Just tested out v4 on Ubuntu/fglrx and a couple R9 380 cards. No material change. Reducing MAX_SOLS from 2000 to 500 reduces the amount of data xfer over the PCI-E bus, and gives me about a 3-5% speed improvement running sa-solver. In my fork I have OPTIM_SIMPLIFY_ROUND turned on, and the MAX_SOLS reduced. https://github.com/nerdralph/sa-nrWatch out you are missing solutions if you reduce MAX_SOLS. It might be ok if it correspondingly reduces the Equihash runtime, but you have to look at how many solutions you lose on a good sample >1k, such as "./sa-solver --nonces 1000" |
|
|
|
As I see FGLRX constant has been removed from params.h. So, for 15.12 fglrx leave params.h as is?
The parameter was renamed. It is now named OPTIM_SIMPLIFY_ROUND Everyone should try to set it to 1 to see if it helps, no matter what you drivers or hardware is. It is worth a try. See https://github.com/mbevand/silentarmy/blob/master/TROUBLESHOOTING.md for instructions |
|
|
|
|
And me, Ubuntu 16.04, and amdgpu-pro 16.40. So latest drivers might have slightly degraded performance...
|
|
|
|
|
Zero difference with or without OPTIM_SIMPLIFY_ROUND on my test system (RX 480 and R9 Nano.)
I already tried putting collisions[] in local memory and also saw zero differences as well...
Weird. What is your OS & drivers?
Pascal is a different story.
|
|
|
|
Thanks eXtremal! I should say I put ZERO EFFORTS into optimizing for Nvidia. Silentarmy v4 is a straight port to Nvidia and nothing else. I hope to have time to work on optimizations in the near future. Also, to all those testing Nvidia, I have reports that setting OPTIM_SIMPLIFY_ROUND to 1 increases performance by +25% on some Nvidia GPUs. See https://github.com/mbevand/silentarmy/blob/master/TROUBLESHOOTING.md for instructions |
|
|
|
Fresh install of Ubundu 16.04. Using the instructions on the github website installed the SDK only. Can't install the drivers because I am running R9 280x and AMD doesn't make drivers for 16.04 for 280x so was going to use the default ones. I get here and I am stuck. Any help will be appreciated!
mine@mine:~$ cd silentarmy
mine@mine:~/silentarmy$ make
echo 'const char *ocl_code = R"_mrb_(' >_kernel.h
cpp input.cl >>_kernel.h
echo ')_mrb_";' >>_kernel.h
gcc -std=gnu99 -pedantic -Wextra -Wall -ggdb -Wno-deprecated-declarations -Wno-overlength-strings -I"/opt/AMDAPPSDK-3.0/include" -c -o main.o main.c
gcc -std=gnu99 -pedantic -Wextra -Wall -ggdb -Wno-deprecated-declarations -Wno-overlength-strings -I"/opt/AMDAPPSDK-3.0/include" -c -o blake.o blake.c
gcc -std=gnu99 -pedantic -Wextra -Wall -ggdb -Wno-deprecated-declarations -Wno-overlength-strings -I"/opt/AMDAPPSDK-3.0/include" -c -o sha256.o sha256.c
gcc -o sa-solver main.o blake.o sha256.o -rdynamic -L"/opt/amdgpu-pro/lib/x86_64-linux-gnu" -lOpenCL
/usr/bin/ld: cannot find -lOpenCL
collect2: error: ld returned 1 exit status
Makefile:20: recipe for target 'sa-solver' failed
make: *** [sa-solver] Error 1
mine@mine:~/silentarmy$
The amdgpu-pro driver provides libOpenCL.so, so if you didn't install it, it's normal that you get this error. The solution to your pb is install Ubuntu 14.04 which will provide the fglrx driver (works with your card), and will provide libOpenCL.so. But, for those who want to try a hack, I wonder if you could get a 280X to work on 16.04 by simply using the free amdgpu driver installed by default (don't install amdgpu-pro), and manually installing only the libopencl1-amdgpu-pro and opencl-amdgpu-pro-icd packages that come with amdgpu-pro (you need some dpkg/apt hackery to ignore other dependencies). Of course this is a big hack, but I wonder if it could work... |
|
|
|
and do you get daily cuts of his dev fee? seems unfair if you dont.
Well it's the intent of my choice of the MIT license. I want to allow such uses, at the condition my copyright notice is included. |
|
|
|
I have a question for you: why does 1 of your 2 servers behind equihash.eu.nicehash.com returns a 17 bytes nonce (37.58.117.214) and the other a 5 bytes nonce (5.153.50.217)? Only the latter is supported by silentarmy.
