EIP: Modify block mining to be ASIC resistant.

[pipermerriam]

Disclaimer: My area of expertise does not lend well to me suggesting how to make things ASIC resistant. I hope there are some informed opinions floating around out there who can help fill in the how.

According to "the internet" there is an ASIC based ethereum miner on the horizon.

this may be the original source of that news

If you believe the analysis in the comments on this reddit thread BitMain may already running these miners.

I believe it is the accepted wisdom that ASIC based mining leads to increases centralization when compared to GPU mining.

This leads us to two questions:

1. Should we hard fork to make ASIC mining harder and to demonstrate a willingness to hard fork any future ASIC based ethereum mining.
2. What specifically changes do we make to implement this increased ASIC resistance.

Should we fork?

I propose that people indicate support/opposition with a simple 👍 / 👎 on this main issue. I would prefer this conversation not devolve into deep discussions around this subjective topic so my request is that people refrain from commenting on that specific question here.

How do we implement improved ASIC resistance

This is the primary issue that I think needs to be addressed, after which we can have an informed discussion about whether we should actually do it.

[pdaian]

At the end of my text, there can only be one conclusion: Hardfork, now! Either we change Dagger so that ASICs wil brick, OR we move immediately to PoS (which won't be here yet)

There are quite a few opinions in this thread expressing a doomsday-like forking fervor from people who are already mining (which I view as a conflict of interest), and that worries me from a governance standpoint.

To comment on the "how" portion of this issue, I don't recommend slapped together changes. Please seek the review of at least 3-4 professional cryptographers; questions of efficacy against ASICs here and even just safety against attackers are difficult, beyond my intellectual pay grade, and probably beyond the intellectual pay grade of 99.9999% of the developers inside the community.

Also note: does anyone have actual details of the hardware of the ASIC in question or its optimizations? This is key for evaluating any proposed change to target specific hardware (which is what this is). If Bitmain can reuse most of the same R&D and reconfigure their production a bit, a fork would be giving them an even bigger advantage in selling these miners and pre-mining with the next crop of chips. In general I oppose this kind of hardware-targeted change, but if the community strongly supports, what can I say. I don't think the proposed constant changes will prove sufficient to defeat the general optimization techniques in the E3; I doubt there is much DAG-specific optimization there, and if there is I wouldn't be surprised if it were generated programatically and trivially reusable for any hash algorithm in the same family.

[CryptoBlockchainTechnologies]

I want to first say, if you are here trying to help Ethereum develop a better ASIC resistant algorithm thank you, please do not take wrong what I am going to say.

This thread has asked for people to come forward with ideas that could help create an improved ASIC resistant algorithm. We need to assume that Bitmain could try to influence our direction to make it less or even not effective by bringing their own people in here to push "solutions".

So all I am asking is that Piper you personally vet the people here that have influenced the decisions to ensure they are not linked in any way to Bitmain. At the end of the day we are putting a lot of trust in these guys so we would hate to see that trust misplaced.

[tonymorella]

I know this is a little off topic but I am part of the https://www.rendertoken.com/ (GPU 3D decentralized proof of render) community and while there are still many unknowns it seems like OTOY/RNDR (plus other like work based utility based tokes) will pull a bunch of Nvidia cards off ETH when it goes live late 2018. If the value proposition holds true I could imagine a scenario where it would create a centralization of ETH miners based on "ASIC" that are typically going to be used by larger miners (and not usable for CGI) vs GPU's that are spread around small, medium miners (like myself) and large miners alike, starting around Q1 2019. FYI the roadmap also calls for AMD support sometime in 2019. Comments?

[pdaian]

So all I am asking is that Piper you personally vet the people here that have influenced the decisions to ensure they are not linked in any way to Bitmain.

While this does get at a core issue (regulatory capture of hash selection process by miners) with repeated anti-ASIC PoW change, I don't think it's fair to pin mitigation responsibility on Piper. He's just one dev, and has no real way to evaluate whether anonymous Internet people who may have contributory ideas do or do not have relationships with Bitmain (or any other equally serious conflicts of interests). Proceed with caution.

