Obelisk Tech Clients Breathe a Sia of Exasperation as Bitmain Releases Antminer A3 Miners

Obelisk Tech Clients Breathe a Sia of Exasperation as Bitmain Releases Antminer A3 Miners

Last night Bitmain announced that a new miner, the Blake2b-mining Antminer A3, is now available for sale. This beats the Obelisk.tech SC1 miner – being designed and manufactured in partnership with the Sia development team – to market by almost four months. While  surprisingly a less efficient miner, Bitmain’s head start will allow them to flood the market with miners which will have a serious, if not catastrophic impact to early Obelisk clients. Linear mining calculators can not be trusted to make accurate profitability projections, and the way in which Sia/Obelisk and their respective communities responds to this competition will set major precedents for the cryptocurrency ecosystem at large.

The New Antminer A3

With little to no advance warning in the run-up to the announcement, Bitmain announced at roughly 12:30am EST on January 17th that a new miner would be released and on sale starting two hours later at 2:30am EST. These brand new miners, the Antminer A3s, would compute the Blake2b algorithm. According to their announcement, the miners will be capable of 815 GH/s at roughly 1275 Watts.

The website currently displays that units were sold out minutes into the sale, though this is a behavior that has been taken in the past when their e-commerce website has experienced technical difficulties, often caused by inundation of demand. Bitmain has since clarified that the “Sold Out” link is merely a deterrent for larger redistributors.

Why this release matters

SiacoinBy some respects, this is a bit surprising move as Blake2b is not a widely leveraged cryptographic function. The largest platform/coin to utilize this algorithm is Sia, a distributed, encrypted file storage protocol. While Sia has seen some recent attention in the markets, Siacoin sits (as of today) as the 24th largest coin by market cap with a global hash rate of only 528 TH/s.  While Sia is very promising technology, it begs the question as to why a multi-billion dollar ASIC manufacturing company would suddenly turn their attention to a coin so far down the proverbial totem pole?

To those paying attention, the answer is quite straightforward. This highlights and reinforces Bitmain’s strategy of anti-competitive practices to control production of all ASIC mining hardware. The Sia development team had partnered with Obelisk.tech to engineer the first ever Siacoin mining ASICs. The way in which these units were pre-sold baked in an exclusivity period in which the early purchasers would have the only ASICs on the market and reap the benefit that the order-of-magitude-more-efficient machines would produce. These hopes were Dashed (pun intended…) with the release of the Antminer A3s, which if they are in fact distributed to purchasers within 10 days, will be out in the wild four whole months before the delivery date of the Obelisk SC1s! Bitmain will establish an early and perhaps irreversible lead in the Blake2b mining ecosystem that could have serious consequences for Obelisk.tech and the Sia community.

How many A3 miners will Bitmain release?

Tl;dr – Profitability should be projected as 1 / total number of Antminer A3s and early adopters should not trust linear projections as they’re using the current, not future, denominator representing global hash rate.

Nobody other than those close enough to Bitmain’s production schedule – which likely includes some individuals at Taiwan Semiconductor Manufacturing and perhaps some of Bitmain’s hedge fund investors – know the true answer to this question. But here’s what we do know:

