EIP-3368 | A more critical look at GPU Proof of Work Security
An understanding
This model write-up is going to consist of a range of potential scenarios that if played out could have a range of effects within a proof of work environment on and around Ethereum. Before I get into the content, I want to provide context of the way I personally approach the formulation of a security risk within a cryptocurrency blockchain that is based in Proof of Work. As a long time GPU miner I have experience with CPU, GPU, ASICs and FPGA mining and risk associated within the space. My perspective remains attentive with brining maximum participation to proof of work mining through multiple mediums and social media, guiding individuals to the tools and starting knowledge on how to participate.
When I was reviewing the EIP-1559 set of changes, the majority of the changes are great for Ethereum. Early on, I came out against the Feeburn as a the initial premise presented was ‘this is good for the hodl, it helps the SoV” narrative. I know now it’s more than just a monetary move and by design some of the gaming mechanics indirectly create an environment where the incentive would be to game the new mechanics to always max the blocksize and not change much in the basic behavior within the mechanics. I rescinded my negative opinion to EIP-1559 and started to move on. This was when I started to run some numbers to see how much potentially would bleed off ethereum if the cut was to much, what networks start to get the hashrate if something like a price drop was to occur at the same time this fee burn went in. I put forth a first model attempt at taking 10 various scenarios that could play out, with a majority of them (7 of 10) not affecting ethereum that much at all.
I have said, on multiple occasions, these few scenarios are highly unlikely, but still, a larger issue lurks in the background related to total network size and growth; more on that further down below.
Much of the base calculations in all threat models are from the aspect of a few key variables for an individual mining algorithm, in this case (Ethash), with focus on singular token (ETH) as example. This normally centers around yield (Block Reward + Basefee + Inclusion Reward), respective difficulty on the network (calculation based on total hashrate vs. block target time), in Eth’s case 15 seconds and a spot price rate for the token. Ethereum established early on it wanted to maintain a MNI Minimal Necessary Issuance, to ensure token inflation was at a moderate rate (4.5% with 2.0 BR) derived as a ‘security’ budget to maintain best in class Proof of Work incentive. Ethereum’s monetary policy aims to curb excessive emission of new currency while balancing an calculated approach at providing an incentive for security enforcement on the chain through hashpower. To date, Miner’s have not challenged this and while some to be expected grief on twitter, the world has continued to turn without incident.
The excitement and growth
As Ethereum’s price discovery has found new heights in late 2017, the depth of participation had grown through 2018 as a rolling wave of new miners came online, peaking just under 300TH by mid 2018. The seemingly common cycle of volatility suppressed the need for more miners throughout the remainder of 2018 on through early 2020. During this time Ethereum’s network and many other GPU enabled PoW algos reduced the need of mining through lower market price and additional yield cuts (Ethereum’s Constantinople updated taking issuance from 3 to 2). The bullrun of late 2020 however, coupled with significant increases in total yield due to the success of Defi and NFT space, Ethereum has created a significant opportunity for new participants in nearly all regards.
Extensive Expansion in Mining
Due to Ethereum’s current mining algorithm, Ethash, the newer generations of GPUs higher speed memory and bus width along better efficiency that enables smaller footprints with much higher hashrates. As the global supply chain works to catch up, a significant a range of new participants across the world have joined in the opportunity of expanded yields and price discovery. This modern day gold rush creates organic attention for Ethereum and benefits from the significant uptick in news, stories and social media exposure brought to the chain from these activities (and apparently the negative too). Ethereum success has not only brought nearly 2.5x increase in total nethash, but many other networks now have also increased exponentially their participation and nethash. Ethereum currently rest well ahead, leading the nethash within the shared GPU space by a significant margin.
As the bullrun continues, more are drawn in as participants. This typically is a great thing, more organic growth, more having a one to many connection with other individuals that may partake in mining, or may just take a position in the coin itself. The success of Ethereum’s growth expands its participation and we see that with a 430 Terahash nethash with no sign of slowing down. The expansion at its current rate will continue to grow as there is depth in the price and yield to allow more participants the opportunity. The barrier to entry is low, creating a low friction, easy way to start. Individuals can take their existing hardware or go out and pick up a GPU to become part of the ecosystem. (Yes, easier said than done right now with lack of supply)
The sleeping dragon
With the extensive growth (measured in total nethash), the excitement around the space generates the discovery on how to participate. That discovery often leads people to various aspects of cryptocurrency, including how to mine. Mining within the Ethereum ecosystem is a low barrier of entry to participate as folks with existing hardware can download open source tools, OS, many online resources including detailed walkthroughs on YouTube across multilingual channels. A few hours of research, (for some people getting lost on the internet for an evening), gets them to a point of starting a miner and earning Ethereum as their participation in adding security to the network results in a small payout. This directly increases the entire size of pie in the category of GPU mining (potential) as once an individual understands how to mine and participate with Ethereum, natural progression is learning about ethereum’s roadmap and how mining will transition to Proof of Stake.
This naturally triggers an interest to look for alternatives as a way to understand the landscape. This is the distinct advantage for GPU mining as you are LONG the crypto PoW Proof of Work space, not just one particular token, it comes down to economic incentive to provide hashrate for the other network. This dynamic is both good and can be perceived as bad for any crypto project as the ‘security detail’ is by design naturally selfish for overall profitability. Now, I have always communicated on my channel that there is more to the ecosystem then just mining the ‘most profitable’ coin and one should balance it from a stack of priorities and potential strategies. This includes calculating the likelihood of a project succeeding in the long term, even if it is not the most profitable at the time or is not economically viable, some will operate for the future speculative results. This type of strategy works outside of mining and is the exact rational for someone wanting to become a Proof of Stake recipient too. If you believe in the project and want to support it, you can take a stake in the coin and become part of its validation layer. As part of any calculus, that reality must be accounted for, but not wholesale relied upon if the security risk hits a point of high risk chance of 51%.