[AirSquirrels]

From my side, I’m just here to help. Anyone else is welcome to lead the charge if someone better vetted needs to step up. For the avoidance of double, my identity is well published as is my company (airsquirrels.com) We make screen mirroring software/hardware for schools and businesses - David Stanfill. I’m active in the number theory community on mersenneforum.org as well as part of the verification team of current largest known primes. Feel free to actually reach out directly.

[rorsch]

Do not forget miners. Many of you are attacking PoW mining with moot points when miners / hodlers are the strength behind Eth. Miners have much invested and have provided many industries substantial capital for creating a new business sector.

Bitmain is the issue. Asics are a means to disrupt the people's power to become involved while corporations are stealing away the network. The PoW era is pushing tech to new limits and need for more efficiency for gamers/miners - both. We need to band together and work towards a hybrid method with PoS/PoW in a way to prevent Asics from ever existing. Perhaps PoS can some how validate the miners PoW to determine if its Asic? Maybe use CPU power to encapsulate GPU work as a signature? Or is there something in bios to identify cards and determine manufacturer info / serial number as a validator? I am not a low level hardware expert but feel there has to be a simple solution staring us in the face.

[pipermerriam]

@CryptoBlockchainTechnologies thanks for voicing your concern. Our next step is a formal EIP which will lay out specs for changes to the mining algorithm at which point I'll be seeking broad input from experts on the merit of that spec and assessments of whether it's likely to brick existing ASICs (even though we may not know their specs).

[ouillepouille]

Hello ! Just passing by to say that the Monero community had the same discussions about 2-3 weeks ago (and is still discussing, of course) when it was made public that Bitmain had been using (and was now selling) a Cryptonight ASIC... I published a concept that, AFAICT has core similarities to what @AirSquirrels is proposing. I have exposed the idea in 2 or 3 reddit threads, but the essential is here.

Disclaimer : I haven't coded in many, many years and I used to code at a high level (Cocoa). So this is all purely conceptual without any real computer-science backbone. Essentially, food for thought. If the PoW code requires to emulate and leverage the generic capacities of a CPU, we are closing the gap, and/or ASIC designing actually becomes profitable to the entire world of computing. So, maybe we need to take an inception-style approach to fight ASICs. But, as @AirSquirrels said it already, it is probable that this isn't the proper route to take (too complex, too many risks, too slow)... but it probably deserves more experiments and maybe even draft implementation.

Keeping mining decentralized is just essential in my view... it is one of the core features that makes crypto interesting in my book.

PS. : People here might want to take a look at the work done by curie-kief on the SumoKoin PoW. Fireice-uk (= curie-kief) claims it is more ASIC-resistant than the new Monero PoW code.

[jjviscomi]

I think responding to this externality with a means of trying to subvert it is a poor decision, unless there is a real imminent threat to the overall system, and there should be a well thought out plan for events like that. Everything right now is speculative conjecture ... not a very good "strategic plan".

[AirSquirrels]

@ouillepouille it is worth noting CryptoNight7 essentially didn’t even try to make the algorithm more ASIC/FPGA resistant. They merely tried to brick the current ASICs and make minor mask fixes impossible by introducing new information dependencies - to ensure a full ASIC R&D cycle. The same reason I suggested changing all three constant multiplication FNVs to separate values.

The cryptonote-heavy design is interesting but there are some points in it I disagree with from a hardware design standpoint, at least at first blush. At least in designs I’ve looked at for Cryptonote the AES round calculations turn out to be the barrier well ahead of scratchpad/memory bandwidth. You can’t get enough parallel pipelines to exploit the bandwidth you have already. Lowering from 513k to 16384 seems... an odd suggestion.

The vulnerability in the implode step is real, but that’s specific to XMR. Honestly not a lot of that is relevant to ETH, which remains significantly harder to compromise/optimize as is.

The FNV changes are proposed for the mining loop, not the DAG item generation, because they would make the mining calculations wrong in the current ASICs. We all agree this wouldn’t prevent new ASICs from being made, but it would brick every existing chip.

Suggestions have been made for more detailed and specific changes, but the argument was made that could be done in the future as the solutions are better understood. There are also next to no optimization’s going into the marketed E3 to make that hashrate. It’s just dedicated chips and memory on a board.

