Topic: [RFC] Potential release of AMD Lyra2z miner (Raw speed: Vega64 9.8 MH/s) (Read 7171 times)

And I bet there would be thousands willing to buy his product just to not to deal with Linux.

Long post, please read through if you have an interest in Lyra2z and is an AMD GPU owner, otherwise you can stop here.

Background: Lyra2z is commonly regarded as an "NVidia-only" algo since there are no public miners for available AMD GPUs worth the time. In reality, it's just a bullshit statement. You can pull off an highly optimized Lyra2z kernel that runs circles around the NVidia miners available, but you need to use techniques not available in OpenCL. GCN ASM is the only solution, and you need to be quite skilled working your magic.

My miner, which currently is a pure GCN ASM-replacement that I drop into sgminer, does 4.5-5.5 MH/s on a RX 580, 7-8.5 MH/s on a Vega 56 and 9-10.5 MH/s on a Vega 64, intervals depending on how you aggressively you set your clocks. So yes, a RX 580 now equals more than 1.5 1080ti and a not-too-overclocked Vega 64 can crunch more than 3x1080ti.

I only have tons of Vegas and RX580s of all sorts at my disposal to test with, but the implementation should work for any device >= GCN 1.4. The hashrate is pretty much a linear function of the nr of CUs that the card provides.

My question to the (zcoin and other tiny lyra2z altcoins) communities right now is: in what form do you want this miner released? Some choices are a classic dev fee miner, selling an unlimited amount of kernels for a set price (sp-mod style) or selling a limited amount of kernels to the highest bidders.

As you can figure, my own edge right now mining with my rigs is quite nice. The reason I'm writing this is an upcoming move and I will probably have to shut down my rigs. Personally, if I make this available at all, I'd prefer a public miner. I don't like selling to a few select people/farms. Also, the (supposedly) limited time left for ZCoin before the MTP fork renders an upfront investment obsolete in a not-too-distant future. The issue is that this will not be a 2% dev fee miner. Releasing it to the public is not worth it for me with anything less than a 25% dev fee, it's just simple math given the size of my mining rigs and my current edge. However, I know the shitstorm such a suggestion sparks in some entitled peoples' heads. Of course, the kernel will also be dumped and copied within a week regardless of how much I work on protection, and then my current edge is gone forever, which is also factored into the 25%.

So, before the kneejerk reactions take over completely, let me present a few arguments. Taking the obscurity of Lyra2z into account, I still don't believe there has been any other new miner instance where blood, sweat, tears and engineering ingenuity has improved the best known public miner for e.g. a stock RX 580 by 1800-1900% in one disruptive release and at the same time also completely crushed comparable NVidia GPUs across the board. That and the assumption that my edge will be short is what the 25% is about, not ripping people off. Even WITH 25% dev fee, mining zcoin with a Vega 64@180w theoretically yields 7 MH/s and $2.59/day net profit with this miner and has approx the same hashrate as two 1080tis. At the time of writing, this is > +140% profit/day compared to the current top entry in whattomine.com's standard list for a Vega64, and then you'd have to go to Claymore's Neoscrypt miner that consumes even more power and some complete shitcoin that makes ZCoin look like BTC. Presented with an obvious win/win, if you're still one of those guys that don't see the work and value involved and is pissed for feeling "ripped off" just because you have to adhere to the conditions that I, the rightful owner of the intellectual property, has stated, then trying every trick in the book to circumvent the dev fee, well, congrats, you're the reason these kernels stay hidden or sold exclusively to big farms. I could have presented this with a non-disclosed 50% dev fee, said that a Vega is now 4.5-5 MH/s, and you would all still have been impressed. THAT would have been a rip-off imho. Instead, I'm being fully transparent, which is a common thing to ask when new unknown miners are released.

So, the choice is yours. If I can be convinced enough people will actually honor a 25% dev fee, I might take the time to rewrite stratum code and other things to not violate GPL by stealing from sgminer, the miner codebase I know best. It's a serious investment in time. The easy route for me would be to just to sell the kernel: no rewrite of a serious codebase, no support, no dealing with imbeciles complaining about potential malware, but as stated above, this is not my preferred choice. Or I just continue keeping it private until the ZCoin MTP fork, no big loss for me really. This thread and/or PMs will decide.

A FWIW-style reference: I'm letting my dev workstation with a RX 580 and Vega 64 LC mine at Mintpond for a while now. You can verify the hashrate with the address below and see that it matches the attached pic (not a junior member, no inline pics...).

https://i.imgur.com/gDSNJLZ.png

https://mintpond.com/#zcoin:aPm4DXfXWLf59qvMMnQ7tHBzoYFNKCxFAD

EDIT: added Mintpond link

You should try to talk with The Doctor Tekcomm. He would show how it is spending 140h of coding just for pure pleasure sharing his work with others for free.
Greed is the root of all evil...

I tried it and it works with 1 RX Vega card. With two or more cards, the machine refuses to boot and throws PCIe errors. If someone is able to run more than one RX Vega, please share your findings.

I was planning to try this on Sunday. Literally have not much experience with linux but the plan was;
1. Install ubuntu
2. Install ROCm 1.8
3. Install tdxminer

And hopefully it will recognize the 8 vega 64s for me to point it to a Lyra2z pool.

Sgminer's lyra2z kernel is unoptimized.

There is also tdxminer but I have not tried yet, have you?

There is also tdxminer but I have not tried yet, have you?