Discussion: Bounty for new GPU solver with better efficacy than a G1 mini

Mining
1

The premise of PoW mining is to get as close to a the thermodynamic limit as possible and ASICs are the best way to achieve that. However, with the progression of GPUs and no further known develop of Grin ASICs, GPUs could now ( or soon) better reach that limit.

All Grin mining software is outdated, built on solvers that are not optimized for new Gen hardware. Are we already at the point where new gen GPUs provide better efficiency than current gen ASICs? This is not a question of if, but when. Based on the current climate it also seems unlikely that new Gen ASICs will be shipped anytime soon?

For reference, on specs a G1 mini produces ~0.01GPS/W. A G1 has better efficacy, but no one is going to run one of them in their house and no hobbyist is going to pay the full cost either.

Speaking with [@Lolliedieb ] he mentioned that “I am pretty confident that with a completely rewritten solver a RX6900 could reach 1.2 g/s and a 4090 certainly about 2 g/s when well done”

I’ve then asked about wattage and he’s said “180W (that 6900XT) / 280W (the 4090) maybe? For the first rounds we need quite some core and memory, therefore can not go too deep with the settings. On AMDs it massively helps though that we can control voltage down manually”

For reference, if a 6900XT could achieve 1.2GPS @180w, then Grin would be the most profitable coin to mine on that GPU- The same will likely be applicable to other new gen GPUs aswell.

No miner dev has bothered to release new public software, because, there’s currently no incentive to do so. Grin’s block subsidy is so small in $ terms that it would take years for them to recoup the time spent in miner fees

However, If for example someone could design a new solver that achieved better efficacy than a G1 mini, then would it be worthwhile considering whether a bounty should be created and open for the first developer who achieves this?

Negatives:

  1. If more GPUs switched over and started mining Grin, then price could fall, because, it might create additional selling pressure( if new miners dumped more Grin than current miners).
  2. If price remains stable and network diff goes up, then Grin might not remain the most profitable coin to mine, therefore few miners will consider switching over.

Positives:

  1. Creates a fair market. knowingly having outdated public mining software is dangerous and creates a potential unfair advantage, since selfish miners could be running their own private software. Sure, this a risk for any PoW, however, it’s different in this case since we can already make the strong assumption that publicly released mining software can be significantly optimized*
  2. Creates a stronger network
  3. Creates new interest in Grin- Grin’s monetary policy is designed for miners in the sense that Grin’s emission today is the same as it was when it launched. A new influx of miners could help the market better appreciate this design.

*Question: Does not addressing this potentially undermine Grin’s “Fair” emission premise?

3 Likes
2

I will post my opinion here, since I won’t be able to attend next week’s meeting.
In short, I am skeptical that we should spend funds on promoting GPU mining. I would expect Ipollo to come with a new miners with either smaller procede chip (more efficient) and or to double or quadruple memory to boost efficiency for next gen G2 (mini). This is not hard to do and does not require any redesigninging of parts. They might already have some more efficient miners on small scale without us knowing🤫.
In any case I am skeptical about putting a bounty for redesigning a GPU miner, I believe the future of Grin mining is in ASICs since they will always be more efficient.
Also I would not be surprised if there already is mining software for the latest generation of GPU’s that can efficiently mine Grin.

1 Like
3

This is not just a consideration for the CC, but for the OC too. I just added it to the CC agenda so it would get formally addressed.

fyi there’s already a bounty here for the M1 ultra : Bounty - 1 BTC + 100 Grin for a MacOS M1 C32 Open Source Miner with ≥0.5 Gps - #2 by tromp

The main point is that if mining software can be built that significantly improves GPU efficacy to the point where it now(or soon) outperforms current gen ASICs, then shouldn’t we be incentivizing for that software to be public?

3 Likes
4

Correct me if I’m wrong, last time, the main reason that we moved from GPU to ASIC because we prevented the hashrate attack?

