ProgPoW Audit Delay Issue

EIPs
61

Happy weekend @Sonia-Chen,

I am responding to several of your comments including:

    1. potential ASICs for Ethereum;
    1. Why Ethereum doesn’t scale with process nodes;
    1. Other information/FUD

1. Follow the Money – why small companies are eyeing Ethereum?

There’s saying in North America that encapsulates what you are pointing out here about multiple companies eyeing Ethereum ASIC’s … “Follow the money!”

To elaborate, Ethereum is the 2nd largest coin and represents over $700M USD in mining rewards. Currently there are only 2 serious contenders, AMD and Nvidia (and counting the weak E3 effort from Bitmain) and in theory, ASIC’s should have an overwhelming advantage over GPU’s. So multiple companies are relishing the opportunity to make bijillions of dollars by building Ethereum ASIC’s.

This is no different than how the market evolved with Bitcoin ASIC development. The path to Bitcoin success is littered with defunct companies, failed efforts and $100’s of millions of dollars of lost investments. I point you to this Bitcoin Wiki which shows:

  • 28 ASIC’s going to market
  • 10 ASIC’s being canceled
  • About 50 ASIC’s announced but status is unknown

Ultimately only Bitmain, Canaan, Innosilicon/Holong and Whatsminer/Bitwei were successful. Even a giant with deep pockets and early access to 7nm nodes like GMO failed and withdrew from the market.

Ethereum is a much more complicated design and has the added cost burden of memory dependency (that I will post on a different thread one day).

As you well know, Ethereum ASIC’s are not easy to build. Even with a superstar designer like your CEO, Chen Min, Linzhi is at least 11 months (Aug 2018 to June 2019) into development with no tape-out date in sight. Assuming you are taping out at the end of July and the earliest you will see “super hot lot” samples is Nov. 7. Assuming another 2 weeks testing and qualification, you are another 80 days from production depending on fab availability. Oh and add another 2 weeks for OSAT (Outsourced Assembly & Test) which takes a little longer due to your use of an interposer and HBM stacked die.

Linzhi’s total project time is “guestimated” to be 19 months (i.e. August 2018 to Feburary 2020). That’s a far cry from your original presentation in September that was projecting April 2019 projection or 9 months (August 2018 to April 2019)

2. Semiconductor densities translate to non-linear improvements

Sonia, my friend, let’s debate using facts and empirical proof.

Your comment makes no sense as Ethereum is bottlenecked by memory bandwidth and not semiconductor logic; therefore, Ethereum hashrate will improve but only to the extent of available memory bandwidth (in other words, zero improvement, unless memory subsystem changes).

Case in point is the 7nm Radeon VII at 90MHs (1400c/1000m) compared to the 14nm Vega56 at 38MHs (1138c/800m). Performance increases by 2.4x which is the increase in memory bandwidth by 2x bus width (ie. 4096-bit vs 2048-bit) multiplied by 1.25 or 1000MHz/800MHz (increase in memory clock).

38 MHs x 4096/2048 x 1000/800 = 95MHs (close to the measured 90MHs)

To reiterate, moving from 14nm to 7nm and increasing the core clock by 23% (ie 1400MHz/1138MHz) DID NOT improve hashing performance beyond the increase in memory bandwidth.

3. ProgPOW and BSV conspiracy

What point are you trying to make about Kristy, Calvin and Craig? Are you suggesting that Bitcoin SV is going to adopt ProgPOW as their POW instead of SHA256 and nChain? ROFLMFAO

4. Help from friends

Send me a PM. We can help you with design of your ASIC and procurement of HBM memories. My team is from AMD and has extensive experience with HBM through 2 generations of Vega design and 8 generations of GPU design. Even the Beatles know that they need to get by “With a little help from my friends.”

3 Likes
62

Thanks @epic.henry for referencing my benchmark. I’m glad I did them getting some truth. I haven’t been active because this whole damn progpow discussion continues to go around in circles, I’m sick of it all.

I want to know a straight answer from the Devs working on ProgPoW Audit
A) Is it 100% agreed that “Audit must be complete for ProgPoW to be implemented?” I’ve heard this thrown around but I’ve haven’t heard it as the official statement for inclusion.
B) Do we have ANY time-frame now because of this company that dropped out to do the ASIC tests?
C) By April 2020 The next “Ice-age” will start effectively putting ProgPoW and the Iceage together for miners meaning it’s damn near disaster. What is the point of the ice-age until PoS is few months out(which it isnt)?

