Topic: Phi2 miner for AMD (Read 4823 times)

Thanks to @crytotamer01 for introducing this opportunity.

thanks for the update

Beta5d release, much faster than beta5c, especially for 580.
https://github.com/fancyIX/sgminer-phi2-branch/releases/tag/5.6.1.3.b5d

Thanks for the info.
Their algorithm uses 4x4 lyra2, which has already been fully optimized by many miners and is much easier to optimize than 8x8 lyra2.
My miner is good at optimizing 8x8 lyra2, which is much harder, but may not be the best for 4x4 lyra2.

Nice work, getting closer!

Beta5b release:
https://github.com/fancyIX/sgminer-phi2-branch/releases/tag/5.6.1.3.b5b
Speed compare of release (beta5b) than previous release (beta5): 580 +25%, Vega56 +35% for phi2, and 580 -15%, Vega56 +35% for lyra2Z .

Beta5 release, about 30% faster:
https://github.com/fancyIX/sgminer-phi2-branch/releases/tag/5.6.1.3.b5

Lyra2z somewhat better than the old sgminer by djm34, but still has a way to go compared to mkxminer (https://bitcointalksearch.org/topic/mkxminer-fast-lyra2zlyra2rev2-miner-for-amd-gpu-zcoin-gincoin-vert-mona-2360168)
Hope you can do much better

Lyra2Z hashrate is 1620khs on my vega 56, on Ubuntu 18.04 with AMDGPU Pro PAL driver (https://support.amd.com/en-us/kb-articles/Pages/Radeon-Software-for-Linux-Release-Notes.aspx).
The hashrate on windows should be the same. I tried using newest driver.

Bro, please share the driver version and hash rate for your vega on Lyra2z.

all my cards have biosmod for eth, and mclock 2150 and cclock 1130

I mean the rate on your card is less than my card. I don't know why. My card has eth bios mod, but I am not sure if that's the reason. Does your card use default frequency? Did you bios mod?

hello @fancyIX you recommend me to change the biosmod eth of my cards to their original bios, or there is some specific configuration of watches that you recommend for your miner

My vega 56 has over 3300khs. My 580 has over 1800khs.

They are lower than I expected. My Sapphire nitro+ 580 has over 1800khs. It's bios mod for eth though.

Testing new version
GPU0= MSI Armor rx480 4GB
GPU1= MSI Armor rx570 8GB
GPU2= MSI Armor rx470 4GB
GPU3= MSI Gaming rx580 4gb
GPU4= Asus RogStrix evo rx560 4GB


sgminer v5.6.1.3 - Started: [2018-09-09 17:35:08] - [0 days 00:15:37]
--------------------------------------------------------------------------------
(5s):6.471M (avg):6.022Mh/s | A:374  R:6  HW:0  WU:23.019/m
ST: 2  SS: 0  NB: 15  LW: 1234  GF: 0  RF: 0
Connected to yiimp eu (stratum) diff 8.000 as user
Block: a5f3c712...  Diff:2.48K  Started: [17:49:25]  Best share: 7.65K
--------------------------------------------------------------------------------
GPU 0:                | 1.502M/1.501Mh/s | R: 10.9% HW:0 WU:5.398/m I:22
GPU 1:                | 1.388M/1.389Mh/s | R:  1.7% HW:0 WU:5.287/m I:22
GPU 2:                | 1.320M/1.320Mh/s | R:  2.9% HW:0 WU:4.461/m I:22
GPU 3:                | 1.498M/1.497Mh/s | R:  0.0% HW:0 WU:5.508/m I:22
GPU 4:                | 447.8K/335.0Kh/s | R:  0.0% HW:0 WU:1.267/m I:22
--------------------------------------------------------------------------------
[17:47:51] Accepted 067bf42c Diff 39/6.000 GPU 3
[17:47:53] Accepted 099f32d5 Diff 26/6.000 GPU 4
[17:48:02] Accepted 0208a4fd Diff 125/6.000 GPU 3
[17:49:12] Accepted 0f5853d7 Diff 16/6.000 GPU 1
[17:49:12] Accepted 190cdafe Diff 10/6.000 GPU 3
[17:49:16] Accepted 1494b5e1 Diff 12/6.000 GPU 0
[17:49:17] Accepted 04ad1704 Diff 54/6.000 GPU 4
[17:49:18] Accepted 123c2077 Diff 14/6.000 GPU 1
[17:49:20] Accepted 29a78d6d Diff 6.146/6.000 GPU 2
[17:49:25] yiimp eu difficulty changed to 8
[17:49:42] Accepted 0e768533 Diff 17/8.000 GPU 1

And what AMD RX hashrates info you have  ) ? Can you share .