As I see it getting the E3 in hand won’t give us any more information than we have now, unless someone is planning to de-lid and electron microscope scan the chip and reverse engineer it.

[pdaian]

The FNV changes are proposed for the mining loop, not the DAG item generation, because they would make the mining calculations wrong in the current ASICs. We all agree this wouldn’t prevent new ASICs from being made, but it would brick every existing chip.

It really bothers me that this is being sold as "Modify block mining to be ASIC resistant" (see title) when all it is admittedly doing is discriminating against one particular model of ASIC (note that it is not even discriminating against a design, since the same design can be trivially re-used).

I'm OK with proposing a fork that discriminates against a single hardware model in this way, but can we at least be politically honest about it? The naming of this EIP is objectively inaccurate, and quickly entering "PATRIOT Act" territory when dumb users think this buys them some actual ASIC resistance.

I also have no idea what specs you've seen, but I don't personally trust them to be accurate.

[goobur]

The goal is to currently kick any ASICs off the network and force a redesign -- a more thorough analysis of ASIC resistance and and a proposal to redesign the algorithm will likely come next.

[pdaian]

The goal is to currently kick any ASICs off the network and force a redesign -- a more thorough analysis of ASIC resistance and and a proposal to redesign the algorithm will likely come next.

Whether such a redesign is even possible is a scientifically open question. If this is merged under the name "Modify block mining to be ASIC resistant" and the first proposed change is specifically E3-elimination, I'm going to be fairly upset and vocal about what I see as the rug being pulled out from under users with slick wording.

I propose naming any EIP that contains any mining loop, DAG structure, or other hash algorithm changes designed to target the E3 to make this clear. If the plan is to have this be the meta-issue and have the specific changes proposed here spun out into EIPs with more appropriate titles, I can support that.

[DrizztVD]

Would it not be best to wait-and-see? Bitmain seems to be trying to force wide availability of the chips by only allowing one unit per user. No idea how circumventable this will be, but may certainly prevent centralisation. ASICS are not the problem, centralisation is. So we do should not be working on trying to prevent ASICS, since THAT IS NOT THE PROBLEM.

[AirSquirrels]

@pdaian re: specs, you did see that these are actually for sale now right? The specs are known. Also the change would stop all current existing ASIC chips from all manufacturer, known and unknown, which are the time consuming part to produce. New board designs could always be done quickly.

I personally agree and would rather see long-term resistance, but the signaling here has been that that is a big change unlikely to happen in a reasonable timeframe.

[jjviscomi]

@DrizztVD I am with you ... Bitmain will make money by selling their Ethash ASIC miners. The effect might be an unplanned hard fork to no or little avail. Cycle could repeat and Bitmain is the winner, not because of centralization but because of causing chaos and destabilization for Ethereum. Let the market decide ...

[Arachnid]

@AirSquirrels The FNV changes are proposed for the mining loop, not the DAG item generation, because they would make the mining calculations wrong in the current ASICs. We all agree this wouldn’t prevent new ASICs from being made, but it would brick every existing chip.

Are we sure about this? Given that ethash is not computationally-bounded, it's not unreasonable to assume they might have used some spare ASIC surface area to build in more programmability than they could if they had to optimise purely for compute cycles.

As I see it getting the E3 in hand won’t give us any more information than we have now, unless someone is planning to de-lid and electron microscope scan the chip and reverse engineer it.

On the contrary - looking at the board and analysing the firmware would both tell us a lot about the construction of these miners and their limits.

@pdaian This issue (not an EIP) was opened by @pipermerriam to explore potential ASIC-alleviation options, without one particular one in mind. Specific proposals have focused on bricking existing ASICs, but this topic is broader than that.

Were someone to open an actual EIP on a specific proposal, I'd be right there with you making sure it's accurately named.

[dadyal]

Yes for sure ASIC is not good for ETH we do not want another bitcoin always on thredt by china and other countries with ASIC miners

[AndreaLanfranchi]

Resistance or resilience ? Literally any algorithm can be turned into ASIC : it's only a matter of time. Thus "Resistance is Futile" (cit. the Borgs). ASICs on one side are more efficient with better Hashes/Watt ratio and, somehow, may help address the energetic footprint of mining.