5 Likes
63

@xazax310

Thanks for all your hard work pulling the analysis together. It must have taken days of benchmarking, reconfiguring systems and tuning memory clocks. I thought your article would have put an end to the AMD vs Nvidia argument but the FUD and mudslinging continues. (Btw, are you going to update the data for DDR6 and Navi? I have some boards on order so I can send you test data when I get the cards. Unfortunately, it won’t be an apples to apples comparison as Nvidia uses 384-bit wide memory as opposed to AMD’s 256-bit wide bus)

I too am sick of the controversy around ProgPOW. Ultimately my company, ePIC Blockchain decided to step in and contribute our expertise, which consist of both GPU and ASIC design, to help with the ProgPOW audit. See this post to Hudson Jameson after all requests to emails, PM’s, to ECH (Ethereum Cat Herders) and Least Authority were ignored for a week prior to my post.

You ask good questions since the whole audit selection process has not been transparent as Anlan and other pointed out in this post.

I have little faith in the audit as the goals and methodology has not been made clear. For a more robust audit, the methodology also needs scrutiny and input. ECH and the core devs are not ASIC and GPU experts and, therefore, are not the best people qualified to assess whether the methodology is valid and complete. ePIC offered to provide input and help with the audit, even as a contributor, to fill in the gaps but that offer was also met with silence.

There are a lot of open questions to address in the audit and expertise needed which is not generally available, including ASIC development tools and GPU architecture and driver knowledge including the ability to code to the metal. @timolson noted in this post that the audit is a thankless job, that loses money and needs tools that many firms don’t have. I would also add that the Linzhi Open Chip Design needs to be synthesized to show if their claims are true and not.

I finally received an email 9 days ago that ECH had selected a hardware auditor but that has not been announced yet. Minutes from the core dev meeting #62 stated that ProgPOW is likely to miss Istanbul due to the hardware audit slipping.

Based on current pace of ETH 2.0 development, it is almost impossible for PoS to go mainnet in 2020. The pace of research is slow; there is a lot to be done to ensure security of the network and to prevent a hardware arms race.

EDITS - to fix links and typos

2 Likes
64

Sorry, but unfortunately I think @xazax310’s benchmark is not what is needed, and it’s probably misleading.

First off, it looks like he only tested one program (a single block height.) Any single program may have a non-representative distribution of math operations, so many possible block heights / programs must be tested and an average taken.

Secondly, as I mentioned above, merely optimizing the hashrate & power per card does nothing to show whether the card is near 100% utilization on all resources. The ProgPoW authors could have tuned the loop constants for Nvidia cards, and xazax310’s benchmark would not show the bias.

This is not FUD. I have zero interest in Ethereum politics. I own zero ETH and do not mine it. If you want to claim that the ProgPoW authors have no Nvidia bias, follow the proposal I wrote above, which requires modifying the loop constants and testing GPU resource utilization NOT total hashrate or hash-per-watt. Do the ProgPoW authors’ proposed constants cause higher utilization on Nvidia cards vs AMD’s? xazax’s benchmark does not answer this question.

65

Anyone can follow these steps to benchmark resource utilization. I agree with Tim Olson, we have been very clear on how to benchmark ProgPoW.

Please note, this was, admittedly, on 0.9.2 of ProgPoW, so it is outdated.

3 Likes
66

For those that were curious on a independent review of 40+ different GPUs using 8-10 different settings/configurations per GPU, all recorded live on twitch.tv and recapped on YouTube for any audit/review here is a Google Sheet covering my 0.9.2 benchmark results.

Additionally, on the Channel “BitsBeTrippin” you can search “PROGPOW” and find each of those GPUs testing details.

I am about to go through a refresh of testing with the latest architecture (RDNA based Radeon 5700 XT) and revisit the entire RTX lineup with latest miners avail. If there is something specific the community would like me to cover, I have over 70 different GPU types in my studio. If it makes sense for any selected auditing party to do testing, I will make available this lot of GPUs for their independent testing. This includes nearly the entire lineup from AMD and NVIDIA since 2012.

