[XMR] Monero Mining

[Oscilson]

What is the theoretical up limit of GPU performance in XMR?

For AMD R9 290X, the hash rate is 0.81kH/s using Claymore miner.

[owlcatz]

I'll pledge 2000 XMR to the release of this highly optimized miner code (if it in fact exists). And if it doesn't, well there should be a bounty for it anyways so mine remains open for anyone who wants to work on it.

Totally. Never understood why no one but Claymore has one coded yet. I'd add to the bounty even, just to make more xmr.

[DirtyUniverse]

I'll pledge 2000 XMR to the release of this highly optimized miner code (if it in fact exists). And if it doesn't, well there should be a bounty for it anyways so mine remains open for anyone who wants to work on it.

You can use Claymore's miner but is not open source.

[Lloydimiller4]

I will of course provide a bounty to having an open source AMD miner, I believe we need one. Is the CUDA miner opensource? I know there is an optimized CUDA miner for people with big rigs(at least I heard there was), but we need this stuff open to the public and get it more decentralized.

I would provide 500-1k XMR additionally for a bounty, easily, for helping GPU miners get more involved with Monero.

[GingerAle]

okay. As I mentioned, the theory is that some ridiculously optimized miner may exist.

chocolatebar mentioned goalposts for the bounty i proposed. I've never done bounties before, so I have no idea. So I'll just throw something together and we can start hashing it out.

I'm particularly interested in nvidia (so the cuda miner), mainly because the maxwell architecture is very efficient hashes / watt, at least with the last version of tsivs ccminer.

As mentioned in my original post, I proposed the full bounty for a 40.4X factor in hashrate. So, with my 750 ti card, that would be 280 h/s * 40 = 11200 h/s. Yeah. 11.2 kh/s. Man my rig would fly.

Granted, obviously, that might be ridiculous or impossible.

Smooth mentioned some stuff here https://bitcointalk.org/index.php?topic=653467.msg11571118#msg11571118 regarding GPU stuff.

For the 290X, as mentioned below, https://bitcointalk.org/index.php?topic=653467.msg11730818#msg11730818 , this card kicks out 810 h/s. So a 40.4X of that would be.... 32724. 32.7 khs. Yum.

SO:

If this code exists, this deserves the full reward. Thus, the primary bounty is for these numbers. Because these numbers SHOULD be impossible, this bounty should be ridiculous. So far we have:

GingerAle: 200
Jwinterm: "about 3.5 XMR" which I will interpret as anywhere between 3.5 XMR and 3500 XMR.
Chocolatebar: 2000
owlcatz: some yet to be determined amount
Lloydmiller4: 500 - 1000

Thus, at the extreme high, we have 200 + 3.5 + 2000 + 1000 = 3203.5 XMR


Lets use the number from my analysis 137.7 kh/s, and assume that thats the hashrate of our miner-with-awesome-code (MWAC). Yes, this assumption is flawed, because MWAC could have 80% of the hashrate and the other 11 people have the other 20%, but whatever.

Option #1:

Full bounty for code that can make my 750 ti spit out 11.2 kh/s, as calculated by my 750 ti existing hashrate X 40.4. This number actually comes close to a scenario where MWAC has 2 full rigs, with 6 GPUs per rig. So each card is spitting out 11.47 kh/s.

Option #2: (because option #1 is ridiculous)

Yah know what, we can just use 40.4 as ceiling, and any improvement in hashrate below that will get a bounty = that factors percentage of 40.4 X the full bounty...

a 2-fold increase in hashrate would be 2/40.4 * full bounty. So, at the current full bounty, thats 158 XMR.
a 3-fold increase in hashrate would be 3/40.4 * full bounty. So, thats 237.88
4 = 4/40.4 * 3203.5 = 317.18

... you get the idea.


