Innosilicon's Grin ASIC Miner or Paper Tiger?

Innosilicon “announced” their intent to make an ASIC miner for Grin. Very strange thing to do as there were no specs disclosed. Is this a paper tiger intended to stall Obelisk’s Grin1 miner?

I’d be skeptic of Innosilicon based on their history of secret mining and lacklustre performance on memory hard architectures like ethereum. Their A10 miner wasn’t much better than several GPU’s; it seemed to be an ASIC DDR controller (which they are well known for) with a bunch of cheap external memory. Which means Innosilicon will be severely memory bound or have massive boards very wide memory buses.

However, it certainly seems that Grin has the attention of the hardware community with Sapphire 16GB Grin GPU, Obelisk Grin1 miner and now Innosilicon.

@timolson, what do you think of innosilicon’s announcement?


Thanks for posting on this. I have always been a fan and client of INNO. I am almost sure whatever they engineer will certainly have high performance yields, they certainly have a solid track record of that. As a miner and customer we all look forward to asic’s that enable return on investment with calculated risks. This being said this is a very serious topic for the GRIN dev team to follow carefully as well as the GRIN community with emphasis to set specific guidelines on, “Transparency”. Here is where we are with transparency and ASIC manufactures thus far.

Manufactures have long been known to operate industrial scale mining farms while self mining the same ASIC’s being sold to customers (miners). Most all manufacturers thus far have done little to nothing in disclosing audit-able asic production runs on batches sold and unsold. Additionally absent of self mining disclosures related to said production run quantities that would account for otherwise direct self mining in secret, competing with its customers and reducing any chance of roi or ability to quantify the real risks associated with buying into a new asic as such. As a large scale manufacturer there are unfair advantages that can harm and have harmed coins in the past as well as customers (miners) and trust across the miner communities around the world.

A good example of lack of “transparency” has resulted in DCR, BYTOM and others to basically lose the majority of its’ miner community over the asic manufacturers that opted to self mine in secret against its own customers inability to mine and reach roi. Worst part would have to be the practice used to cripple other manufacturers whom are also producing new ASICS for the same coin and rendering their sales efforts to fail as the network hash of said coin raises by a factor of 2x in a very short time thus causing the value of newly released ASICs to be worthless along the way directly competing with their customers (miners). The advantage manufacturers have over customers, coin network, and mining overall comes with great responsibility and trust.

Here is my suggestion and I hope others agree particularly all those whom can relate to the above as miners:

• An ethical policy adoption, mandated by the Coin devs and community towards setting the right rules for manufacturers early in to ensure that the coins’ network is not rendered un-mineable by its community of miners and supporters. The disclosure of actual and factual quantities of ASICS being produced for every batch, quantities shipped as well as a full transparent disclosure of intent to self mine and what percentage or number of ASICS allocated to said practice. This helps miners buying into new ASICS, as well as the protection through policy for asic makers to ensure protection to the coin network with factual transparency enforcement.

• Furthermore setting timeline based caps on quantities to be produced by individual manufacturers to allow for fair balance across manufacturers and sustain a real audit-able scope for the Cuckatoo31 Protocol and GRIN coin network. X amount quota per 12 months per manufacturer to be produced…

• Penalize bad actors and unethical practice via forking those ASIC’s and future ASICS from said manufacturer off the network in the event of discovery of unethical and advantageous practice with intent to mislead and disclose misleading quantities and false information.

One good examples of “Transparency” being put to to use by Obelisk. They disclose on their website actual batch quantities sold and shipped. This has proven to further protect Sia coin and its valued customers and mining community in that they are able to factor in risks based on quantifiable quantities of ASICS on the coin network and not get blind sighted by manufacturers ill intent to compete with its own customers.

In closing my thoughts are to up vote a need for clear, strict policy adoption to protect the longterm mining roadmap the mining community needs in a way that protects the coins’ network from abuse as well as its means to balance the playing field for its community of miners. What we do not want is a repeat of the mentioned coins above.

Make sure to checkout my work on my non-profit ASIC realtime profitability framework which gives a complete breakdown on profits per day, cost to operate and much much more. Help by sharing with other fellow miners to quantify risk by monitoring in real time!

Lastly I manage a large all miners discussion group on telegram called “Blockchain Association of Miners”. Please come join our public group:



Could you give some examples on how the should be done in practice? Preferably in a way that is not prone to cheating?

What impacts do you see this having on the centralization of the project?

Honestly, I don’t think ASIC manufacturers owe anything to their customers except to deliver a product that hashes the algorithm at the declared specification.

It’s totally up to the customer to decide if they want to buy and it’s up to the buyer to calculate the risk themselves.

Calculating the risk is not the job of the manufacturer. Why would a manufacturer sell miners in the first place? Why don’t they just build them for themselves only? Why does the public think they deserve to buy something that has a predictable ROI? Isn’t it the mining communities own fault for having expectations that are too high?