The paragraph header “The sleeping dragon” refers to this dynamic that is not easily calculated or a model built against. That model includes a series of events and key variables resulting in a fundamental and quick shift in any one network, especially Ethereum. My previous attempt to communicate this at the ECH Ethereum Cat Herders discussion failed to articulate the risk as the scope remained in the existing Ethereum scope of current and range of future price and yields. One of the key scenarios I brought up depicted a LOW Eth Price (sub $700) + Deep cut of up to 50% of the fees (basefeeburn) (resulting in (BR -Burn @ 2.1ETH) could result in a sizable decrease of incentive that would instantly alienate eth miners as the economic viability would be not there to stay on eth’s network. Calculus assumes this may make other networks more viable and/or the hashrate ending up on a brokerage and sold to the highest bidder.
The reality of what I was trying to convey through a failed attempt was the total size of pie of GPU hashpower now greatly exceeds any previous time in history by a substantial amount, 2.5x at the current press time and that much potential is very susceptible to price and yield elasticity. While the earlier days of GPU mining, the hashrate flowed with the ebb and flow of the market, rescinding as expected when price and yields lowered. Much of that behavior would be expected again if we saw a slowing of the bullish sediment in the ecosystem and price slowly started to decline over several months. Yields are subjective as we live in a reality now that Defi and NFTs are not going anywhere and the activity would remain speculatively high, thus creating additional MEV. Where this concern fully rest is in the unlikely event that a rapid price pullback, coupled with a basefee burn in 1559 provides a one-two punch to participation causing a situation where a would-be nefarious opportunist looks to take advantage of the new exponentially large amount of hashpower providing an competition fee through a brokerage service. While this sounds like an extremely improbable set of circumstances, to discount it as not a potential attack vector is premature. This is exactly how ETC, using the same algorithm was impacted, multiple times. Ethash has no direct protection on a block reorg/double spend event.
Out from left field
These concerns are not just about ethereum’s hashrate, it’s the entire ecosystem of hashrate. The key figure to classify as a ‘class’ of force projection. The classes would be split up across ASIC (by algorithm), FPGA (by bitstream) and GPUs (all). CPUs I am leaving out of this as CPU mining is not an existential threat to specialized devices such as GPU, ASIC or FPGAs. It typically is the other way around, where something that is mined by a CPU (as bitcoin originally was in 2010, which promptly switched to GPUs throughout 2011 and quickly transitioned to FPGA and ASICS by 2012). The GPU scene currently covers PoW for more than 30+ Algorithms (70+ Coins) of which the top 8 have a growing % of the network outside of Ethereums dominance. As many have claimed, not my quote, Bitcoin’s growth raises all boats within the crypto sea. The primary factor here is total network potential and how elastic is the hashrate. The current incentives are supremely aligned right now in Ethereum’s favor meaning there is little to no risk at any double spend/reorg to Ethereum based on the economic incentives greatly out way the nefarious collusion to attack the network; incentives are working as intended. As other tokens grow in popularity and network value, they will continue to shed ethereums PoW participants. This isn’t a discussion on survival of the fittest or what coin is better than another, we can model it in a much more linear way of price * yield vs block time. Due to some coins monetary design, the overall yield amount is significantly higher, coupled with their price has risen (in Ravencoin example, nearly 25x since December) it naturally creates a division of security labor as it were. Make no mistake, I firmly believe Ethereum will remain the primary PoW token until it’s transition to Proof of Stake, however, to dismissed this competition on primary hashing class is another uncalculated risk. As some have said, Ethereum will probably only get in its own way than any other competitor knocking it off the top list. The reality is this potential should be accounted for as part of any calculus when threat modeling.
Cool story and stats, but wtf is up with a EIP about raising the Block Reward?
EIP-3368 was written a few days ago as a potential risk avoidance IF we have one of those bad scenarios play out. I have since corrected the rational to reflect this as an optional path, if negative scenarios started to play out for ethereum. This was created as much of this analysis has been provided to the Ethereum Dev R&D Discord and other mediums, however there was a disconnect on how these type of changes can actually get ‘heard’. Presenting the risk doesnt mean ACD are going to act on the risk and jump in to try to solve everything anyone is concerned about. They rely on the community to put forth a technical proposal and solicit feedback. The further deep dive into the modeling, reviewing total size of pie in the GPU mining space (not fully discounting there could be a significant portion of ASIC miners on ETH either), the risk remains very low, even after EIP-1559 implementation. EIP-3368 is a way to ensure there is a tapered path of hashrate, while keeping the total emission aligned to the same current inflation curve over a two year transition period. This proposal does not make sense IF the low ethereum price and large fee burn scenario do not play out, thus this should not be included immediately after EIP-1559. Nowhere in the EIP does it state it has to, however I have mentioned that if the scenarios look like they are playing out, they may want to consider. There is a high chance this is all a moot point, but to not bring up the risk would be irresponsible at this point. Living through multiple ETC hacks as an example tends to change your perspective. The community feedback is pretty loud on this one, the risk does not look to be clear and present. If this proposal is a non-starter no matter what, accepting full risk of a 51% attack than ever increasing the block reward, then I would suggest as a backup plan MESS is looked at as a viable solution. This would prevent any inflation increase and would inherit the effort already done on other networks. All data associated to this analysis has been sourced and added to the following link. If you are timed out, give it time, someone will eventually leave and let you have access. I am working on transitioning this to a proper website.
https://docs.google.com/spreadsheets/d/138M4R1-_zS-OLBsl2VJeN_anfTSCRCFc6EguYUVG-yA/edit?usp=sharing