Grin Improvement Proposal 1: put later phase outs on hold and rephrase primary PoW commitment

Before further discussion, I want to emphasize again that for official information please refer to our website:
Now back to the topic around GPS & memory architecture. First of all, even though DRAM needs to be refreshed every few milliseconds, when using at full speed it is typically less than 10% of the total power consumption. Most of the energy is spent moving data from point A to point B. Let’s examine access energy of two popular high speed DRAMs: GDDR and HBM.


HBM is a few times more efficient than GDDR because it is connected to the Core ASIC through an interposer. The energy to access a bit in HBM2 is about 4pJ/bit, most of that is data movement both on the DRAM die and from the base layer die to reach the I/O pins. To visualize this let’s look at a picture showing how data travels:

Currently HBM is still too expensive for mining ASICs. On the other hand, TSMC has announced the availability of Wafer-on-Wafer (WOW) technology which allows you to stack two logic dies and connect them through TSV. This is a cheaper alternative than using HBM. Also because there is no interposer it allows higher bandwidth. This opens the possibility of building an SRAM chip and stack it with your ASIC core.

In order to design an energy efficient ASIC for GRIN, one has the task of minimizing data movement. I will make a few observations here. It is not meant to disclose our particular method but will stimulate thinking. For example, if you goes with single chip ASIC, you can’t treat 512MiB of SRAM like one big memory chunk. The SRAM size on 16nm will be over 22mm on each side. A completely random access of this SRAM block is not efficient. This is an area where AI technology overlaps with GRIN Mining. Ideas like processing in memory (PIM) has to be considered where logic is mixed with SRAM to minimize power consumption and increase speed. Again, instead of focusing on building high bandwidth, you should at the same time think about minimizing movement. This is true both for both single and multiple chip design.

thank-you for the contribution. As an FYI, here is the publication where the diagrams would have originated. Sharing so we know this is public information and nothing confidential is being posted again.

Glad to see after 3-4 days, the thread is not being mis-directed.

At this point, there may be 3 ASICs that have invested resources into C31 or C32 designs. If anyone knows of others, let us know.

Incorrect, Linizhi Corp will be releasing their Lavasnow Ethereum miner using HBM in June. I should note that Linzhi’s performance is 3x the performance of Inno’s A10 miner.

@asic_king, thank you for the memory lesson. I noticed you did not give Obelisk the benefit of the doubt in the previous threads when it comes to memory segmentation and optimization. I guess those dumbasses at epic who built 10+ generations of GPU don’t know how to build memory controllers and high performance, low power ASICs.

Btw, you still didn’t answer how many ASIC’s are in your rig. Why is there any reluctance when you have released your CC31 rig performance but have only provided the CC32 at 2 gps for a single ASIC?

A few things that have been on my mind:

Innosilicon as best we know never shipped an Ethereum miner. Though I do believe that its claims and performance did damage Bitmain’s sales. I know people who ordered them and never received them. I’ve heard other rumors saying that the Ethereum miner from Inno does exist, but Inno only mined it internally.

During the sale of Obelisk’s Decred miner, Obelisk had the most energy efficient miner on the market, even ahead of the Innosilicon Decred miner. While the Obelisk sale was still going, engineers claiming to be working on the firmware for Inno’s Decred machine said that they had changed the firmware to boost the hashrate to 4.4 TH/s on only 1000 watts. This put the Inno machine very firmly ahead of the Obelisk machine in terms of cost and efficiency. Note: there were no official statements by Innosilicon, and also the connection between Inno and the engineers making the claims was never officially confirmed. Similar to this conversation here actually.

This claim transparently hurt Obelisk sales. Though there were no official numbers saying the Inno machine was better, the rumors were clearly impacting our sales. When the Inno Decred machine shipped, it was only doing 2.4 TH/s at 1000w. The 4.4 TH/s claims were a paper tiger, and not even from official sources, but the damage was done.

As it happened, Whatsminer had been making a 40 TH/s miner anyway. Neither the Obelisk nor the Inno machines were viable, so it ended up not mattering so much anyway. But this was not known at the time.

There’s another super important variable that isn’t being discussed by Innosilicon, and has never been discussed by Innosilicon, and also has been used dramatically to their advantage and their customers disadvantage at multiple points in the past: total production quantity.

