tdxminer lyra2z/XZC Miner for AMD GPUs on Linux

[todxx]

tdxminer has become teamredminer
Please see the teamredminer thread for newer releases.

tdxminer v0.2.2.2 Beta Release
This is an optimized miner for AMD GPUs running on the Linux.

This is a BETA release and may be unstable on some hardware.

Algorithms: lyra2z
GPUs Supported:

• RX 580/570/480/470 with amdgpu-pro or ROCm 1.8 drivers
• RX Vega 64/56, Vega FE with ROCm 1.8 driver

Download available here: https://github.com/todxx/tdxminer/releases/download/v0.2.2.2/tdxminer-v0.2.2.2.tgz
Brief readme available here: https://github.com/todxx/tdxminer

NOTE: This miner does NOT monitor GPU temperatures.
      It is up to the user to ensure that GPU(s) are functioning within safe power and temperature limits.

Software Requirements:

• Linux (preferably Ubuntu 16.04 as that is supported by ROCm and amdgpu-pro)
• Supported GPU driver (see GPUs Supported above)
• libjansson4

This miner includes a 3% dev fee.

Rough lyra2z user hash rates on stock GPUs:

GPU	SCLK (MHz)	MCLK (MHz)	Hash Rate (Mh/s)
RX 580	1167	2000	3.27
RX Vega 64	1401	945	6.66
RX Vega 64 LC	1408	945	7.07

If you are considering installing ROCm, please check that it supports your hardware first.
Note that ROCm versions prior to 1.8 do not support typical riser setups.

For reporting bugs and/or for features requests please open an issue in the github issue tracker,
or make a post here and I'll do my best to respond.

Any feedback would be appreciated.

------------------------------------
Changes in v0.2.2.2
- Changed output to be line buffered for easier use with piping.
- Fixed month off-by-one bug in printing the date.
- Fixed hang that could occur when pool connection is lost.

Changes in v0.2.2.1
- Added better/faster handling of non-responding pool connections.
- More network bug fixes.

Changes in v0.2.2.0
- More network bug fixes for pool connection.

Changes in v0.2.1.1
- Fixed bug with handling network errors in dev pool connection.

Changes in v0.2.1.0
- Fixed bug with handling network errors in pool connection.

Changes in v0.2.0.0
- Improved lyra2z performance
- Support for Ellesmere GPUs under amdgpu-pro drivers
- Dev fee set to 3%

[todxx]

To answer this upfront since it's been asked a few times:

There is no windows version available.

[soothaa]

Curious why this version is 40% slower than the previous thread indicated..?
My bad.. different dev/thread. I was mixed up.

[ArcticUK]

What speeds were indicated before? It's the first time I have seen this miner and it is damn quick, best think since sliced bread. Well done dev and alpha release is very stable for me. I'm actually getting better results than indicated here.

[soothaa]

What speeds were indicated before? It's the first time I have seen this miner and it is damn quick, best think since sliced bread. Well done dev and alpha release is very stable for me. I'm actually getting better results than indicated here.

9MHs+

[yrk1957]

What speeds were indicated before? It's the first time I have seen this miner and it is damn quick, best think since sliced bread. Well done dev and alpha release is very stable for me. I'm actually getting better results than indicated here.

9MHs+

That thread was a different dev/miner.

[soothaa]

What speeds were indicated before? It's the first time I have seen this miner and it is damn quick, best think since sliced bread. Well done dev and alpha release is very stable for me. I'm actually getting better results than indicated here.

9MHs+

That thread was a different dev/miner.

Oh my bad..

[Elder III]

This is quite interesting. I'll follow this thread for any information about a future Windows version.

[todxx]

What speeds were indicated before? It's the first time I have seen this miner and it is damn quick, best think since sliced bread. Well done dev and alpha release is very stable for me. I'm actually getting better results than indicated here.

Thanks Arctic!  I'm glad it's working well for you.
I have done my best to test this release on my hardware for stability and have run it for several hundred hours on my rigs.  But I can't really call it more than an alpha release until I get some feedback on how it runs on other hardware.  I've already had a couple reports of it not being able to find/load GPUs on certain hardware, so hopefully I can get to the bottom of those problems soon and get a new release out.

[N2DCRYPT]

Definitely interested, but worried about the PCIE Gen 3 requirement. My larger rigs definitely require Gen 2 settings in the bios due to the number of cards, so don't know how feasible it will be to run this miner except on a couple one or two card computers.

Btw todxx, have you checked out the other thread on potential release of Vega XZC miner? Guy seemed like he was doing it in GCN ASM instead of OpenCL or ROCm... maybe you could pick something up from that thread to optimize your miner/make it compatible with Windows. Assuming he wasn't bs'ing, but the numbers he was throwing around were phenomenal.