If I were to plan my revenues out of a PoW mining enterprise (no matter the size) why not consider specific machinery instead of generic one ? It makes no sense. At the prices of E3 (800 Usd - right ?) plus power supply why should I buy a heavy and complex to handle mining rig for at least 3x price ?

Real problem is concentration : wether or not E3 will be sold in large stocks to big players is yet to be seen ... but for sure Bitmain has already taken all the advantages they could while keeping their miners silently operating for their mining ops.

Like it or not as we talk about money (as this is the underlying aspect - ah ok let's call it incentive) concentration is nowadays inevitable. Bitmain is not what they produce: is what they mine before going public.

Resistance is futile ... resilience is temporary. Either we change mindset (and logic) about mining to render it "concentration-resistent" (if ever possible) or these discussions are meaningless

P.s. I personally do not think PoS will do the job either.

[ouillepouille]

@AirSquirrels @pdaian If the PoW includes at least a pass where generic CPU capacities need to be taken into account, maybe it'd make ASIC-designing for PoW less relevant.

An example would be to use the [code]->[opcode] translation done by a Virtual Machine as a form of pseudo-hash-pass (e.g. cryptohash the resulting opcode for instance), and have another pass execute the actual opcode inside the VM (and pass the returning value for further hashing). The main issue would become speed ? Because alongside these non-crypto-(pseudo)-hash functions you will want to keep hard crypto hash rounds.

Anyway, that was my idea, not really sure it's feasible or even a good idea. The core principle behind it is to make ASIC-designing for PoWs become akin to designing CPUs (which is a bit absurd).

I'm out as a contributor of this discussion, but I will remain a reader !

[AirSquirrels]

@Arachnid I agree - I've long indicated that the outcome of this war on ASICs is going to be a larger reliance on programable logic by Bitmain et. al, either as configurable parts of the ASIC or through FPGAs. I even suggested above that the E3 might be using FPGAs instead of ASICs between memory chips. For that reason I suggested larger scope changes, but in general the feedback has been that we would want to spend time (months?) analyzing those due to risk and the prevailing views in this thread have asked for minimal changes that would primarily be to send a message.

Most ASIC synthesis would treat multiplication by a constant in the FNVs an optimized operation and not a 32x32 multiplier with a configured value. Certainly they could have done it with a configurable value anticipating it may be an easy breaking change, but it seems unlikely they would take all those steps. It seems even more unlikely they would expect each of the three (or up to 7 if we wanted to split to compress FNVs) constants to change separately and anticipate/make that configurable unless the entire chip was configurable logic.

What firmware do you hope to analyze? If these are truly ASICs they don't have much firmware, and the control board firmware isn't likely to give many hints on optimizations. If they are FPGAs they would be foolish not to use bitstream encryption. It just passes headers and nonces to boards and gets back winners.

The physical construction would be nice to analyze, but I would be genuinely surprised to find anything other than memory chips connected to ASIC or FPGA chips. To be honest the count of memory chips and type isn't really something that would help us know what optimizations they would have taken. 180 MH = 1.4 TB/s collective memory bandwidth, there isn't any way around that. The large chip count of memory chips is supported by the large form factor of these boxes, and GDDR5 or DDR3 are the only commercially viable variants. Knowing which doesn't really provide us any information useful for resistance.

Now it is true that we could find out they've used GPU chips or FPGA chips paired to their memory bandwidth, and minor changes would be ineffective. With that said, nothing is going to be effective if they are GPU chips (unlikely coming from Bitmain), and for FPGA chips or configurable ASICs our only option is to drastically change algorithms to ones that significantly reduce the number of (still memory bound) hash pipelines one can place on a chip to the number of cores (true cores, not SIMD units) feasible in silicon for CPUs/GPUs.

[romka1977]

Just a thought - with GDDR6 adoption on the horizon in GTX 11 series these cards might have a comparable performance/efficiency/price of E3 ASIC miner....so it might be no benefits for these ASICs except easier operation.

[Arachnid]

A reminder that the subject here is potential changes to or replacement of ethash to make it ASIC resistant. This is not the place to express how much you hate bitmain, or whether you think making a change is a good idea; use upvotes/downvotes for that, or take the discussion to Reddit.