In the case where a gynormous bounty is created, and someone releases 2X mining code, we could continue the bounty. We would just subtract that fold from the original number.... ah shit this is some complicated mafs. All right, ...

ori_hash = original has (280)
target_hash = MWAC theoretical hash (11200)
fold = original max fold (40)
bounty = original total bounty (3203.5)
relfold = released fold (2)

bounty_release = relfold / fold*bounty

new_bounty = bounty - bounty_release
new_hash = ori_hash * relfold
new_fold = target_hash / new_hash

nah, but then the new fold would be 20. And if someone made a 2 fold on that, they'd get 2 / 20 * new_bounty, which would be significantly more than the first release... but I guess that might make sense.

If someone wants to map out the function of what I was getting at above...

ah nevermind, We can cross that bridge if we get to it.

I think option #2 might cover it, because it encourages 2 things - a ridiculously high primary bounty which effects the lower bounties.

STIPULATION / PROPOSAL I propose that we donate 10% of the bounty to core development, because Monero.

[Oscilson]

Maybe those high hash miners use a proxy. They send the hash combined with many miners.

[jwinterm]

okay. As I mentioned, the theory is that some ridiculously optimized miner may exist.

chocolatebar mentioned goalposts for the bounty i proposed. I've never done bounties before, so I have no idea. So I'll just throw something together and we can start hashing it out.

I'm particularly interested in nvidia (so the cuda miner), mainly because the maxwell architecture is very efficient hashes / watt, at least with the last version of tsivs ccminer.

As mentioned in my original post, I proposed the full bounty for a 40.4X factor in hashrate. So, with my 750 ti card, that would be 280 h/s * 40 = 11200 h/s. Yeah. 11.2 kh/s. Man my rig would fly.

...blahblahblah

The 40.4x factor does seem a bit ridiculous and arbitrary. To be honest, I think even a 2x improvement at this point is asking for the near impossible. Maybe the way to structure it would be to have a 2x or 4x improvement take home all the cash, and so then a 10% increase in hash rate would be worth 5 or 10% of the current bounty. Just a suggestion, because I (from a pretty uninformed perspective) really think 40x is asking for well beyond the impossible.

[GingerAle]

well, an update: the operator of monero.net contacted me and provided information suggesting that there is not some super amazing optimized code.

I agree with jwinterm - a 2X would be significant enough to warrant full bounty, and subsequent percentages of 2X that get corresponding percentage of bounty.

However, we should be aware that a 10% improvement already exists - somewhere in the ccminer thread, but the dev hasn't released it publicly. Ah i can't find the link.

[jwinterm]

I do remember seeing some chat about a 5-10% increase in hashrate with a private version of ccminer posted either here or on irc. Probably worth it to give the person that releases that speedup 10% of the bounty, whether that's the author or just someone who paid for it and now just wants to release the binary for extra $$$XMR. I'll solidify my contribution to the bounty at 35 XMR, so a 10% speedup would be worth about 3.50 XMR

[photon_coin]

this is a hard coin to mine , that is all I know

if someone has a better miner it would not be the first coin where it was not shared with others

[GingerAle]

well, an update: the operator of monero.net contacted me and provided information suggesting that there is not some super amazing optimized code.

I agree with jwinterm - a 2X would be significant enough to warrant full bounty, and subsequent percentages of 2X that get corresponding percentage of bounty.

However, we should be aware that a 10% improvement already exists - somewhere in the ccminer thread, but the dev hasn't released it publicly. Ah i can't find the link.

Any explanation of the discrepancies in your analysis?

Basically, someone is mining with access to thousands of PCs. This could either be botnet or very lucky system administrators with system users that don't need 100% of their CPU time.

In either case, someone is getting boatloads of Monero for free. My assumption is that these coins are dumped to extract the most profit. But who knows. They could be hodlers.

My only hope is that these operators are getting this message - if you want to continue reaping profit from one of the only CPU mineable coins that is profitable, you should really donate a large percentage of these coins to Monero core development. After all, the only reason your whole enterprise is possible is because Monero exists.

To me, this is just further evidence that we need better GPU software. Yargh. I don't know what the best solution is. I support the whole everyone should be able to mine thing. But yargh.