[todxx]

Definitely interested, but worried about the PCIE Gen 3 requirement. My larger rigs definitely require Gen 2 settings in the bios due to the number of cards, so don't know how feasible it will be to run this miner except on a couple one or two card computers.

Btw todxx, have you checked out the other thread on potential release of Vega XZC miner? Guy seemed like he was doing it in GCN ASM instead of OpenCL or ROCm... maybe you could pick something up from that thread to optimize your miner/make it compatible with Windows. Assuming he wasn't bs'ing, but the numbers he was throwing around were phenomenal.

Yeah, ROCm is a bit of a double edged sword right now.  Their toolchain is good, but their run-time has some overbearing requirements that other drivers don't.  I can see how they ended up there as ROCm was initially targeted for data-center cards, but now that they are starting to use it as the default driver for desktop cards, it's causing some headaches.

Regarding the other thread, I don't know what to tell you man.  I'm pretty sure you could get pretty close to those numbers with my miner if you properly tuned/OC'd your GPUs.  The numbers I posted are for bone-stock GPUs (stock bios, voltages, clocks, etc).  The only thing I did was set a fixed power-state so I can get reliable performance measurements.  Given all of this, and the fact that he posted his rant about half a day after I released my alpha release on github makes me feel like it's someone who just ripped my kernels and ran them on some very OC'd cards.  I even have an idea of who it might have been, as I had asked this person if he could test out my miner right after posting the release.  Maybe I'm just being paranoid.  But I can't get over the fact that I've been working on this for 4+ months, and the day I post a release, coincidentally someone else figured out how to do the exact same thing later that day.  I guess we'll find out if his kernels are the same if he ever releases anything.  Ugh, sorry for the rant.

Since the request for supporting windows/amdgpu-pro drivers is the most common feedback I get, I will focus on figuring this out.

[todxx]

Turns out converting my kernels from ROCm to amdgpu-pro/windows is a lot simpler than I expected.
Unfortunately, I first need to waste some time hardening them so we don't have another "coincidental" announcement.

I should have a new version that supports amdgpu-pro drivers on linux in a day or two.  This should make the miner compatible with the typical linux rigs running RX 580s/570s, such as HiveOS and the like.  Windows will take a week or two longer, since I don't even have a windows environment to build/test in right now.

[Iamtutut]

This is quite interesting. I'll follow this thread for any information about a future Windows version.

Same here, I only mine with windows.

[N2DCRYPT]

Turns out converting my kernels from ROCm to amdgpu-pro/windows is a lot simpler than I expected.
Unfortunately, I first need to waste some time hardening them so we don't have another "coincidental" announcement.

I should have a new version that supports amdgpu-pro drivers on linux in a day or two.  This should make the miner compatible with the typical linux rigs running RX 580s/570s, such as HiveOS and the like.  Windows will take a week or two longer, since I don't even have a windows environment to build/test in right now.

Nice! yeah I'll definitely be waiting for windows as well, since the undervolting/overclocking with Vegas is much easier in windows. Windows 10 is free, and you can do pretty much everything without activating it... not sure what's holding you back but definitely looking forward to this version!

Edit: Also, that's very interesting/coincidental about the other thread for sure... If true, I'd definitely use yours... and that would make his whole rant quite hypocritical. If it is yours, sweet! That means the numbers for this miner are great  but that would also mean your devfee is quite high 

Hopefully, if it really is as simple as dropping your kernel into something like SGMiner, we won't have to wait too long for a windows version. Very excited about this, been in XZC for a while but never had a good reason to mine it.

[ginterh]

Hi,

Is there anyway someone could upload a linux image with this installed? It's quite simple using this http://hddguru.com/software/HDD-Raw-Copy-Tool/

[todxx]

Turns out converting my kernels from ROCm to amdgpu-pro/windows is a lot simpler than I expected.
Unfortunately, I first need to waste some time hardening them so we don't have another "coincidental" announcement.

I should have a new version that supports amdgpu-pro drivers on linux in a day or two.  This should make the miner compatible with the typical linux rigs running RX 580s/570s, such as HiveOS and the like.  Windows will take a week or two longer, since I don't even have a windows environment to build/test in right now.

Nice! yeah I'll definitely be waiting for windows as well, since the undervolting/overclocking with Vegas is much easier in windows. Windows 10 is free, and you can do pretty much everything without activating it... not sure what's holding you back but definitely looking forward to this version!

