Mining Operation Cost Model


#13

The question is not whether ASICs can be developed. The question is can an ASIC be developed that has a clear advantage over GPUs, both current and next-gen cards, in terms of price/performance/power consumption. Then there's resale value ... you can't play video games on an ASIC after it's mining days are over. :slight_smile:


#14

Point well taken!!!!


#15

I think the ZCash team have said that they can change the algorithms in future which would render ASICs useless - which I think is probably enough of a threat for people to not invest in developing ASICs for ZCash.


#16

Out of curiosity, do you have a "below $X, it doesn't make it profitable to mine ZEC" built into your cost model? And what would you do if ZEC falls from it's current $37-$40 range to this

For me, this figure is about $22 per ZEC to break even. Above that, I make a "paper" profit. Below that is a "paper" loss.

I'm taking a long term view, so thinking that if ZEC continues to drop, even to $10, I could keep mining and then regain profits once the value goes back up.

With any kind of investment, profits are only realised when you actually sell. Otherwise they are "paper" profits.


#18

Out of curiosity, for those trying to do a longish term cost model, what did you estimate for the rise/fall in difficulty?

I was expecting an average 5% increase in difficulty each month - but the difficulty has dropped by about 7% in the last 2 weeks. Putting the new figures in makes the future forecast change wildly. I know a lot of this is finger in the air estimates of "what do you think ZECUSD rate will be at some point in the future" and no-one can predict that with any accuracy.


#19

The only thing you can say with certainty is that whatever model you, or anyone else, comes up with, it will be far from the actual reality that unfolds, and for reasons that never even occurred to you. This whole endeavor is, by design, a never ending crap shoot. Think about sitting there, last May, coming up with cost model projections for ETH. To say that there were a couple of surprises would be a massive understatement. Anything can, and probably will, happen. If you are operating under any other assumption, you're playing the wrong game, in my opinion.


#20

indeed lots os surprises possble, good and bad

if you start a farm you nee to play the long game

and make sure you have a good power contract, saves you a LOT!


#21

Yes I've been depreciating the cards. I'm not sure how often new cards will come out and what effect that will have on the hash capacity of the whole network.

I think of the network as all the GPU miners-- across all alt coins. Whatever coin is most valuable they will switch to it. So we're in a sense in competition with each other.

ZEC will continue to decline, I think, until there's enough capacity that it reaches its fair market value. I think it's still constrained by lack of availability.

One thing to consider is that you can depreciate your motherboard and power supply over different time periods... I think PSUs are often warranted for 10 years, and a MB should last you more than a GPU.


#22

If the expected ZEC is lower in value than your cost of mining, you shouldn't mine and instead of giving that money to the power company, you should just buy ZEC on the open market (if holding ZEC is what you want to end up with.)

I suggest everyone be prepared to mine other coins than just ZEC. While mining for ideology is commendable, the mining community is following what's valuable.

When a coin is more valuable than its hash power dictates, that implies its security is vulnerable, and more hash power is desired. When less, then too many people are mining it, and people should switch. IT's an active market, just as the exchange rates with BTC shift too.


#23

ASICs are not ideal for Equihash. This is why Equihash was designed that way. In fact, its designed to leverage DRAM which is already an ASIC-- eg a memory oriented chip. So noboy will have an advantage coming out with an "Asic" to do equihash, because the amount of R&D in the DRAM market exceeds the amount of R&D to do an ASIC by a great deal.

So saying ZEC is "ASIC Proof" is a reasonable approximation.

This holds unless a vulnerability is found in Equihash that allows it to be computation bound (Rather than memory bound) in which case a specialized computing unit (which is what bitcoin ASICs are) might have an advantage.

But unless that happens, equishash is designed to be ASIC proof or at least ASIC resistant.