[5w00p]

I believe the ccminer developer who goes by the username sp_ has an optimized private cryptonote GPU miner that he sells for 0.1 or 0.2 BTC.
Something like that.

I don't know for a fact how much better it is than tsiv's GPU miner, but it is reportedly faster.

Unless I am wrong and it is actually a CPU miner...

[OrientA]

well, an update: the operator of monero.net contacted me and provided information suggesting that there is not some super amazing optimized code.

I agree with jwinterm - a 2X would be significant enough to warrant full bounty, and subsequent percentages of 2X that get corresponding percentage of bounty.

However, we should be aware that a 10% improvement already exists - somewhere in the ccminer thread, but the dev hasn't released it publicly. Ah i can't find the link.

Any explanation of the discrepancies in your analysis?

Basically, someone is mining with access to thousands of PCs. This could either be botnet or very lucky system administrators with system users that don't need 100% of their CPU time.

In either case, someone is getting boatloads of Monero for free. My assumption is that these coins are dumped to extract the most profit. But who knows. They could be hodlers.

My only hope is that these operators are getting this message - if you want to continue reaping profit from one of the only CPU mineable coins that is profitable, you should really donate a large percentage of these coins to Monero core development. After all, the only reason your whole enterprise is possible is because Monero exists.

To me, this is just further evidence that we need better GPU software. Yargh. I don't know what the best solution is. I support the whole everyone should be able to mine thing. But yargh.

The PoW algorithm should be modified so that botnet will be discouraged.

[GingerAle]

well, an update: the operator of monero.net contacted me and provided information suggesting that there is not some super amazing optimized code.

I agree with jwinterm - a 2X would be significant enough to warrant full bounty, and subsequent percentages of 2X that get corresponding percentage of bounty.

However, we should be aware that a 10% improvement already exists - somewhere in the ccminer thread, but the dev hasn't released it publicly. Ah i can't find the link.

Any explanation of the discrepancies in your analysis?

Basically, someone is mining with access to thousands of PCs. This could either be botnet or very lucky system administrators with system users that don't need 100% of their CPU time.

In either case, someone is getting boatloads of Monero for free. My assumption is that these coins are dumped to extract the most profit. But who knows. They could be hodlers.

My only hope is that these operators are getting this message - if you want to continue reaping profit from one of the only CPU mineable coins that is profitable, you should really donate a large percentage of these coins to Monero core development. After all, the only reason your whole enterprise is possible is because Monero exists.

To me, this is just further evidence that we need better GPU software. Yargh. I don't know what the best solution is. I support the whole everyone should be able to mine thing. But yargh.

The PoW algorithm should be modified so that botnet will be discouraged.

But if thats done, then the average user will not be able to mine. That was the point of cryptonight, to have CPU mineable POW to try and achieve "1 PC 1 vote".

[OrientA]

How much of the hashing of XMR is contributed by botnet or larger server farms?

If we use the larger scratch pad size, 3MB instead of 2 MB, then high end CPU will not have bigger advantage. Intel 4170 has 3MB L3 cache, so it is effective for 1 thread; Intel 4770k has 8 MB, so it is effective for 2 threads, the ratio is 1:2. For 2MB scratch, the ratio is 1:3.

[superresistant]

this is a hard coin to mine , that is all I know
if someone has a better miner it would not be the first coin where it was not shared with others

Agree. It probably happened to all existing cryptos since Bitcoin.

Basically, someone is mining with access to thousands of PCs. This could either be botnet or very lucky system administrators with system users that don't need 100% of their CPU time.
In either case, someone is getting boatloads of Monero for free. My assumption is that these coins are dumped to extract the most profit. But who knows. They could be hodlers.

Either botnets or admin computers are far from being free. It requires very high skills and comes with high risks.
It is not free if you consider the time spent in learning all the skills to become a successful botnet owner (it probably takes years) and the time spent on spreading viruses and hacking computers. Not to mention the constant fixing and setting up the most effective way of using bots. Zombies computers are never under your total control, you could lose most of them anytime.
Using the computers of your company to mine is putting yourself at a high risk of getting fired and being sued.