Edit: Also, that's very interesting/coincidental about the other thread for sure... If true, I'd definitely use yours... and that would make his whole rant quite hypocritical. If it is yours, sweet! That means the numbers for this miner are great  but that would also mean your devfee is quite high 

Hopefully, if it really is as simple as dropping your kernel into something like SGMiner, we won't have to wait too long for a windows version. Very excited about this, been in XZC for a while but never had a good reason to mine it.

Yeah, tuning support for Vega on linux is not great.  I believe in kernel 4.17 they will add support for wattman like profiles for overclocking/tuning, but that's a ways out.  I've mostly been running my cards stock, but I also use them mostly for development/testing.

Regarding the other thread, thanks for seeing my side.  Obviously I can't say for sure if it was coincidence or malice, but I guess we'll have to wait and see if he releases anything.  Or even posts again for that matter.
The perf on the current release is already pretty good.  I think people might be getting the wrong impression by the modest hardware settings I used for the hash rates I posted.  I'll see if I can get some numbers from my Vega 64 LC at max powerstate.  I'm sure they'll be impressive and totally useless since that thing burns stupid amounts of power at full throttle.

As for the dev fee, I'm not planning on having a egregiously high dev fee.  I'm currently thinking 3% for the next release.  Even the current relatively high fee in the miner is nowhere near the 25% that guy was talking about.  Unlike what he was saying, I don't need to make quick cash on this algo before XZC switches over to MTP.  I've done work on other algos like lyra2rev2, nist5, skunk, neoscrypt and some others.  I chose to release lyra2z first since my implementation of it has the best perf improvement out of the algos I've worked on (also some like skunk and nist5 are basically dead for GPU mining now, and Claymore beat me to the punch for neoscrypt ).  The next algo I plan to support is lyra2rev2 which should be able to get around ~70Mh/s on a modestly clocked Vega 64.  But first I need to get that windows build out

About the kernels, they are not drop in compatible with sgminer.  Some reverse engineering work and modifications would have to be done.  Before the alpha release, I thought I was being paranoid thinking that someone would go through the effort required to rip them, so I did not spend effort trying to make them hard to rip.  I also think it's a bit of a waste having to spend the time/effort and adding useless junk to the kernels just so they are harder to steal.  Oh well, the next release will have kernels that will have better performance and be much more difficult to rip.

[MinedTangerine]

I also watch this thread with great interest. AMD needs some more versatility against NVIDIA. I think the cards are capable but lack of optimized miners is holding AMD back. 3% dev fee is fine if you will really make such an improvement. You get paid for quality work. I have no problem with that.

[N2DCRYPT]

Turns out converting my kernels from ROCm to amdgpu-pro/windows is a lot simpler than I expected.
Unfortunately, I first need to waste some time hardening them so we don't have another "coincidental" announcement.

I should have a new version that supports amdgpu-pro drivers on linux in a day or two.  This should make the miner compatible with the typical linux rigs running RX 580s/570s, such as HiveOS and the like.  Windows will take a week or two longer, since I don't even have a windows environment to build/test in right now.

Nice! yeah I'll definitely be waiting for windows as well, since the undervolting/overclocking with Vegas is much easier in windows. Windows 10 is free, and you can do pretty much everything without activating it... not sure what's holding you back but definitely looking forward to this version!

Edit: Also, that's very interesting/coincidental about the other thread for sure... If true, I'd definitely use yours... and that would make his whole rant quite hypocritical. If it is yours, sweet! That means the numbers for this miner are great   but that would also mean your devfee is quite high  

Hopefully, if it really is as simple as dropping your kernel into something like SGMiner, we won't have to wait too long for a windows version. Very excited about this, been in XZC for a while but never had a good reason to mine it.

Yeah, tuning support for Vega on linux is not great.  I believe in kernel 4.17 they will add support for wattman like profiles for overclocking/tuning, but that's a ways out.  I've mostly been running my cards stock, but I also use them mostly for development/testing.

Regarding the other thread, thanks for seeing my side.  Obviously I can't say for sure if it was coincidence or malice, but I guess we'll have to wait and see if he releases anything.  Or even posts again for that matter.
The perf on the current release is already pretty good.  I think people might be getting the wrong impression by the modest hardware settings I used for the hash rates I posted.  I'll see if I can get some numbers from my Vega 64 LC at max powerstate.  I'm sure they'll be impressive and totally useless since that thing burns stupid amounts of power at full throttle.