  • As was discussed last week, banking research reports indicate that Bitmain is planning on producing up to 300,000 units each month over the next few months. It is believed that this quantity encompasses not only Antminer S9s used for Bitcoin mining, but also machines used for mining Litecoin (Antminer L3+s), Dash mining (Antminer D3s), and now Blake2b Siacoin mining. But make no mistake – that number could just be 300,00 S9s with all L3+s, D3s, and A3s being counted separately.
  • Our initial estimates for how many Dash miners would enter the market ended up being overly conservative. While our predictions proved true that 4000 units would be added to the market within two weeks, those are only the units that were turned on in the first two weeks. The hash rate continued to climb until almost 100,000 D3s assuming all Dash miners are either using Antminer D3s or have a contribution that’s negligible. This has produced a lot of consternation in the Dash mining ecosystem as the market would be flooded. Therefore, it is not unrealistic to think that perhaps 100,000 A3s will be turned on in the coming months – whether they hit the market or are hosted privately by “institutional clients or Bitmain themselves.[1]
  • There have been some rumblings that some other ASIC manufacturer have already turned online ASICs given a recent spike in hash rate. With Obelisks not yet in the testing phase, this implies that it must be a different ASIC manufacturer, and there aren’t many that have the ability to spin up ASICs without any marketing around it. With Bitmain shipping the A3s almost immediately (10 days from purchase which means units could be in clients hands by early February), this confirms a few things. For one, this further confirms Bitmain “QA tests” all of their units. The shiny silver sticker on the packaging indicates that your miner is actually at least partially used. You can decide whether you’d prefer that they run your miner before shipping it to you or ship it out untested. You can also make your own judgement for what constitutes an acceptable “QA run” of mining with your rig before you can. This also indicates that they are confident enough in their equipment that their 10 day delivery window is legitimate.

To provide a projection as to how this might unfold, we can simply take a look at the last time this happened: the release of the Antminer D3.

History Repeating Itself

source: https://bitinfocharts.com/comparison/dash-hashrate.html

We issued a warning before the releases of the Antminer D3 as to impact the introduction of the new generation of Bitmain units would have on a new mining algorithm’s ecosystem. Due to the linear nature of most mining calculators, projections show that an Antminer A3 is slated to make upwards of $500 in Siacoin per day. The same thing occurred when the Antminer D3s were released; linear calculators showed profits upwards of $80k per year per unit, which if you’re a current owner of a D3, you might recognize as wishful thinking.

In reality, the hash rate will increase dramatically with the introduction of the new technology, immediately diluting the effect of any miners without an A3. Mining reward will then resemble all other currencies for which Bitmain provides mining hardware, where unit profitability can be calculated simply as 1 / total Bitmain ASICs produced. The current global hash rate for Sia is just under 500 TH/s. This means if Bitmain put out only ~600 Antminer A3s, the Sia hash rate will double, causing the linear calculator to project profitability of half that of what it shows prior to the miners’ introduction.

What will the Antminer A3 really produce?

In short, if Bitmain does produce 100,000 Antminer A3s, each Antminer A3 would mine 1 / 100,000 of the daily block reward. As 100,000 Antminer A3s would yield a new global Siacoin hash rate of 815 PH/s, or 815,000 TH/s – about 1600x the current hash rate – it makes the entire current hash rate effectively negligible. Therefore, using some quick, back-of-the-envelope math:

The current daily block reward is ~$159,000 worth of Sia coin (just set your hashrate = global hashrate on any linear calculator of your choice). Given the above, this implies that if Bitmain sells 100,000 Antminer A3s, each one will net $1.59 per day. 50,000 units sold would yield $3.18 per day. 10,000 units would yield $15.90 per day, which puts it in line with Bitcoin and Litecoin’s recent daily profitability.

Unless of course, the price of Siacoin dramatically appreciates…

Why did Bitmain make the Antminer A3?

One of the most interesting parts of this story are the specs of the miner. With the Obelisk team recently announcing a confirmed hash rate of an SC1 miner of 800 GH/s, this means that it puts it on par with the Antminer A3 in terms of hash rate, yet Obelisk claims that the hash rate can be achieved with less than 500W, whereas the Antminer A3 requires 1.2 kW, indicating it’s less than half as efficient. This point may be moot given that the Antminer A3 is first to market, but it also begs the question as to why the unit would be less efficient given Bitmain’s experience in ASIC manufacturing.

Bitmain uses TSCM which has a 16nm node; all of Bitmain’s other current offerings contain 16nm chips. Obelisks are slated to be 28nm units. Either Bitmain is using a lesser node, or achieved wildly inefficient results (assuming Obelisk can honor its claims of requiring <500W). It would not be surprising that a company took shortcuts to get a unit to market where timing is everything. Is this lazy chipmaking?