5

Most developers make their software freely available but ask 1-2.5% of the mining revenue. It would not be bad if someone develops a better or more efficient solver for high end cards but I believe this model where the software developer takes the initiative and carry the risk is in the best interest of the project. For example, if a developer would not want to invest the time it could mean that:

  1. he/she thinks there will not be a return on the investment because it is technically not possible.
  2. he/she believes it might be too risky since new ASICsight hit market.
  3. he/she might think there are not enough high end GPU miners around to make it worth the time investment.

Whatever the reason is, basically it would mean that the CC or OC would make a risky investment. So it would be more of a for “for fun” bounty.

In any case, it is good to have this discussion, but I do not believe it would attract many GPU miners since it would only work for the most high end expensive GPU’s that can profitably mine many other coins.

6

I support a miner bounty. Bounty should be for open source only with appropriate free use licensing.

I think a smaller bounty for any functional miner, cpu or gpu, would be good. Just a basic framework to build improvements on would be a good start.

Improvement bounties could be issued after that, but I would rather it be a smaller faster goal at first than a drawn out development.

3 Likes
7

Yeah, will be cool to update docs about mining as well, this info is outdated:

I could not find open-source C32 miner.

image
image1092×785 69 KB

8

https://github.com/tromp/cuckoo/tree/master/src/cuckatoo offers lean miners and a CUDA miner (that requires 64K shared memory I think).

1 Like
9

Does that repo have more recent work than grin-miner?

And does it have the functional components to send shares to a grin node or is it just the solvers?

10

It’s just the solvers. The lean miners are completely uncompetitive and the CUDA one is somewhat uncompetitive and has very high shared memory requirements that were not widely available at the time of release (I seem to remember only an RTX Titan would be able to run it).

2 Likes
11

I was fighting alone against the whole forum last week.
Look at the coincidence, we see another example of how the inflation model hurts.

In my opinion, if Ipollo isn’t your company, you don’t have to worry about it. In the competitive market, if the GPU beats the G1, that’s a problem Ipollo has to solve.

However the thing is, if GPUs get involved, they could try to destroy Grin with a bunch of 51% attacks because block rewards are so cheap, they might even try as a joke.

Those who buy Ipollo devices are usually people who trust Grin. However, there may be people on the GPU side who will come from other coins mining and play games with the network.

1 Like
12

As mentioned most miner devs wouldn’t consider designing a new solver for C32 because the network is currently dominated by ASICs. So on paper it would be optimistic to assume their new software would be used by ~5% of the networks miners. A 2% fee on that 5% is ~$3 per day. So it could take years to recoup the time spent.

As allured to with the discussions from @Lolliedieb with new solvers then on paper Grin would be the most profitable coin to mine on multiple new gen GPUs. There’s barely any profit at all in GPU mining atm. So if miner devs are still updating their software then it almost makes sense now to design a new solver for Grin, even without a bounty incentive.

I don’t think it needs to be a large bounty. Also, obtaining the efficacy of a G1 mini is a challenging task and might not even be possible, yet. However, it’s in the best interests of the network to know if/ when it’s possible, otherwise it creates an unfair advantage.

2 Likes
13

I support a bounty for this development.

We have only one company, Ipollo, basically controlling the hashrate. The most effective asic G1 is out of stock and the community has access only to less effective G1 mini.

Introducing an effective GPU solver will give the community a chance to be less dependent on Ipollo and also put some pressure on them to release new miners to be competitive.

2 Likes
14

G1 and G1 use the exact same ASIC, and so are equally effective. The G1 just uses a much larger number of ASICs.

2 Likes
15

我们是否考虑更换新的算法以支持更好的cpu采矿呢？

16

G1 is more effective than G1 mini.

G1: 2800 / 36 = 77
G1mini: 120 / 1.2 = 100

1 Like
17

The difference is in the power supply, not the asic chips. To the average user that means G1 is more efficient.

An advanced user may be able to build a better power supply and/or common power rail across several G1 minis to reduce the impact, but at the end of the day your time/money would probably be better spent buying the G1

4 Likes