Mining farm operators are trying to make a living off of mining a coin, but is that why the coin exists in the first place? Does it exist for the purpose of making its community a lot of money? Or does it exist for the purpose of enabling a blockchain product to flourish?

@spartadata, I understand where you are coming from with your comment on that forum, but at the same time, there are other considerations at play.

If you understand where I’m coming from then you would have touched on the topic of “self mining” And it’s ovious negative implications as I had referenced. The risks are not ethical or fair when the objective is to secret mine the coin network dry - it’s a thing. Keep them honest or fork’em.

It’s good for Grin to have such competition, and at least InnoSilicon promises to have a working prototype with known hashrate before asking for your money. I’m skeptical that it will be profitable for most customers, though. ASIC manufacturers get their profit up-front and leave the end users carrying the risk of making a return on investment over a longer mining period. Historically, early ASIC’s are far over-produced / oversold and they end up losing money on a per-unit basis.

1 Like

This idea is prone to corruption and fraud. To me, it reeks of authoritarianism.

@spartadata, thanks for your reply, certainly you present some interesting perspectives from a miners" POV. There’s a fine line between centralization and decentralization or between dictatorship and democracy.

The Grin community has done a good job in terms of trying to temper ASICs from overwhelming the network while encouraging the hobbyist miner. My fear is the Grin community has set the bar too high for early entry by ASIC providers.

Let’s look at the scenario from several hardware alternatives.

  1. FPGA mining is mostly a non-starter for Grin mining. FPGA memory interfaces (Altera DDR3 and Xilinx DDR4) are too slow.

  2. GPU have limited returns on Cuckaroo29 as competition is high and returns tail off over time. For Cuckatoo31, supply of 2080ti will be limited due to high price and demand from the AI market. They don’t compete well against ASIC miners like Obelisk which are 80x faster than the 2080ti adjusted for fidelity rating. Low cost GPU like Sapphire may be the key here for volume.

  3. I’m surprised at how aggressively Obelisk is pursuing the Grin market given the 2020 phase-in of Cuckatoo32 and the immaturity of the Grin market. The investment in building an ASIC miner is relatively high and risky as the market is small and the Grin price/demand has yet to stabilize. This staring contest between Inno and Obelisk is bad if it scares off the other party or reception to pre-sales are weak and manufacturers are unable to recover NRE cost. It leaves the Grin community in a vicarious position.

@timolson, I’m actually skeptical whether it is profitable for manufacturer to use ASICs with DDR memory. The problem is the price/performance of these rigs if we use ethereum ASIC miners as a case study.

Innosilicon’s A10 and Bitmain’s E3 were duds IHMO. They were basically hash cores with wide memory controllers using cheap external DDR memory. They weren’t much faster, cheaper or significantly more power efficient than GPU’s.

Lingzhi’s ethereum HBM miner looks good on paper but it’s too late and will be limited to a fragment of the Ethereum chain after the ProgPOW fork.

A significant portion of the bill of materials for building these rigs go to memory content rather than to the ASIC where the silicon area will make a significant performance or power improvements.

Equally prone to corruption and fraud refusing to disclose batch quantities transparently as a manufacturer. :mag_right:

Innosilicon has a long history of keeping their products secret either up until they are ready to start ordering wafers, or until one of their competitors announces a product. This is advantageous to Innosilicon because the uncertainty creates risk and FUD for any potential competition, and increases the barrier to entry for the competition.

When they do announce, they like to show that they are as progressed as possible, because the closer that they are to completion, the more it hurts the revenue of their competitors. If Innosilicon had already taped-out, I’m sure they would have said as much, because it would likely significantly harm Obelisk’s chances of bringing a superior product to market. Given that Innosilicon used the terms “currently designing” and “launch… early summer”, and given that Innosilicon “launched” their T3 miner in November even though high volume units do not begin shipping until March, I think it reasonable to assume that Innosilicon is behind Obelisk for their grin miner. If they were not, they would be saying as much, because that would hurt Obelisk much more. They are being vague here because they know the imaginations of potential Obelisk customers will paint a bleaker picture than the actual facts.

A coin is in much worse shape if the developers and leadership are corrupted than it is if the ASIC manufacturers are corrupted. The ASIC strategy has strong incentives protecting the underyling cryptocurrency even when there is a single dominate manufacturer that only self-mines and keeps 100% of the hardware to themselves. But if the developers are corrupt, there are a lot of problems.

Any policy around manufacturers that the developers take should be carefully considered and should be designed to be as hands-off and as objective as possible.

Obelisk has a long term strategy to be the dominant manufacturer for Grin. We intend to demonstrate with the GRN1 that we’ve cleaned up substantially, that we are capable of meeting shipping deadlines, and that we are capable of creating highly competitive ASICs for the Cuckatoo algorithms. We’ve been working tirelessly since the snafu last June to fix our process and ensure that what happened the first time around won’t happen again. And the proof will manifest when we ship our Cuckatoo31 machines.