Perhaps the 815 GH/s is only a subset of the total mining power of the device, while the rest is utilized by the manufacturer (you’d be easily able to prove/disprove this through viewing the internet traffic if using a pool other than Antpool).

Or perhaps they ran into the wall that is the law of Dennardian scaling: shrinking transistor size in a device where all transistors are to be on all the time (like mining) leads to a proportional increase in un-utilized “dark” silicon. This results in situations where larger transistors may be more efficient to manufacture/use when taking Non-Recurring Engineering (NRE) costs into account.[2]

Given the secrecy and rush to market of this new technology, and the (apparently) sub-optimal specifications of the unit, it leads one to believe that the primary intention of the device is to flood the market with units to drown out Obelisk Tech before they can position themselves as competition. It is a bit curious that there is no mention of “Siacoin” anywhere on Bitmain’s website; contrast this with their slogan of “Antminer L3+ – [the] world’s most powerful and efficient Litecoin miner”. Curious indeed…

Bitmain made no secret of what the Antminer L3+ was meant for. So why is the word "Sia" nowhere to be found with respect to the Antminer A3?
Bitmain made no secret of what the Antminer L3+ was meant for. So why is the word “Sia” nowhere to be found with respect to the Antminer A3?

Sia/Obelisk Recourse – The “Nuclear Option”

Here’s where things get a bit scary. The Sia development team knew that this very development was a distinct possibility and said as much in a press release. When justifying the decision to not take preorders from “large buyers”, they wrote:

“If we close ourselves off to large buyers, we may not be able to produce 16nm chips and a competitor may be able to. That means that the competitor will have a monopoly over Sia hashrate the way Bitmain effectively lords over the Bitcoin hashrate.” … “But as soon as one party is able to overcome that [chip design] barrier, it’s pretty much game over for everyone else.”

They go on to detail action that they could take if they found a competitor like Bitcoin in control of the hash rate:

“The advantage that ASICs have over GPUs is enormous, and if the first ASIC chips were made by someone who intended to hold a mining monopoly, likely the only recourse would be a PoW hardfork. There’s a large first mover advantage, because if you are the first ASIC you get to collect 100% of the block reward, and anybody who tries to compete with you will only be able to collect 50% (assuming they even have comparable hardware).”

While we applaud Sia and Obelisk for their decentralization attempts, forcing a fork to disenfranchise mining equipment producers would set a terrible precedent which would have ripple effects (pun not intended) on the cryptocurrency ecosystem. While we just discussed how Bitmain mines with their own equipment, a large number of participants are average people who purchased equipment in good faith who would find themselves mining an algorithm without a coin. Such a kill-switch[3], or providing the power for the users/developers to invalidate the miners, is a very dangerous path for those putting their faith in the immutability of code.

The likely winners here as these two companies battle over the Blake2b hash rate, are the holders of Siacoin. Keep an eye on how Sia’s price with respect to fiat fares as the mining ecosystem has money poured and attention poured into it. It seemed to bode well for Litecoin and Dash.

Update: As this blog was being published, the Sia team put out an official response to the news. They confirm that the Bitmain unit is less efficient, acknowledge the “kill-switch” built into the SC1, and concede that the impact of such an action would have a greater impact on Bitmain’s customers than Bitmain themselves.

[1] The bottleneck in most mining ecosystems doesn’t appear to be the inability to procure mining rigs, but instead finding places to turn them on. With China pushing to reduce the strain on their power infrastructure by having miners relocate overseas, there appears to be a run on colocation space open to the little guy. We know for a fact that many people are sitting on miners without a home for them and therefore current hash rates may underestimate the total possible hash power. The Miners’ Union is working hard to provide a solution to this…

[2] This could easily be the topic of a future blog or even a series…

[3] Such a “kill-switch” would have to be a developer or user-activated soft fork, as what miner is going to vote altruistically to invalidate their own expensive mining equipment?

  1. Won’t this destroy the price of SIA?
    All the new A3 miners, are going to mine SIA, and flood the market selling them to buy bitcoin.

Leave a reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.