3 Likes
67

To @timolson point on changing the loop constants or as us less technical have been calling it “turning the knobs”, this was already done. IfDefElses original spec was 0.9.2. Community testing led to 0.9.3, this was not a change dictated by IfDefElse it was a result of a community effort.

1 Like
68

Hey all :wave:
We would like to bring this topic of discussion up during Magicians Council in Berlin. Is anyone interested in to talk about it @greerso @fubuloubu @timolson or anybody else :slight_smile:

2 Likes
69

Hi! :wave:
Which aspect interests you the most?

70

I unfortunately will not be in Berlin, but I would recommend a discussion about this audit and revisiting the topic of “ASIC resistance” and its importance to the Ethereum design as specified in the Whitepaper. This issue has many different opinions on both sides, with sentiment shifting and mutating over time, especially as we learn more about the nature of mining hardware and it’s manufacture.

Thanks for taking the initiative @anett!

71

More than simply “ASIC resistance”, the reason fixed function ASIC’s are currently considered undesirable for Ethereum.

ProgPow proposes to more efficiently convert energy into network security neutralizing the threat of not fixed function ASIC’s but also proprietary GPU and FPGA hardware and software innovations that are cheating the current PoW algorithm. A ProgPow fixed function ASIC should be welcomed as it would only further decentralize the hashing power because it would be no more efficient than 1.2x like cost hardware.

Ethereum has hard forks on the roadmap already. In the unlikely event that, like cost, fixed function hardware successfully hashes greater than 1.2x, Ethash-ProgPow has tunable knobs. Ehash-DaggerHashimoto has successfully kept fixed function hardware at bay this long, only now are weaknesses in the algorithm starting to be exploited, ProgPoW removes these known weaknesses.

The audit should be confirming that ProgPoW does not break anything/create security issue and no more. Trying to prove that something that does not exist could not do a thing does not sound sensible. Ethash-ProgPow is different to Ethash-DaggerHashimoto and is tuneable, that is already an improvement over status quo.

Core devs have already stated that for good reason this hard fork should be separate to any planned fork, discussion around Istanbul is irrelevant. There is no doubt the community want it, all is left is to make sure that it is bug free and secure then implement.

1 Like
72

@salanki @shemnon @Anlan @epic.henry@OhGodAGirl @bitsbetrippin @sneg55 @xazax310 all more qualified than I.

Maybe this guy? https://twitter.com/TrustlessState/status/1148359466070372352

Skip over my opinion in this thread for some links to interesting people on many sides of this discussion.
edit:
(oops, link) Is ASIC-resistance good for Ethereum?

73

The audit should be confirming that ProgPoW does not break anything/create security issue and no more.

Earlier in the audit process there was talk of including an analysis of whether the goals of ProgPow were worth pursuing / possibly counterproductive to Ethereum’s security. Is it pretty accepted now that this won’t be part of the audit at all?

There is no doubt the community want it, all is left is to make sure that it is bug free and secure then implement.

What is the best evidence that “there is no doubt that the community want it”?

I am concerned that because most of the objectors don’t have a personal stake in the outcome, most of them (including myself) stopped participating in the discussions due to fatigue long ago. Yet ProgPow advocates are continuing to push for this. This might give the impression that sentiment is becoming more favorable to ProgPow when it isn’t.

The data I’ve seen seems consistent with ProgPow being favored by a highly vocal minority, rather than the community in general.

If sentiment has shifted in favor of ProgPow, there should be people we can point to who used to oppose it who now favor it. Who are those people?

Btw, I will be in Berlin during the Magicians Council and am interested in discussing ProgPow as it relates to how decisions are made for Ethereum.

74

One thing we all seem to agree on is that Cat Herders need to do a much better job communicating what exactly is going on and give updates.

I strongly disagree with the ‘vocal minority’ in favor, the opposite is true as shown by evidence below. The few very big mouths against aren’t even against ProgPoW for technological reasons but more out of a want for an immediate switch to PoS and ignorance about the serious threat of centralization to the current PoW.