Mining is a zero-sum game. The amount of reward that exists get split between the miners evenly, and so if more miners exist, the reward for each miner is smaller. Innosilicon has sold miners in the past at extremely high prices, even using phrases such as “you can’t put a discount on dollars”, not disclosing production volumes, not disclosing the fact that they would be mining themselves at-cost, and ultimately resulting in their customers losing substantial sums of money even though Innosilicon had the best miner.

As far as GIP1 goes though, it seems like everyone is on board with delaying the phase-out of CC32 indefinitely, and stopping at CC32. And it seems like delaying the phase-out of CC31 is out of the question, and that Grin is committed to moving to a 7nm-required algorithm for their cryptocurrency.

Yes, it’s possible that competitor statements about their tech had an impact on your presales.

However, since your presale specifications changed multiple times, including the delivery dates, isn’t it just as possible (if not more likely) that the way you executed the presale program could, itself, be the main reason for whatever impact on presales Obelisk experienced?

1 Like

At the time, Obelisk had the most efficient machine on the market (notwithstanding the whatsminer machine, which hadn’t yet been announced), even despite missing spec. Sales dropped off tightly corresponding with the rumor that the Innosilicon machine was actually 4.4 TH/s.

No we never shipped any Ethereum miner. The first chip came back functional but there is some performance/stability issue that required a fix to address. We now have the fixed chip back and it is looking promising.

What complicates the situation is that there are multiple players each producing miners. So it is very difficult to model the right amount of miners that will maximize profit. What often happens is that you will be forced to drop prices and sell your stock before the competitions do. It is often very dynamic and chaotic.

Miner companies like Innosilicon typically don’t want to do self-mining because it stretches out the capital cycle and put stresses on cash flow. It is usually forced into self-mining if the market price for miner is depressed thus self-mining might be more desirable. Another reason is that there will be sub par machines that is not good for sale but can be used on a self-mining basis to recover a portion of the investment.

It’s good to do a post mortem on this, and I’ll help provide a perspective on how this works as a consumer but also behavioral economics.

Let’s concede that you had the best performing unit on the market, but even if they the units were technically the best performing they did not meet the performance expectations that you set with consumers. And yet, your customer approach has still been “it’s the best soup on the street and it’s late you’re going to like it.” You used a handcuff approach to customers and customers do not like that – or their shacklers.

So yes, in that context, a competitor comes into the market and says “we have another product and it’s going to do X or Y,” then consumers make choices based on credibility. On that front there are no saints in this business. But customers have real data at that point about how you handle them as customers and they’re reconciling your prior promises with your treatment then.

For example, you’re still sending Batch 5 orders of DCR-1s to customers today, even with dozens and dozens of people on your Discord posting that these devices are losing money and aren’t being used. (The environmental damage of this alone will be a topic I’ll turn to next: you come from a generation that should be protecting our environment and producing thousands of computers that go immediately unto the trash is a tragedy).

This is how consumers behave: you set the expectations with customers on what you’re selling and how you’ll do business. You are selling your credibility. You’re losing that credibility quickly by your practice, and even if competitors’ practices aren’t ideal, consumers are making a comparative choice. Here are your stats:

  • SC-1: Only worked b/c Obelisk monopolized the platform and bricked competitors. Made purchasers of Obslisk happy at extreme expense of any other purchaser (also creating significant environmental damage by bricking the other devices prematurely);
  • DCR-1: Didn’t meet any technical or delivery metrics. Still shipping units now even though useless 8 months ago.
  • GRN-1: Another presale. Obelisk is having difficulty making the design that was pre-sold work.

You see, you’re looking for externalities but you did not do what you said you would do on any of these things and you’ve only delivered an effective product (the SC-1) in a context where you cut out the competition from the field of play.

With the GRN-1, you still have an opportunity to do the right thing now. Don’t make the mistake that you made before by digging in and producing rigs late that customers don’t want because they’re not what you promised and then moving on. You won’t make your delivery times, and you should start now to offer GRN-1 refunds.

It’s never too late to do the right thing.

Thank you for your honesty. That finally solves the question about the availability of the mysterious A10.

