read some answers that were 10 years old in regards the the cost to creating ASIC chips but I'm hoping costs have come down since then. Hoping someone can shed some light on this subject, thanks in advance!
-
2\$\begingroup\$ The manufacturers capable of designing and bringing such an ASIC to production usually have better ways to earn money than mining bitcoin. \$\endgroup\$– Eugene Sh.Commented Apr 14, 2021 at 19:52
-
3\$\begingroup\$ @EugeneSh. I think he's wondering how much it would cost him to make an ASIC, not "Why don't ASIC houses make bit coin miners to sell?" \$\endgroup\$– AaronCommented Apr 14, 2021 at 19:58
-
1\$\begingroup\$ @Aaron Yes, but my point is that if one has the resources to produce such a chip, then they probably won't mine enough to cover the costs. \$\endgroup\$– Eugene Sh.Commented Apr 14, 2021 at 20:00
-
3\$\begingroup\$ If you expect more performance than high end multi GPU card, what costs differences would you expect between the latest NVidia CPU and a typical Altera high end ASIC with lots of libraries. \$\endgroup\$– D.A.S.Commented Apr 14, 2021 at 20:19
-
1\$\begingroup\$ @TonyStewartEE75 Especially with NVidia dedicated mining chip. \$\endgroup\$– MapleCommented Apr 14, 2021 at 20:49
3 Answers
An FPGA would be my first choice. FPGA logic design requires specialized skills if done properly. This could be done for under $5000 for a prototype (parts and a circuit board) if you do the PCB layout and coding (i.e. coding costs zero). If you need to hire an FPGA coder, that could run USD 50k to 100k, perhaps more. Coding an FPGA is very different from coding a CPU since you are doing logic design.
If doing an ASIC, you need volume to make it worthwhile, perhaps quantities in the thousands. Full custom ASIC will probably cost you around USD >1.0M for design, layout, and a few prototypes. Most likely, you would test the design out on an FPGA before committing to silicon.
-
\$\begingroup\$ Good point for FPGAs. Very often designs towards ASICs actually start from them because trial and error is "free" (except for HW engineers work time, of course) whereas it can cost as much as a brand new design on ASICs (i.e. hundreds of thousands, millions of $ too). \$\endgroup\$– edmzCommented Apr 14, 2021 at 20:59
-
\$\begingroup\$ FPGA flow is different when using them as the end-product vs. as an ASIC prototyping tool. An ASIC prototype flow is for validating logic optimized for the target process library, whereas the FPGA-as-product will try make maximal use of the FPGA resources. There will be things that you can do in ASIC that don't model efficiently in FPGA, but nevertheless will result in a higher-performance chip. \$\endgroup\$ Commented Apr 14, 2021 at 21:07
-
\$\begingroup\$ For a particular task of mining, considering cost of power to run FPGA-based miner per acquired revenue, FPGA becomes not the best choice. That's the primary reason to use ASICs, and not one-time engineering cost. \$\endgroup\$ Commented Apr 15, 2021 at 9:49
tl; dr answer: Hundreds of thousands for a simple chip on mature technology, to multiple millions for an advanced process with sophisticated IP blocks.
For a chip destined for production, the steps include:
- logic design and verification (usually behavior model and FPGA prototype)
- ASIC logic synthesis, test insertion
- physical design
- timing closure / AC signoff
- Tape-out, masks
- package design, tooling
- prototypes (preferably on a shuttle)
- validation (functional test, parametric test)
- test vector and fixture development
- reference designs
- software support
Not to mention the EDA seats (Synopsys, Cadence) and IP licensing you might need.
You can outsource much of this to an ASIC house who, given a proper spec and reference model, could produce a complete chip for you. Or you can do some or all of this in-house. Given the nature of your question, seems like you'd need a complete turn-key solution, for which you will pay a premium.
With reference to How much does it cost to have a custom ASIC made? -- What part of the costs would have come down in only 10 years? There are still substantial costs of materials handling (and hazmat disposal) and development costs. Time hasn't gotten cheaper either.
Not only that, but a wafer fab requires a lot of expensive pieces of equipment and a supply of some very nasty chemicals, and requires lots of dedicated infrastructure. It's expensive even when it is standing idle, which is hopefully never...
Consumer products experience price reduction over time because the NREs get amortized (accounting translation: Non-Recurring Engineering costs are incurred at the start of the project, but rather than say the first unit costs $10M and the second and subsequent units cost $1, the accountants instead spread the NRE costs over some initial period. So after selling 10k-100k-1000k units, it's figured that the NRE costs have been recovered by that time, and prices start to come down). But the initial NRE is still substantial for any new product development.
There are also "house label" projects that might be requested by one large customer, who by paying the NRE costs, gets to demand that the product is not sold to any other customer. Often you see these mysterious chips in "white goods" like washer/dryer appliances and other embedded systems. Might be based on some general purpose chip with slight modifications and a custom label, or it might be something wildly different. Whoever pays the NREs basically owns that design and gets to control whether it is publicly available.
There's also currently (2021) an industry-wide capacity crunch partially due to Covid plus unexpected automotive demand. Just because "demand is greater than supply" doesn't mean that there is also pricing support for what it would actually cost.