I don’t have 570 card. But I guess it’s around 1700 -1900 khs

What hashrate for phi2 expected for RX570 4GB ?

Beta4 release: https://github.com/fancyIX/sgminer-phi2-branch/releases/tag/5.6.1.3.b4. About 40% faster on my rx 580 and vega 56 cards. The speed is improved using pre-built binary kernels. They are only used if you use "-w 256". Please don't delete them otherwise new kernel will be compiled from source code and will be slower than those pre-built kernel. Please help test if interested.

I just updated my tutorial benchmarking RX 470 4Gb, RX 570 4GB, RX 580 4GB and RX 580 8GB here the results https://1stminingrig.com/how-to-mine-luxcoin-lux-with-nvidia-and-amd-gpus/

Nice work all, further than we were a few weeks ago.

Setting up a test machine tomorrow a d can start testing some releases with a few different cards.

beta3 is memory bond, and beta3b/c is core bond.
That means if your core is slow, you'd better not use beta3b/c.

Beta 3c, about 10% better:
https://github.com/fancyIX/sgminer-phi2-branch/releases/tag/5.6.1.3.b3bc

For me beta3 is faster than beta3b and c.
610khs against 540
used worksize 64

The newest release is about 2.1Mhs on my Vega 56 card. See
https://github.com/fancyIX/sgminer-phi2-branch/releases/tag/5.6.1.3.b3b

Edit:
beta3c can get 2.3Mhs.

Beta3b release, about 25% faster than previous release on my Vega 56 card. Please help testing if interested:
https://github.com/fancyIX/sgminer-phi2-branch/releases/tag/5.6.1.3.b3b

What is the vega 56 speed with the new beta?

1.2 mgh - Rx588. But it is still very weak...

~610khs on R9 280x. Intensity 18, worksize 64
win7, adrenalin 18.5.1

Beta3 release, about 40% faster than previous release on my RX 580 cards. Please help testing if interested: https://github.com/fancyIX/sgminer-phi2-branch/releases/tag/5.6.1.3.b3

Hi, fancyIX, any further optimization possible on lyra2 kernel? Is the feature branch #12 faster than master?

Same here. 2xVega 64 not recognized.

[01:42:38] CL Platform vendor: NVIDIA Corporation
[01:42:38] CL Platform name: NVIDIA CUDA
[01:42:38] CL Platform version: OpenCL 1.2 CUDA 9.2.147
[01:42:38] Platform devices: 1
[01:42:38]      0       GeForce GTX 1050
[01:42:38] 1 GPU devices max detected
It's not recognizing the 2nd platform. I have 3 AMD devices installed (Vega 56, RX 580, RX 560)

According to sgminer FAQ: " Run sgminer with the -n option and it will list all the platforms currently installed. Then you can tell sgminer which platform to use with --gpu-platform."

You can go to https://github.com/fancyIX/sgminer-phi2-branch/issues and file a bug.

Yes, the beta is working strange too....

[20:44:18] Probing for an alive pool
[20:44:28] Waiting for work to be available from pools.
[20:49:17] Stratum connection to eu.bsod.pw interrupted

I've already updated drivers, checked the firewall settings, without success.

Even with the beta2 release, if I try and specify "--gpu-platform 1" it exits with no error message.

Maybe beta2 works for you:
https://github.com/fancyIX/sgminer-phi2-branch/releases/tag/5.6.1.3.b2

Beta no longer works. Please use beta2 instead:
https://github.com/fancyIX/sgminer-phi2-branch/releases/tag/5.6.1.3.b2

Running a vega 56 rig...run the miner and it hangs then shuts down on

[05:53:28] Testing us.bsod.pw
[05:53:28] Probing for GBT support

On my vega56, set gpu threads to 2 will gain about 15% speed up

good work phi2 miner on amd?