I’m very surprised at how long it took Innosilicon to fix the problems with the A10. You are delayed by at least 7 months as initial shipments were scheduled for September 2018. Why the long delay? How do you guarantee people who pre-order your Grin rigs that Innosilicon won’t be have the same delays? Otherwise, they are better off buying both the Obelisk CC31 and CC32 miners.

On a different note, how will A10 be competitive with Linzhi’s upcoming ETH miner? It’s 3x your performance and uses HBM. Chen Min is shipping in June.

Interesting dilemma! How does your CEO, Gordon Ao, delineate the difference between mining with defective mining rigs or commercially viable rigs? I mean if you already have the infrastructure for mining, how does Gordon resist the urge to make more profits by secret mining.

You raise some good points here (I’ve been picking on Obelisk but you are right to point out the issues here with Innosilicon).

One suggestion that a community like Grin might consider is to have a Code of Conduct that vendors must sign up to in order to have tech-coordination discussions with community founders. Vendors gain entry to privileged discussion with coders in exchange for agreeing to abide by certain expectations.

The Code of Conduct could be as simple as this.

  1. Vendor agrees to a refund policy that allows refunds if vendor misses a technical specification or an expected shipping date by more than 30 days.
  2. If conducting a presale, vendor agrees to maintain sufficient funds on hand to handle refunds for consumers as required.
  3. Vendor agrees not to use pre-announcement tactics for new products unless vendor pre-announces with a shipping date (i.e., to discourage misleading market, any announcement requires a shipping commitment tied to willingness to refund).

These would be binding on any vendor that chooses to collaborate in the same way that NDAs and other instruments are. It would hold the vendors directly accountable to the community by virtue of some simple rules. Vendors could still develop products independently without signing on but the benefit of agreeing to the Code of Conduct should be apparent for access.

Obelisk is not having difficulty making the Cuckatoo31 design work. Everything is on-schedule, and on-target for performance metrics.

Your open letter included a demand for a full refund. The terms of service that you agreed to upon buying the miner indicated that refunds would not be possible. We offered a buyback program whereby you could receive partial compensation in return for releasing us from the obligation to finish manufacturing and shipping your units (several hundred dollars per machine). As you skipped over this buyback option, Obelisk still had a legal obligation to deliver the units that you ordered.

We have not delivered hazardous waste, we have delivered fully functional DCR1 mining machines, per your order.

We have substantially upgraded and refined our processes at Obelisk to ensure that we are capable of meeting delivery targets. We’ve posted about this at length already, but we’ve made changes all the way from our firmware strategy to our chip specification simulation to our manufacturing pipeline.

This is a non-answer. Innosilicon at some point decides how many wafers to order. That number should be disclosed to customers, as that number directly impacts the returns that the customers will see. Especially if machine margins are not known, and the manufacturer both has the capability to self-mine and a history of self-mining, there is a big risk to the customer that the manufacturer will produce more machines than make sense at the customer price.

If Innosilicon refuses to disclose total production volumes, its customers are fully trusting that a reasonable and profitable number of machines have been made. Every machine that gets made reduces the profitability of all of the other machines. Without the critical variable for total machines produced, customers cannot make accurate models to determine whether the purchase makes sense for them.

The graph size upgrades were introduced … as a means to resist single chip ASICs, which were deemed “not meaningfully different from pure computational ones like Bitcoin’s sha256".

All this was heavily put into doubt when Obelisk posted details of their GRN1 miner in … revealing simulation data of a single chip C31 ASIC as well as planned single chip ASICs for C32, C33, and C34.

Based upon the long discussion here, I didn’t see the obvious confirmation/s from other ASIC firms. So may I ask explicitly?

  1. Does everybody convince single chip ASIC for C32,C33,C34 are indeed feasible?
  2. The original graph size upgrading plan doesn’t help to avoid (or make it hard) to make single chip Grin ASIC?

If we maintain our commitment to the C32 phaseout, and have ASIC manufacturers make significant investments into C33 miners, then we lose the option of freezing the primary PoW at 32 edge bits. Which is a very natural size that’s easier to design circuits for. Also, there’s something nice about the memory requirement being a round 1GB.