[ofpcarnage]

Apologies

On Tue, 3 Apr 2018, 20:28 Nick Johnson, notifications@github.com wrote:

A reminder that the subject here is potential changes to or replacement of ethash to make it ASIC resistant. This is not the place to express how much you hate bitmain, or whether you think making a change is a good idea; use upvotes/downvotes for that, or take the discussion to Reddit.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/ethereum/EIPs/issues/958#issuecomment-378368024, or mute the thread https://github.com/notifications/unsubscribe-auth/AWHySlO0AVrHk_L6rC79HF7Ij4xtGJA-ks5tk81cgaJpZM4TAuxE .

[quyquan]

https://www.coindesk.com/bitmain-confirms-release-first-ever-ethereum-asic-miners/?utm_content=buffer61d78&utm_medium=social&utm_source=twitter.com&utm_campaign=buffer

https://shop.bitmain.com/product/detail?pid=00020180403174908564M8dMJKtz06B7&utm_source=twitter&utm_medium=social&utm_campaign=e3-announcement

[andre-brongniart]

I left a comment here. I thought it was an open discussion. Not a centralized controlled discussion... So much for ETH and being naive about that. Bottom line is Bitmain has the Asic and has been mining ETH way before any news leak about this. If it is indeed an ASIC with GPU's in it, a fork of ETH will not change anything as far as I know.

[int03h]

I've been reluctant to chime in on this- mainly because there has been nothing concrete to go on. Technically I doubt there will be any SIGNIFICANT advantage, unless they have managed to overcome the laws of physics and/or come up with some sort of brand new architecture. Moving bits around on a bus or on silicon is an expensive business. ASICS are great at doing fixed tasks, but flipping bits is all the same thing regardless how you do it. Fast/Wide/Fast and Wide .. whatever .. its all about charging capacitors. Perhaps they may have created something which foregoes the need for some of the more expensive components on a graphics card that ETH doesn't use, but the cost of memory remains along with the laws of thermodynamics used to keep those caps charged and those bits flipping. Anyhow - I imagine the simplest solution would be to increase the size of the DAG file to the point where it is bigger than the available memory on these miners. Yes, it will more than likely hurt the GPU community too, but it is the simplest way to "correct" this "problem".

For those of you who are new to the ASIC game - over promising and under delivering is how it done. "Unexpected" problems ALWAYS arrive a few weeks before deliveries are promised. I have ZERO doubt that this will be delivered late.

Personally, I think Bitmain has managed to find a way to make money simply by shorting ETH, NVIDIA and AMD stock and announcing a half baked product which WILL be late. Bets anyone?

Finally, the only risk here is Bitmains ability to "retard" the network like they have with Bitcoin. i.e. controlling consensus and slowing the network down because they are behind the Chinese firewall. We need to make sure we clamp the QoS on the network protocol to avoid the same situation. (if it isn't already).

IF they have managed to innovate and they do deliver ontime, then GPU's are obsolete and competitors will come onto the scene - cool! It will still take a MASSIVE investment to grab 51% of the power from the network.

[7runks]

@dreusskis, that is what I think too but like I said before, that is not the real eth asic from bitmain, the real bitmain eth asic miner is hidden so that is why is important this fork.

@int03h, yes "Personally, I think Bitmain has managed to find a way to make money simply by shorting ETH, NVIDIA and AMD stock and announcing a half baked product which WILL be late. Bets anyone?"

Bitmain announced 45 days ago and eth price crashed 70% and now bitmain announced the real specs of its useless eth asic miner then price goes up.

[int03h]

@7runks If it is anything, it is not a GPU/ASIC, but a bunch of memory controllers stitched together. Technically a memory controller is an ASIC.. but w/e .. lets just use the term as loosely as possible since this is a fact free zone anyway.

I am sure you have noticed the price of all cryptocurrencies have been taking a hammering. You can't seriously be suggesting that Bitmain is the cause of this ?

[Nsomnia]

Would love a hard fork just enough code/algo changes to throw off ASIC mining, although for the sake of humanity and gamers, all mining should be GPU only if it were possible, unfortunatly not. Been thinking lately about how many fossel fuels are burnt for mining which is crashing lately.