#24

This whole "ASIC proof" has always escaped me. I have been involved with the development of multiple ASIC's. My industry lives and dies by ASIC's. The turn around time on a digital ASIC is short, and with creative collaborative mask development not expensive. For example: we and others would routinely piggy back our ASIC design into the mask (the expensive part) of someone else's mask via the help of the foundry. Then every time the main had a fab run we would get x number of chips / wafer at a set cost from the foundry. We would use these as beta fab runs. Once the design was solid we could spend the money on our own mask or go slow and increase the chip count / wafer on someone else's mask. There is always someone willing to split mask costs with another group and the foundries promote this. Having said this, there is still probe testing, wire bonding, and packaging that can be no small challenge. My point is: if there is money to be made, those with money will put in the work to make it with ASIC's, and leave us in the dust. Today, only Bitcoin make sense to do this, for but in the future, who knows.


Let’s talk about ASIC mining
Let’s talk about ASIC mining
#25

Please explain to me how you can build an ASIC to "leave us in the dust". You say you are in the industry. Do you understand the difference between the SHA256 proof of work and Equihash?

I submit that it is not possible to create an ASIC that would do so, unless someone found a vulnerability in the algorithm.

SHA256 is computation bound, thus an ASIC which doesn't run an ILC could in theory produce a hash every clock cycle. You can't do that with equihash-- unless you can fit 1GB of RAM on that ASIC, in which case you're competing with DRAM vendors for performance.

Or, as I said, the DRAM is already the ASIC for EquiHash. Short of compromising the algorithm you're not going to find a shortcut that lets you magically get significantly more performance, because performance is limited by memory bandwidth to a large amount of DRAM.


Let’s talk about ASIC mining
#26

In addition, Zcash CO has stated that, if ASICs become available for Equihash, they'll change the POW algo to break them. In that light, why would anyone make the investment to develop the ASIC in the first place?


#27

Thank you for pointing that out, I needed a better understanding of the Equihash Algorithm.
I took the time to go through the Equihash paper in detail and I now agree with almost all your points on the difficulty of an ASIC implementation.

However;
According to the paper (VI Parallelism, page 10). The problem for an ASIC implementation is memory bandwidth for parallel sorting algorithms. If you look at section (c) Potential ASIC implementations: They talk about an example ASIC requiring an estimated 1TB/sec memory bandwidth. The last two sentences in paragraph two (and particularly the last sentence) would be my cause for concern. I quote: "the highest reported bandwidth applies to large (a few GB) memory chips, for which it is not so hard to place sufficiently many readout pins to get the bandwidth."; "For smaller chips (700MB) we presume it to be much harder."

So it is my understanding that interconnect technology is the main limiting factor in producing a hybrid ASIC for the Equihash algorithm with extremely high memory bandwidth. Interconnect technology is changing at a very rapid rate driven by multiple industries. There is no rule that the memory must be on the same Die as the ASIC. Multi-layer and direct bond ASIC's are becoming available and even common. Just two years ago bonding a 128 X 128 matrix of 70um pads was nearly impossible. Today we have 256 X 256, 40um pads bonded on a 14mm die (65,536 discreet, high speed, low noise signal connections in 14mm^2). These kinds of connections are not possible with traditional wire bonding technology and multiple completely new techniques had to be developed.

However, yields and fabrication cost are still quite high but falling. So in time, with the right financial incentive, I would expect someone to give it a try. So while I agree its hard, I still do not believe in the term "ASIC Proof".


#28

If you do make an ASIC for ZEC then why would you make it public? Just keep it private and mine and the devs won't have a good reason to change the POW.


#29

Who knows, really. If you made that kind of investment, and needed to keep it secret under threat of it becoming worthless, the size farm you'd build would be hard to keep under the radar, when it came on the network, IMO. ASICs are like having a poster printed: the first quantity 100 are expensive, but it's a lot cheaper per piece for quantity 1000, 5000, .... I don't think the ASIC developer would stop at some modest quantity that no one would notice. But again, who knows.