I am not encouraging botnet mining or system administrators abuses. I just pointing out that it is much harder and complicated than what most people think.

[GingerAle]

this is a hard coin to mine , that is all I know
if someone has a better miner it would not be the first coin where it was not shared with others

Agree. It probably happened to all existing cryptos since Bitcoin.

Basically, someone is mining with access to thousands of PCs. This could either be botnet or very lucky system administrators with system users that don't need 100% of their CPU time.
In either case, someone is getting boatloads of Monero for free. My assumption is that these coins are dumped to extract the most profit. But who knows. They could be hodlers.

Either botnets or admin computers are far from being free. It requires very high skills and comes with high risks.
It is not free if you consider the time spent in learning all the skills to become a successful botnet owner (it probably takes years) and the time spent on spreading viruses and hacking computers. Not to mention the constant fixing and setting up the most effective way of using bots.
Using the computers of your company to mine is putting yourself at a high risk of getting fired and being sued.

I am not encouraging botnet mining or system administrators abuses. I just pointing out that it is much harder and complicated than what most people think.

I get that. I just mean in blanket base terms - cost of hardware, cost of electricity. A botnet owner can take that skill set (that cost time) and use it for another purpose. My electricity is gone, to never be used for another purpose. My GPUs could be used for another purpose, sure.

I'm not denying that its probably difficult to manage a botnet. I would just argue that the "resources" used to mine skew some of the assumptions made that underlie the purpose of POW. But, this is a phenomenon that I'm sure any POW encounters at some point.. bitcoin, litecoin.

Essentially, though, one could posit that these are the market makers. They have to effectively control the price of the coin to maintain a profit level but at the same time keep it suppressed in order to prevent the development of ASICs. Interesting dynamic.

[GingerAle]

How much of the hashing of XMR is contributed by botnet or larger server farms?

If we use the larger scratch pad size, 3MB instead of 2 MB, then high end CPU will not have bigger advantage. Intel 4170 has 3MB L3 cache, so it is effective for 1 thread; Intel 4770k has 8 MB, so it is effective for 2 threads, the ratio is 1:2. For 2MB scratch, the ratio is 1:3.

Well, using the data "thousands of PCs", we'll assume ... 2,000. Lets assume 1 mining thread, which on a modern cpu is about... 50 h/s? Thats probably low. Even my xeon 771's can get ~30 h/s per thread. Those numbers give 100kh/s.

[superresistant]

How much of the hashing of XMR is contributed by botnet or larger server farms?
If we use the larger scratch pad size, 3MB instead of 2 MB, then high end CPU will not have bigger advantage. Intel 4170 has 3MB L3 cache, so it is effective for 1 thread; Intel 4770k has 8 MB, so it is effective for 2 threads, the ratio is 1:2. For 2MB scratch, the ratio is 1:3.

Well, using the data "thousands of PCs", we'll assume ... 2,000. Lets assume 1 mining thread, which on a modern cpu is about... 50 h/s? Thats probably low. Even my xeon 771's can get ~30 h/s per thread. Those numbers give 100kh/s.

One zombie computer produce much less hashrate. Not to mention the high stale mate and connection problem.
Most bots are in China and Russia where the Internet connection is one of the worst in the world.
Chinese bots require a proxy to mine that could go offline anytime and add a huge lag time.

Honestly, bot-mining is not worth the investment of time and the hassle. There are better way of getting money if you've got the skills.

Even without the skills, it would be much easier to find free electricity or steal electricity than bother setting up ten thousand bots.

[GingerAle]

I'm just goin on what the pool operator told me, so...

But anyway.

Are we all in agreement? Maximum bounty is for 2X hashrate on existing hardware, and any increase beyond that is a proportional percentage of the maximum bounty? So if that guy publicly releases the one he has (which is a reported 10%), he'll get 10% of the bounty?