This is why manufacturing volume disclosures are extremely important. We’ve seen time and time again that when manufacturers aren’t telling you how many machines they are producing, it’s because they are producing so many that buying their machine at the listed price is a bad deal. In the case of innosilicon, they use their markup to build machines for themselves to mine, which essentially guarantees that their customers will not be able to ROI.

I’m happy to talk about this at length if people do not understand why, but it is a really bad idea to buy mining machines from a manufacturer who is not disclosing total production volumes. When manufacturers do not disclose production volumes, they are the only ones who know what the fair value for their miner is. You do not know, and therefore you should not participate. You are fundamentally disadvantaged, and the manufacturer is fully exploiting that fact.

Mining is different from other manufacturing industries because mining is a zero-sum game. The sum of all miners have to share the same block reward. If there are more miners, there is less block reward that can be given to each miner. Comparing best practices with Apple or Intel is a bad idea, because Apple and Intel don’t operate in a zero-sum world. It doesn’t hurt Dell if Intel uses massive amounts of their own processors for machine learning and Intel didn’t disclose that to Dell. But it does hurt a mining farm if Innosilicon uses massive amounts of their own ASICs to mine and does’t tell the mining farm about it.

And I can’t stress that enough. If you do not want to lose money buying mining machines, do not buy mining machines from manufacturers that don’t disclose total production volumes.

Hi Taek, IMHO this is not the complete picture.

When there is a bad actor producing ASICs for a given alogorithm, buying an ASIC for that algorithm from any manufacturer is a bad idea. Beacuse you don’t know how many machines the bad actor is producing, you have no way knowing the right value of other manufacturers’ machines, given the zero sum nature of mining.

So unless, other manufacturers are significantly discounting, read giving away almost free, their machines value would also be indeterminate. Hence equally bad.

So a single bad actor will destroy the whole pool for that algorithm.

1 Like

@Taek, I think it’s commendable that Obelisk stood behind it’s products, offered compensation and fixed the problems. That speaks highly of your character and your company.

Good luck with your Grin rigs. The mining community needs a high quality supplier that provides great product, early access to product and a supplier that doesn’t compete with customers.

I tried to bite my tongue, but…

This is exactly what Obelisk did to us at altASIC when we announced a Decred miner. You were building yours in secret until we made an announcement, then you advertised completely fabricated delivery dates, hashrates, and power consumption numbers. Decred users funded your offering, thinking they would get a 1.5 TH miner using 500w delivered by possibly May but definitely by June. Thats’s what you said. Your customers actually got a 25% slower miner using 10% more power, delivered in October.

But hey, that’s ok when you make false announcements in order to crush competition, right?

At least InnoSilicon isn’t making up numbers then asking for money up-front.

It remains to be seen how they approach a Grin ASIC. I think mean miners are probably the way to go, in which case it will still be DRAM bandwidth limited. The power advantage will be more significant since Cuckoo uses a lot more hash computation compared to Ethash’s mixing function.

At least in the case of Sia and Decred, the problem actually wasn’t that Innosilicon overproduced. The machines that they were making were profitable to manufacture and mine if you were paying a price that was close to the manufacturing cost. The problem is that most of their customers were paying prices that were 5x to 10x the manufacturing cost, and they overproduced for that price point. The well wasn’t poisoned for a machine that had competitive hashrates.

That said, there are also cases like Dash where the manufacturers do overproduce even if you can get the machines at-cost, in which case the well is poisoned for everyone unless someone comes out with a substantially better machine. If customers buy from manufacturers who overproduce, then the customers lose money and the manufacturers do not. But if the customers do not buy from manufacturers who overproduce, then the manufacturers lose money and they will be less likely to do it again.

We were fully up front with all of the information that we had to the best of our knowledge. We were not making numbers up, we were using results that we had from simulations that were run on the chip design. We based our shipping dates on the tape-out timelines that were being presented by our engineers, and basing shipping dates on the amount of time our PCB teams were representing would be required to turn prototype chips into a functional unit. And we added significant buffers to both timeline and simulation results to account for misses that we knew had been experienced by other mining rig manufacturers.

We ended up missing both spec and shipping date for a number of reasons that have all been thoroughly addressed for the GRN1. We provided our customers with full compensation for the missed shipping date, and even though we missed spec we came out well ahead of the altASIC estimated specs. altASIC was advertising miners performing at 700j / TH for $2500 / TH. Obelisk’s final units came in at 420j / TH for $2100 / TH. Especially taking into account the compensation for late shipping, Obelisk was clearly the better purchase of the two (though the real winners ended up being the 14nm Innosilicon machine and the 16nm Whatsminer machine).

1 Like