Please revise the official stratum documentation at https://slushpool.com/help/#!/manual/stratum-protocol. There are no such limitations, a nonce can be even larger. The fact that you're receiving different nonce sizes is not related to a particular NiceHash stratum endpoint. It is simply a fact that NiceHash is forwarding work from different hashing power orders, which are working on different target pools. Anyway, 17 bytes nonce is the largest (max) that you'll receive from NiceHash stratum and even at this size you still have 3 bytes for iteration. So there is no reason why not to accept this kind of work. Please fix this as soon as possible, since our users would really love to use your miner and are now not able to do so due to these limitations of your miner. Thank you and keep up the good work! I am trying to be conservative/safe. Because the fewer bytes you have to iterate the nonce, the higher the chance is that a fast multi-GPU rig will have 2 GPUs that accidentally work on the same nonce. I am planning to release a significant speedup where an Equihash round might take 20ms or less. With 7 GPUs running 3 Equihash instances, they will exhaust 1% of the 3-byte nonce space in 150 seconds. So there is a ~1% chance that duplicate work is being performed by 2 Equihash instances. In practice this is not so much a problem because new mining jobs are pushed more frequently than every 150 seconds, but still... this does not give me the absolute confidence of correctness that I like SILENTARMY to have. Anyway, I pushed a nicehash-compatibilty update to tolerate stratum servers fixing 17 bytes of the nonce: https://github.com/mbevand/silentarmy/commit/0d371a9582159627b11941c14ca6cbc3da359cb1Perhaps I will eventually permanently address the issue by allowing the last 12 bytes of the nonce to be non-zero... |
|
|
|
I've did some research about error, that i posted before: Solver 3.0: unexpected banner "Maximum number of clients reachedMaximum number of clients reachedSILENTARMY mining mode ready" miner is reached Xorg max clients limit - in Debian it's 256. When i look into lsof -U output, i saw, that every sa-solver opens 17 connections, so we have 17 (per sa-solver) * 7 (GPUs) * 2 (thread per GPU) = 238 connections just for solvers and a few connections from main process: http://sprunge.us/UgXB - may be more just after startup because i see only 16 connections from last 2 threads. So, when i removed banner checking from silentarmy script 17 connections per sa-solver? On an Ubuntu 16.04 system with the amdgpu-pro 16.40 drivers, I notice each sa-solver opens only 1 connection. Example with 4 sa-solver instances: $ lsof -U | grep sa-solver
sa-solver 2121 mrb 0u unix 0x0000000000000000 0t0 342442 type=STREAM
sa-solver 2122 mrb 0u unix 0x0000000000000000 0t0 342446 type=STREAM
sa-solver 2123 mrb 0u unix 0x0000000000000000 0t0 343439 type=STREAM
sa-solver 2124 mrb 0u unix 0x0000000000000000 0t0 341927 type=STREAM
|
|
|
|
I just sent a donation to your address.
Thank you for the great work!
Thank you very much! 0.1 ZEC is very generous! |
|
|
|
Let this thread serve as a public record to show that Claymore and I resolved this licensing issue. It is no longer a problem. Some people ran "strings" on Claymore's Zcash binary and saw function names and variables that matched my SILENTARMY miner: https://forum.z.cash/t/silentarmy-v3-now-a-full-miner-multi-gpu-stratum/3580/629I confirmed Claymore copied my code beyond doubt: https://forum.z.cash/t/silentarmy-v3-now-a-full-miner-multi-gpu-stratum/3580/641Because SILENTARMY is released under the MIT license, Claymore was obligated to include my copyright notice, but he failed to do so. So I privately asked him to do so in a private message: Claymore, you took and reused my blake.c code and didn't include the copyright notice as required by silentarmy's MIT license: https://forum.z.cash/t/silentarmy-v3-now-a-full-miner-multi-gpu-stratum/3580/640I request that you cease infringing by: (1) either including the full "License" section of silentarmy's README.md along with your distribution of your zcash miner, (2) or stop using my blake.c code. Of course this is only ~100 lines of code, and we are all lazy programmers and I understand why you did it Smiley but you ought to do the right thing. Thanks, -Marc Bevand However some users were very bothered by this because Claymore charges a 2.5% "dev fee" in his miner, so they took the issue publicly in his Zcash miner thread. He replied a few hours later, publicly in his miner thread, admitting fault and looking for a resolution: I had to remove v1.0 and v1.1 from google drive and mega, because I was pointed that I used a small part of mrb's work.
It will take up to 24 hours to revise my code and create own implementation, then I will upload updated version.
Details are:
[mrb's message]
Oh, you are correct, I really missed this point. I was in a big hurry to release first version asap, so I concentrated on main Wagner's algo,
blake is not interesting for me at this point since it takes only 10% of total GPU time and it's trivial in fact, blake algo is not that I can call something new.
But to make some start point asap I included your blake.c into my sources and then forgot about it and did not create my implementation. That's what happens if you are in a hurry.