I am working on some optimizations. Theoretically it can be 2x faster or more

RX480 (with Uber 3.1 timings)
-w 32 -I 20

average 872.8 kh
84w as reported by Afterburner

current profit minus electricity = -0.05$ 

didnt measure it, but online socket plug show around 650W for 4x vega so like 80W less than on monero (got 730w on monero)

Thank you, so there is no profitability at the moment

can i know the consumption please ?

W10 rx vega 56 1,3Mh/s each
rx 580 / 570 around 700kh/s each

worksize 32, intensity 20

~250khs on R9 280x. Intensity 18, worksize 12
win7, adrenalin 18.5.1

Wow great ! you should make an own thread to get more tester. Will test it Asap.

Beta release with win64 and linux x86_64 support:
https://github.com/fancyIX/sgminer-phi2-branch/releases/tag/5.6.1.3.b

May be lux dev can ask zawawa

Yeah, total 4.5mhs.

tdxminer is very good at lyra2z, so if they can work on this the result should be very impressive with amd cards, even better than nvidia.
Anyone has knowledge of GCN ASM may be interested on optimizing this miner.

Im guessing total, if not two 580 is performing like a 1080 ti With a miner made for fun that would be very impressive on a Nvidia friendly algo!

4.5mh each card or total??

Early preview release: https://github.com/fancyIX/sgminer-phi2-branch/releases

I made an AMD miner for PHI2 based on sgminer: https://github.com/fancyIX/sgminer-phi2-branch

I did this just for fun. It may have bugs so don't expect too much on it.

My rig has 6 RX 580 at 1170 Mhz and the hashrate is about 4.5 Mhs.

The most recent I saw in Disc:

"The people (plural) who were offered a bounty to produce a phi2 sgminer failed to do so. As I said before, it's beyond the Lux team's purview. Not much we can do about it except hope another sgminer dev steps forward."

With that...back to hoping some enterprising mind steps up...

Correct afaik. Hope here was there's be some other enterprising Dev working on something. I.E. a Claymore or other.

I also believe they are hoping the AMD miner will come from a 3rd party like sgminer. I don't think any of the Lux team are personally working on it - I may be wrong, but that is what I was getting from their discord channel.

nothing new to report, lux devs have been busy to get the chain back on track...

Bumping and hoping for any news, any real input appreciated. Thx all.

it's more than simply an algorithm change, the blockheader has also jumped from 80 bytes to 144 with the addition of two fields.. thats a big cubehash512

The other half of this--Phi2 isn't Phi1612. If someone gets one working, tunes, analyzes, and has the data to say "ain't gonna happen"...that's different, that's acceptable. At that point there's no choice but to move on.

So far all I have to go on is whattomine estimates. Does two rigs of 1070s produce more than same number of 580s? According to 1612 #s yes, comes out to about 165 vs 200.

So if you're in it for efficiency then sure, AMD not a viable option. But realize some are playing the gamble in hopes it increases in the future. If someone can still cover costs, even make a miniscule profit, but achieve the goal of banking some coin cheaper than buying...then it's worth it to them based on their strategy. Looking at the future potential of what is mined, not the current going rate.

That said--i can't buy the "there's no need" line. There's hash to give and Dev fee to be had if someone produces something, that's the bottom line.

phi algo is so tight on nvidias, there is no needs to do an phi amd miner. Like neo miner. It exist, but not many people use this

sorry, wrong discussion

I think you should ask the AMD Lyra2z miner  tdxminer  devs, because lyra2 2nd kernel takes 75%+ GPU time on PHI2 algo so again it should be faster on AMD vs Nvidia.

Also...this is mainly regarding  https://luxcore.io

Any input appreciated!

Anyone working on, or have plans to work on, a Phi2 miner for AMD gpus? There is a version of ccminer and cryptodredge for NVidia and there are supposedly attempts to get SGMiner working but no timeline. Anyone else aware of anything in the works?