The developers decided 9/2018 then later, after the fore-mentioned vocal minority spoke up, asked Cat Herders to get community feedback. The signals that mattered were:

Hash voting 100% in favor (voter turnout 77.1%)

CarbonVote 93.6786% in favor (29,37,726.9575 to 198237.5702 ether)
Verifiable at https://etherscan.io/address/0x90a3aa83b0da90abe2c33fadbe829cf74dfa59d9 and https://etherscan.io/address/0x975f4ac23fcc9c39228cc20a2d4897c7b9bb39cc

Tennagraph, 73% in favor coin voting, 87% in favor gas voting

All this heavy favor despite many lies, false sponsored attacks, and blatant attempt to manipulate vote outcome by gpushack by adding a false statement and a link to vote to ethos distro’s dashboard (@5chdn deleted the screenshot but I’m sure I can find it if somebody wants to see).

Your previous posts link to outdated information about the benefits of fixed function hardware on a network, since those studies we have seen and had opportunity to reconsider and the discussion has never been about whether the Ethereum network has had a change of heart on its statements made in both white and yellow papers. Even if we were at a point where fixed function hardware had broad and fair distribution, see point made by @fubuloubu.

1 Like
75

I don’t find your evidence that those in favor of ProgPow are more than a vocal minority very convincing. Here’s why:

There are 18 entities that voted, all of them are existing miners who we should expect to be part of the vocal minority who wants to preserve GPU mining.

About 3% of total ETH was represented in this vote – an extremely small percentage. Likely people who really care a lot about this issue are over-represented in the sample. We know that existing GPU miners really care a lot about the issue.

Again, extremely small turnout: it looks like less than a couple ETH was probably spent on gas voting, and only a few hundred ETH seemed to have been involved in coin voting. It’s hard to tell exactly how tennagraph is reporting its results though.

All of this data is consistent with miners being extremely in favor of of ProgPow and being heavily over-represented in the metrics that you quoted.

The tennagraph data actually provides some pretty good evidence that should make us question the support for ProgPow among the broader community:

The one metric that’s hardest to skew with selection effects is the voting of top influencers. Among 27 top influencers who voted, 48% are against ProgPow, 22% are for ProgPow, and 30% abstained.

1 Like
76

I do not understand your point about 100% yes from 77% of the miners. Pools aren’t entities in the way you insinuate. Miners can switch pool if they disagree with the vote.

With most eth is locked up in contracts. The turnout wasn’t too bad and if the vote was a little closer you’d have a point.

“Influencers” are literally the definition of a vocal minority. They made up some ground for the nay but still in the minority.

1 Like
77

Existing miners in general are a minority of the community, so even if 100% of miners favored ProgPow it could still be disliked by most of the community.

3% seems quite low to me. The point is that when there are strong selection effects at work the distribution of votes doesn’t tell you much. Consider a national vote about whether one penny should be taken from every citizen of the US and distributed evenly among 15 people. Those 15 people are going to vote yes, and perhaps no one else will vote at all because the hassle of voting isn’t worth saving a penny to them. So you might see a vote of 15-0 in favor of this plan but can’t infer that it’s a good idea or that almost everyone supports it.

The point is about whether top influencers are more likely to be representative of the community as a whole, as opposed to these other metrics. It’s not completely clear, but there are arguments for why they would be: they’re more likely to weigh in on hot topics even if they have no direct personal interest in the outcome, unlike with the other metrics.

Why would we expect top influencers to be biased either for or against ProgPow?

1 Like
78

Because it’s like asking congressmen and women to weigh in on detailed tech industry issues – they have no clue what they’re talking about, and will probably make poor decisions if left to their own devices.

The voting situation in Ethereum is really bad, since it’s very difficult to identify who exact composes “the community”. That’s partially by design. We don’t believe in plutocratic coin votes making final decisions, although we do use it to help aid sentiment analysis. The most relevant vote here is the miner vote because they are the most affected by this proposal, and have the most to lose by it’s failure to be implemented. 77% have said yes, with 23% abstaining, which is probably about as good as it gets.

As you stated, the only metric that is down was the “influencer vote”, which is akin to representative democracy except we didn’t vote for these people and some of them don’t really have credentials to back their opinions. In fact, most if not all in the “no” camp reference a blog post written years ago by one person, which may or may not be accurate as it contains several unverified opinions and is not a peer reviewed paper. The influencer sampling also did not take into account the result of the audit, which I think is an important factor for many and would affect the results if it were done again afterwards.