If I understand well, we’re going to reverse direction: to make it easier for ASIC miner. Then, my question is why we don’t directly change the C31 phaseout plan? since the C31 make the ASIC even more easier.

And to protect the existing investment on C32 ASIC, we can keep both C31 and C32 but change the C31 phaseout plan to make it more fair between C31 and C32 ASIC miners.

According to current phaseout plan, the C31 miner has obviously shorter life time than C32 and it’s not fair for C31 miners. (mainly because the C31 miner come out a little bit late than expected!)

The Grin developers currently have no intention of changing these rules in the foreseeable future, unless Cuckatoo is shown to be broken (e.g. by exhibiting a sublinear time-memory tradeoff), or unless the phasing out of smaller sizes should be delayed in order to ensure public availability of >= 1 GPS miners.

^ According to our commitment on Cuckatoo31+ (im)mutability, even the Scheduled PoW upgrades proposal was added later than that commitment, I’m afraid we’re opening a back door to start breaking our commitments. I’m not against the necessary change/improvement, but just want to emphasize that we MUST be very careful and make the community well informed about this discussion, and please DON’T hurry to make decision unless we’re 100% sure.

No, not everybody is convinced that C32 chips have sufficient yield to be economically feasible. But we should find out within a year, as Obelisk is planning to put them into production, following their C31 chips.
The original phaseouts do make single chip ASICs harder. But as Moore’s law marches on, with 5nm and 3nm processes in the pipeline,
it’s not clear if it will ever be too hard.

Keeping our commitments is precisely the reason we shouldn’t change the C31 phaseout plan, which is less than a year away, and upon which current investments have been made.

That’s why my proposal only affects C32 and later phaseouts, which are much further into the future, beyond the 18 month deadline that I’m proposing the term “foreseeable future” should refer to.

@asic_king Update for the prospective buyer? Is 100GPS @ 500W achievable as the final spec?

Time sure flies. The other reference point that is confusing for many of us is the announcement back in January. The expectation was that final specs would be available BEFORE the pre-orders start. That factoid was from the company themselves. But no specs that can be validated so far??? And they are also onto batch2 of the pre-orders. This does not pass the red-face test.


No new details on C31 specs, but this tweet from a few hours ago is suggesting they are pointing everyone to the C32 version and telling everyone to wait one more day.


We have released the spec. Please refer to our website for official announcement.

Some observations:

  • fewer graphs per second than Obelisk machines

  • worse energy efficiency than Obelisk machines

  • higher cost per computation than Obelisk machines (even with the presale discount)

  • 2.5x TMTO to hit CC32. Slean mining maybe? Certainly indicates that CC32 is a second-class algorithm on this machine, if it was first class the TMTO would be 2x at worst. That’s also going to impact it’s competitiveness in May, when there’s been plenty of time for manufacturers to come forward with a new CC32 miner.

  • No disclosure around production volumes. Which in my opinion matters more than any of the other points. Innosilicon has had multiple points in history where their miner was absolutely dominant, many times better than the next best competitor, and yet because of overproduction everyone who purchased from them still lost money.

  • No mention of what type of mining it does. If it does mean mining or slean mining, it should be capable of targeting Cucakroo29, which means the planned hardforks will be important.

Awesome. We will be sharing our observations with the customers that have hired us for a technical assessment. Hopefully, with their permission, some of that insight can be released to this community, so informed decisions can be made. And thank-you for taking down the first copy of the Grin PCIe card from your Site. It was obvious that it was a Nvidia Tesla card. That could have mis-led the community that Inno was now re-branding a GPU card from them. The 2nd version may be confusing too with the graphics DVI-like connector on the back-side. But this time there is a disclaimer that it is just for illustration purposes. So we can accept that it is a marketing mock-up.

As a follow up from Tuesday’s governance meeting, it’s been proposed that we continue the conversation regarding @tromp’s proposal to put phaseouts on hold:

tomorrow Saturday @ 15:00 UTC

The aim is to reach a decision regarding the proposal and stakeholders and interested participants in the community are most welcomed to participate.

@tromp has confirmed he’s available.

It’s unfortunately quite short notice. If there’s reason for why tomorrow is not a good time to do this, please advise.

1 Like