[AliAshrafD]

Obviously it is a hoax, Bitmain's claimed ASIC attack on Ethash. check this post of mine in bitcointalk. Main point: Bitmain is playing dirty. By claiming ASIC vulnerability of ethereum's native PoW, it just is manipulating chip manufacturing industry in its interest. @AirSquirrels is 100% right about the inherent ASIC resistance built into ethash. A hypothetical ASIC that is implementing Ethash has no choice other than accessing RAM and the memory bus, (forget about putting 4+ gigabyte of ram on the die it is just impractical ) this configuration can't be significantly more efficient than a gpu installed on a conventional graphics card. In other words: Ethash is a memory access bound algorithm, speaking of ASIC here is irrelevant, you can just claim having better memory access technology if you actually have it and you are a fair player. Neither is true with Bitmain.

[cryptomined]

One thing we have going for us is bitmain states "There are financial risks associated with mining cryptocurrencies. These risks can be related to changes in exchange rate of the cryptocurrency or to changes in the algorithm that is used to mine the cryptocurrency."

Which means that we can change the algo and eliminate their ASICs... we just have to figure out how...

Now that the E3 has been released, expect all the shills to come and try to prevent a hardfork from happening...

[masonticehurst]

The whitepaper specifically talks about how ASICs promote centralization. This, of course, goes against the core principals of Ethereum or any "decentralized" currency for that matter.

"However, this mining algorithm (SHA256 presumably) is vulnerable to two forms of centralization. First, the mining ecosystem has come to be dominated by ASICs (application-specific integrated circuits), computer chips designed for, and therefore thousands of times more efficient at, the specific task of Bitcoin mining. This means that Bitcoin mining is no longer a highly decentralized and egalitarian pursuit, requiring millions of dollars of capital to effectively participate in"

[bernardpeh]

Its official. bitmain already have ethash miners.

"The Antminer E3 is estimated to deliver a minimum hashrate of 180MH/s while consuming 800W of total power.

The very first batch of the Antminer E3 is available to order now and all orders will be shipped as soon the miners are ready. Although this is estimated to be after 16 July 2018 (and before 31 July 2018), we will ship all orders as soon as the miners are ready. "

[QuintLeo]

The E3 has about the SAME performance and efficiency as a well-tuned 6 card RX 480/580 rig with BIOS modes for memory straps. It has slightly LOWER performance and efficiency as a well tuned 6-card GTX 1070 rig (no BIOS mods needed). Bitmain would have to ship well over ONE MILLION of these miners to double the current ETH network hashrate, more like TWO MILLION to double all ethash total netwok hashrate combined. For perspective, they have probably shipped more than One Million of the S9 but nowhere near Two Million - and have had almost 2 YEARS to manage that.

Waste of time to change the algorithm to fight this miner, they won't have a serious affect on GPU mining profitability if POS gets implimented in the next year or so nor will they have much if any effect on centralization at the low price ($800) they are charging, which is quite a bit cheaper than building a GPU rig of the same performance level AND THUS MAKES IT EASIER FOR SMALL MINERS TO BUY IN.

[yai333]

Antminer e3 batch 1 sold out already.....

[tarzanbigcity]

I would recommend looking at the x16r and x16s algos as they are higly tuned for asic resitance.

[7runks]

@QuintLeo, come on, what they will be doing here is to stop hidden asics which exist already. Bitmain really wants what you are saying, reason they are sending this trojan horse. Bitmain does not want the algorithm to change, period!

[Buttaa]

@7runks @yai333 @QuintLeo @bernardpeh
The vote whether or not to fork has been done on the first post on this thread by simply using 👍 / 👎. We should stop spamming this thread, so the dev team doesn't need to read through all of this to get the valuable info, which will waste even more of their time and probably delay the development. just my 2 wei

[pipermerriam]

@Arachnid or someone with permission. I propose we close and lock this issue. I've got an actual EIP pull request incoming soon where we can hopefully have a more technical discussion without all the noise that this issue currently has happening.