As for the dev fee, I'm not planning on having a egregiously high dev fee.  I'm currently thinking 3% for the next release.  Even the current relatively high fee in the miner is nowhere near the 25% that guy was talking about.  Unlike what he was saying, I don't need to make quick cash on this algo before XZC switches over to MTP.  I've done work on other algos like lyra2rev2, nist5, skunk, neoscrypt and some others.  I chose to release lyra2z first since my implementation of it has the best perf improvement out of the algos I've worked on (also some like skunk and nist5 are basically dead for GPU mining now, and Claymore beat me to the punch for neoscrypt ).  The next algo I plan to support is lyra2rev2 which should be able to get around ~70Mh/s on a modestly clocked Vega 64.  But first I need to get that windows build out

About the kernels, they are not drop in compatible with sgminer.  Some reverse engineering work and modifications would have to be done.  Before the alpha release, I thought I was being paranoid thinking that someone would go through the effort required to rip them, so I did not spend effort trying to make them hard to rip.  I also think it's a bit of a waste having to spend the time/effort and adding useless junk to the kernels just so they are harder to steal.  Oh well, the next release will have kernels that will have better performance and be much more difficult to rip.

Impressive! 70Mh/s on Lyra2Rev2 would beat out mkxminer... I've gotten it up to 67 on a V64 with high overclock, but 70 @ stock settings? Niceee.

As mentioned above, 3% is quite reasonable. People will pay for good miners, especially new ones where there is no real competition.

As for ripping kernels... don't know much about that, although I do recall Claymore calling out Phoenix's ETH kernels fairly recently. I think you're focused on the right thing, as far as developing goes. There is/will be (at least from my perspective) high demand for more algo flexibility on the AMD side. And the gpus are quite capable if you can fine-tune your kernels to exploit the hardware, particularly the Vegas.

I've got a fair amount of AMD cards, not sure what your hardware constraints are, and am I'm happy to test/run anything you throw my way. Got RX 480s, RX 580s, Vega 56s, Vega 64s, and Vega FEs.

[todxx]

New version released!  Now with support for amdgpu-pro Linux drivers.
I've only been able to do minimal testing of the amdgpu-pro support, so some testers here would be much appreciated.

Also if you have successfully tested out the release, please post your results!
I have limited hardware available for testing and am very curious to see what hash rates people get with different setups.

[todxx]

Impressive! 70Mh/s on Lyra2Rev2 would beat out mkxminer... I've gotten it up to 67 on a V64 with high overclock, but 70 @ stock settings? Niceee.

As mentioned above, 3% is quite reasonable. People will pay for good miners, especially new ones where there is no real competition.

As for ripping kernels... don't know much about that, although I do recall Claymore calling out Phoenix's ETH kernels fairly recently. I think you're focused on the right thing, as far as developing goes. There is/will be (at least from my perspective) high demand for more algo flexibility on the AMD side. And the gpus are quite capable if you can fine-tune your kernels to exploit the hardware, particularly the Vegas.

I've got a fair amount of AMD cards, not sure what your hardware constraints are, and am I'm happy to test/run anything you throw my way. Got RX 480s, RX 580s, Vega 56s, Vega 64s, and Vega FEs.

Thanks, I actually started working on lyra2rev2 first, but then quickly realized that lyra2z had much more low hanging fruit for performance improvements.

Here is my reasoning for the 3% fee.  While it is higher than the more common ~2% range for competitive miners, this miner is (to my knowledge) one of the most profitable for AMD hardware currently available.  If you look at whattomine's XZC estimates for revenue based on the hash rates posted in the description, at the time of this post it estimates:
RX 580 - $2.14/day
RX Vega 64 - $4.37/day
These are significantly more profitable than the most profitable coin currently on their front page (currently XMR, with Vega 64 bringing in ~$2.90/day).  And again, the hash rates I've posted are from stock GPUs with no tuning.  Others have reported their rates are higher.  If and when this miner becomes less relatively profitable, either due to competing miners or difficulty increase, I will lower the fee to stay competitive.  I'm not trying to price gouge here.  On the other hand I wouldn't mind getting some return on the several hundred hours of work I've put into this miner.
Edit: I forgot to mention, the miner code is structured to support different fee per algo.  For lyra2rev2 I would go around 2% since I don't believe it would be profitable enough to command the higher fee.  If I can't get it to go significantly faster than a 1080ti, I don't think that justifies much over a 2% fee.

I very much agree with you regarding AMD gpus just not having good software available, and especially for Vega.  Vega has some nice ISA improvements that can yield pretty noticeable perf improvements.  Many of these ISA improvements are not currently supported by OpenCL toolchains, so some hand tuning is required to use them.  For the algos I've looked at so far I can usually get a Vega 64 to perform similarly to, or better than a 1080ti.

Last, but not least, the GPUs you listed are almost exactly all the GPUs the miner currently supports   What OS are you currently running your GPUs under?  If any are under linux, I would be very curious to see how the miner runs with them.  Especially the RX 480 and Vega 56 gpus, as I don't currently have access to any of those.