I get it, sentiment analysis and understanding the “voice” of the community is hard. It’s made much harder by the psuedo-anonymous participation that defines public blockchains, especially when foregoing plutocratic on-chain voting structures. This is still a contentious issue, but I see more and more people moving in direction of support, even this late in the process. It seems to have a lot to do with education on the subject. Hardware design is an extremely complicated discipline, and difficult to describe simply. Most people understand the economic arguments better, and hearing about supply chain issues and access rights to purchase the underlying hardware is helpful too. It’s better to have permissionless entry to the market of mining IMO, and I think many agree as well.

1 Like
79

I may dissent, and a few more links for the studious among us:

  • Voting and Miners

It’s unfair to equate the ProgPoW team with miners. The ProgPoW team doesn’t represent the interests of honest miners at all, and voting was heavily rigged and manipulated in favor of ProgPoW.

One random example, a voice of the suppressed:

"Let’s take Ethermine for e.g. What’s they did?
They just took all availabale hashrate, without any notifications
for miners, and started voting YES. They promised provide dedicated
port for say NO, but didn’t provide it.
In this situation, miners were just hostages of pools.

If you want fair voting — at first prepare two different
dedicated ports for YES and NO, and then start voting.
Using default pool config for “YES” — it’s just fraud!"

James Prestwich got it
https://twitter.com/_prestwich/status/1109207136414781440

In PoW, lowest cost wins.
“Permissionless” access to hardware doesn’t help if you can buy GPUs at retail price in a lot of shops around the world, for a guaranteed loss.
Or you can be the one special partner of the chipmaker who helped the chipmaker exclude other chipmakers, and get the same GPUs at half the price or less in return.

  • Audit

The last audit proposal I’m aware of is still this one
https://github.com/ethereum-cat-herders/progpow-audit/blob/master/Least%20Authority%20-%20ProgPow%20Algorithm%20Audit%20Proposal%20(v2).pdf
Curious if that has changed.
What we’ve learned from the RandomX audits so far is that the first thing you want to see from an auditor is a one-page or more explanation of the difference between a PoW algorithm and a cryptographic hash algorithm, as seen by the audit team.
If the audit team doesn’t see any difference, or is just guessing a bit and largely thinks they are reviewing a cryptographic hash algorithm, that will lead to a disappointing result.

This is not enough.

EIP 1057 makes some key claims that are demonstrably false, and yet were used to rally support behind the EIP, for example

“These would result in minimal, roughly 1.1-1.2x, efficiency gains.
This is much less than the 2x for Ethash”

That’s only the tip of the iceberg of this fraudulent EIP, but one would hope that an audit can keep more damage away from Ethereum.

  • Reputation

The fact that the EIP 1057 author is a close business associate of Calvin Ayre and the nChain camp is public knowledge today, and seems to be accepted for now.
How do these guys make money?
Step 1: Sell ETH short on bitmex and other exchanges
Step 2: Put out press release saying Craig Wright co-authored and patented ProgPoW
Step 3: profit

  • Hardware Accessibility

I was seriously surprised (and learning!) that someone could turn around the positive idea of not wanting to sell PoW hardware to ponzi schemes, scammers or money launderers into a “blacklisting” argument, and use that to instill more ASIC fear! Amazing.

Here’s what happens if a PoW hardware maker acts irresponsibly:

A rogue megafarm is selling hashrate to unsuspecting retail customers at excessive setup and maintenance rates, and thus ‘inherits’ the capex for free after a few months when the difficulty goes up and the inability of their customers (victims) to calculate becomes apparent.
They then proceed to destroy or otherwise manipulate/dominate that coin since their focus is on short-term profits.
A hashrate owner that obtained hashrate for free, through whatever means, is a threat to the coin.

Responsible chipmakers don’t sell to these kinds of farms, because they act against the long-term interests of the coin that is being secured.
Nvidia focuses on the short-term, since everybody knows that GPUs stand no chance in a PoW algo in the long run.

80

Ethereum per se is a network of computers, maintained by a self-selected team of core developers. Our remit is the health of the network. The community is not a polity, but a loose and self-selected anarchy of people with some relationship to that network. What are called “votes” are not about democracy; they are about ensuring that network upgrades are not so contentious as to cause an unwanted fork.

The responsibility of the core developers to the community is ill-defined at best. They don’t elect us and they don’t pay us, so by ordinary standards they have no standing to tell us what to do. So I continue to believe we core devs can best serve the community by seeing to the health of the network as best we can. We are best served by the community when arguments are presented to in those terms.

4 Likes