[SIN3R6Y]

I would highly suggest reading my thoughts here...

https://www.reddit.com/r/ethereum/comments/89jkv6/is_an_antminer_e3_hardfork_a_waste_of_time/

It's likely that the E3 is not asic based in my opinion. This may be a wasted effort.

[7runks]

Hey, you all knew what would happen here, don't blame people for that. Should have opened 2 threads, one for technical discussion and the other for free speech.

The reason people came here is to tell the devs themselves how they feel about it. Writing here means for them their word can be heard.

[jamesray1]

Just for context I asked @pdaian to join this discussion on Twitter, as we need cryptography experts to provide suggestions and analysis before we included changes in a hard fork. He is a student involved with IC3 at Cornell. http://www.initc3.org/people.html. Ari Juels commented via Emin Gün Sirer: "Ari points out that there are off-the-shelf solutions such as Argon2, Equihash of Zcash fame, and Balloon Hashing. We do not have anyone in IC3 who works on memory-hard hash functions atm." https://twitter.com/JamesCRay01/status/981051621424558081>

Here is Phil's comment on Twitter, linking to his comment here: https://twitter.com/phildaian/status/981194693185765376.

I propose naming any EIP that contains any mining loop, DAG structure, or other hash algorithm changes designed to target the E3 to make this clear. If the plan is to have this be the meta-issue and have the specific changes proposed here spun out into EIPs with more appropriate titles, I can support that.

@AirSquirrels you keep saying E3, but it's apparently called the F3.

http://technews.cn/2018/02/12/bitmain-tsmc-dram/ https://www.youtube.com/watch?v=9L-1iG6mdJ8 https://steemit.com/technology/@cloh76/bitmain-to-release-ethereum-asic-miner-the-antminer-f3

Edit: it is called the E3:

https://www.coindesk.com/bitmain-confirms-release-first-ever-ethereum-asic-miners/?utm_content=buffer61d78&utm_medium=social&utm_source=twitter.com&utm_campaign=buffer https://shop.bitmain.com/product/detail?pid=00020180403174908564M8dMJKtz06B7&utm_source=twitter&utm_medium=social&utm_campaign=e3-announcement

[thunder02dragon]

@AirSquirrels @cryptomined I tried to not comment but going through all these made me get an account and comment here. Yes it's official that Bitmain has ASIC now called Antminer E3 with a mere 180MH/s @ 800W. But IMO these are just rebadged version 1.0 of their ASICS that they wanted to get rid of and profit from. Else why on earth they would sell them for mere $800 where a 180MH/s mining rig costs ~$2500 at minimum to build. I am sure they are well aware of this thread and hence planned before any concrete decision is taken.

I am pretty sure that Antminer F3 is real and already behind the huge ETH network hashrate increase. If you want to get some idea about the hashrate increase.. please go ahead and check the Network Hashrate growth chart from Oct'17 until Mar'18. Until Oct'17 to ~ 1st week of Dec'17 the Hashrate was flat which I understand slowly the GPU rigs came online but after Dec '17 see the growth... Unless there are huge machines / ASICS the jump on hashrate from 120TH to 270TH on 26th March 2018 is impossible which is like 2.2x of the total hashrate in ETH Network. Do mind 120TH is the max hashrate entire ETH network had back in Dec and I am sure AMD and nVidia didn't made enough GPU's to double the network hashrate in mere 4Months. - IMO - It's no way.

So based on these stats I fully support the hardfork and would like to keep ETH ASIC resistant. However I do believe the day hardfork takes place if at all and ASICS do breaks... the ETH network hashrate should be around 180TH which is what expected based on past statistics.

To add - have a look on BITMAIN's SOPHON AI Cards.. They are AI cards only for specific purpose they claim. But is that the case? - I don't think so!! They are BM1680 and FPGA chips and should be more than of one use... Well again the day they launched I did put my concern on the same release twitter page. @AirSquirrels Have a look at the product. They have 16GB and 32GB memory respectively but DDR4 but again if we have any plans for memory heavy loops that might not work - Again I am not expert but just speculating.

[yhhca]

My two cents, F3 and E3 are both existing. but bitmain is worrying those ASICs will get banned by hard fork. So they just push out the lower model to ease people's nerves, then they can mine behind by using the high end model. The whole eth network will be controlled by them step by step. The more eths they have the more power they get in the future. Thanks to the PoS.