#30

Hi dlehenky,
I agree. Just keeping the work under the radar with the multiple groups that would have to work on the project would be difficult.......... Unless the fab and packaging house were in on the deal (not likely unless Zcash becomes the second Bitcoin).

However, technology moves rather quickly. If we get into a battle between algorithms and hardware, I choose hardware. How many years has it taken to develop a asymmetrical memory hard algorithm to deal with ASIC's? So what exactly is Zcash going to do to deal with a extremely high memory bandwidth ASIC with scalable memory due to layered direct bonded ASIC technology?

The key is that ASIC technology filters down and is driven by main stream application problems. This will be not different for crypto-currencies. Memory bandwidth is a problem in hundreds of fields (AI and deep learning would be the most glaring example), many are working on the issue right now. Some of those groups have found a solution to this problem, but perhaps not the optimal solution to their specific application (so its not mainstream).

ASIC's are rapidly becoming the norm, generalized multipurpose Silicon is in decline (this is a problem for the fundamental premise of ASIC resistant Algorithms). Technology is driving us to a faster turn time in ASIC's and smaller interconnections are driving complex hybrid packaging for specialized applications. This is only going to accelerate. Software has never kept pace with hardware so why would this suddenly change? How many fortune 500 companies are working on custom ASIC's? Google, IBM, Microsoft this list is long and distinguished.

The whole DRAM R&D argument does not hold water. No one develops an ASIC from scratch unless its highly specialized or you are GE or Google. You use existing designs and or license designs from other parties. Every ASIC design I have been involved with used tech from other groups that we licensed and or paid royalties on. Without this it would have been financially impossible for a small company to develop and ASIC from scratch. ASIC design houses have hundreds of designs to choose bits and pieces from, and lets not forget the national labs that specialize in ASIC's. Commercial groups will also license their ASIC technology as long as it does not compete with their core products (business 101). Many of these groups make their living licensing their core patented ASIC technology.

So if you consider the scope of applications that use ASIC technology (including DRAM), factor in the advanced Hybrid bonding technology being developed in multiple fields, then you can use mostly existing technology for the core of an Equihash ASIC........ That is highly over simplified but within the realm of possibilities.

While ASIC technology is contrary to the core of the crypto-currency fundamental ideology, and would promote centralization. We need to accept that there are groups and organizations that have nearly unlimited resources and will find a way to circumvent decentralized monetary systems if they can. Commercially available off the shelf ASIC's would mitigate this problem. Look at what is happening with Bitcoin. Some say that buying an antminer and joining a pool can make Bitcoin profitable for the average Joe miner again, and could start to decentralize the Bitcoin network again. The problem for us miners is that ASIC's cant mine anything BUT the coin they were designed for, so much higher risk for us. And lets not forget there is at least one lawsuit out there of a Bitcoin ASIC developer accepting purchases from miners and delaying or not delivering the hardware as they wanted to keep it for themselves. Such is what we are up against.

Sorry long post.


#32

I agree with @GPU_Mining. The huge recent increase in difficulty for Bitcoin mining means that unless you have a big mining operation you aren't going to make much money mining on a small scale.

Zcash on the other hand, is in early stages, may never get to Bitcoin levels but as @GPU_Mining said, it looks to be profitable for until at least the first halving of rewards in 4 years.

Zcash miners have also I think reached pretty much close to their efficiency with very little to come in terms of performance gains from Claymore and Optiminer. That in itself does help to make the playing/mining field a bit more consistent.


#33

People don't seem to understand the economics of this situation. You could spend a hundred million dollars building an ASIC for ZEC, but then you would not beat the performance of GPUS. Because people are spending billions on GPU to memory interconnect and standardizing this.

There's a lot more effort behind HBM2 than there will be behind your ASIC. Which is why ASICs are highly unlikely to happen... and thus the GPUs available to gamers will be the best miners for ZEC for the foreseeable future.


#34

Baikal miners can do 5 or 6 different algos and have been incredibly profitable for Dash.