[crossx4real]

not an expert in this field, not by a long-shot but i am heavily invested in ETH, and hate to see it suffer the fate of Sia which is a good coin.

if ASICs can't browse the internet or open a text document, we should exploit this weakness. Gentlemen and ladies, this is war, make no mistake about it. The network can authenticate each mining rig every x amount of time by sending it a file and requiring that details of that file be reported back, rigs that are able to authenticate will continue to earn valid shares while ASIC miners that join the network can contribute to the overall hashrate of the pool but will get zero payouts. in other words big payday for GPU miners. this encourages GPU mining, discourages ASIC mining.

Be a cutthroat, don't be a sheep.

[AirSquirrels]

@thunder02dragon sure - the Bitmain Sophon AI cards as specified would have 166 GB/s of bandwidth, and would be capable of 20 MH peak/each at 25-41W. They also have huge 50 GB/s interconnect bandwidth and 32MB internal cache in a 1599 ball package. That's also not a cheap chip to manufacture by any stretch of the imagination. You're looking at 20-25 of these chips to achieve E3 power usage, and a (unlikely) minimum of 9 to achieve hash rate. For reference their SC1 card shipping in Nov. 2018 is set to cost $589 and contains just one of these chips with 4 channels of DDR4. I don't see them feasibly building and packaging that chip + memory on boards for a mere $89. Most can't even buy the required DDR4 component chips for $89.

It is possible that if they are not ASICs the E3/X3 are based on a prototype chip or older version of the BM1680. With that said, if they are building legitimate processors as a third entrant to the market, then I'm not sure we should be fighting it vs. allowing competition to happen with Nvidia and AMD. Furthermore I'm not sure we CAN block them in that case, not without a very thorough differential analysis of the specific architecture and capabilities of the accelerator card vs. identical functionality provided by AMD and NVIDIA.

It remains worth the (low) risk to implement a preliminary hard fork to guard against other ASIC builders and potential manipulation of information to prevent bricking of their miners.

I do personally generally agree with arguments from @pdaian in this article once you get beyond the one-purpose-hardware https://pdaian.com/blog/anti-asic-forks-considered-harmful/

[krtschmr]

@QuintLeo

For perspective, they have probably shipped more than One Million of the S9 but nowhere near Two Million - and have had almost 2 YEARS to manage that.

actually since december we have about 1 Billion$ , slightly more, added in mining hardware. The BTC Network hashrate added about 17,000,000 THs. on a regular 13.5 THs S9 that would be 17M/13,5 roughly 1,300,000 x S9 Asic. and that's just since Decembers highscore till now !

The selloff of ASICs just recently bagan

Bitmain is easy capable of doing 500,000 of these E3 if people buy it. would be 90 TH/s and add 30% network right now to ETH

[int03h]

Hash has increased for 2 simple reasons : 1) Nvidia used to suck and now it's OK, ie more fabs = more silicon 2) Vega with HBM is better than Polaris

If you factor in some modest ability of both vendors to increase production to meet demand the steep slope doesn't have to be some magical fantasy.

It COULD be .. but I don't think so.

[MicahZoltu]

REMINDER: Ethereum's PoW algorithm was designed to be ASIC RESISTANT, not ASIC PROOF. Ethereum's PoW algorithm is ASIC resistant and has achieved that goal quite well considering it has been years without any known ASICs and the first ASIC to be announced isn't that impressive compared to a GPU. Arguing that Ethereum must change its PoW algorithm because it was "promised in the paper" is not a reasonable argument.

REMINDER: Making miners happy is not a primary goal of Ethereum. Its primary goals are censorship resistance and trustless interactions. ASICs are fine as long as no single actor (by behavior) dominates PoW mining. We only need to make it until PoS, so whatever strategy we embark on only needs to keep mining out of the hands of a single dominant miner until PoS. Note that even with ASICs, due to the resistant nature of EthHash it would still take a lot of capital invested in Ethereum mining to control 50% of hashing power and anything custom-built for EthHash will become worthless once full PoS launches. This is a huge financial risk for anyone looking to take over 50% of Ethereum hashing at this point using ASICs.