It is very unpleasant situation for me, I will do the following:
1. I remove v1.0 and v1.1 versions of my miner. In fact, I already removed current packages from google drive and MEGA, I will upload updated versions soon.
In next version I will remove your blake.c from my sources.
2. I apologize, it's only my fault that I used this part of your code. I was not going to do it in public version.
3. I want to send you some compensation from my ZEC dev fee if you don't mind, just send me your ZEC address.
4. I'm not going to hide my fault in any way, I will post this message in my miner thread in first post.
He first withdrew his package from the Internet, then later re-added them with his readme file updated to include the license: FILES UPDATED TO INCLUDE MIT LICENSE DETAILS IN README.
People started asking me about miner package in PM and some people started to share alternate links for downloading, it's not a good thing because of possible viruses.
So according MIT license I added notice to README and uploaded updated miner packages back to original links. You can also see a license notice in original post of this thread.
I verified that he added the copyright notice and license as required by the MIT license: I had to remove v1.0 and v1.1 from google drive and mega, because I was pointed that I used a small part of mrb's work.
It will take up to 24 hours to revise my code and create own implementation, then I will upload updated version.Details are: mrb> Claymore, you took and reused my blake.c code and didn't include the copyright notice as required by silentarmy's MIT license:
mrb> https://forum.z.cash/t/silentarmy-v3-now-a-full-miner-multi-gpu-stratum/3580/640
mrb> I request that you cease infringing by:
mrb> (1) either including the full "License" section of silentarmy's README.md along with your distribution of your zcash miner,
mrb> (2) or stop using my blake.c code.
mrb>
mrb> Of course this is only ~100 lines of code, and we are all lazy programmers and I understand why you did it :) but you ought to do the right thing.
mrb> Thanks,
mrb> -Marc Bevand
[/quote]
Oh, you are correct, I really missed this point. I was in a big hurry to release first version asap, so I concentrated on main Wagner's algo,
blake is not interesting for me at this point since it takes only 10% of total GPU time and it's trivial in fact, blake algo is not that I can call something new.
But to make some start point asap I included your blake.c into my sources and then forgot about it and did not create my implementation. That's what happens if you are in a hurry.
It is very unpleasant situation for me, I will do the following:
1. I remove v1.0 and v1.1 versions of my miner. In fact, I already removed current packages from google drive and MEGA, I will upload updated versions soon.
In next version I will remove your blake.c from my sources.
2. I apologize, it's only my fault that I used this part of your code. I was not going to do it in public version.
3. I want to send you some compensation from my ZEC dev fee if you don't mind, just send me your ZEC address.
4. I'm not going to hide my fault in any way, I will post this message in my miner thread in first post.
Thank you for doing the right thing. I fetched your new tarball Claymore's ZCash AMD GPU Miner v1.1 Beta - LINUX - Catalyst 15.12.tar.gz and can confirm you included the MIT license as mandated in your "Readme!!!.txt" file, as quoted below. You are in no way obligated to "pay" me. But if you feel this can help re-establish some of your image and reputation, feel free to publicly send to my donation address t1cVviFvgJinQ4w3C2m2CfRxgP5DnHYaoFC You are now fully compliant with the MIT license. I hope my SILENTARMY code was useful to you  ==========
This miner uses some code of SILENTARMY OpenCL GPU Zcash Equihash solver:
# License
The MIT License (MIT)
Copyright (c) 2016 Marc Bevand
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"),
to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense,
and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-Marc Bevand So the issue is now all resolved. |
|
|
|
I had to remove v1.0 and v1.1 from google drive and mega, because I was pointed that I used a small part of mrb's work.
It will take up to 24 hours to revise my code and create own implementation, then I will upload updated version.Details are: mrb> Claymore, you took and reused my blake.c code and didn't include the copyright notice as required by silentarmy's MIT license:
mrb> https://forum.z.cash/t/silentarmy-v3-now-a-full-miner-multi-gpu-stratum/3580/640
mrb> I request that you cease infringing by:
mrb> (1) either including the full "License" section of silentarmy's README.md along with your distribution of your zcash miner,
mrb> (2) or stop using my blake.c code.
mrb>
mrb> Of course this is only ~100 lines of code, and we are all lazy programmers and I understand why you did it :) but you ought to do the right thing.
mrb> Thanks,
mrb> -Marc Bevand
[/quote]
Oh, you are correct, I really missed this point. I was in a big hurry to release first version asap, so I concentrated on main Wagner's algo,
blake is not interesting for me at this point since it takes only 10% of total GPU time and it's trivial in fact, blake algo is not that I can call something new.
But to make some start point asap I included your blake.c into my sources and then forgot about it and did not create my implementation. That's what happens if you are in a hurry.
It is very unpleasant situation for me, I will do the following:
1. I remove v1.0 and v1.1 versions of my miner. In fact, I already removed current packages from google drive and MEGA, I will upload updated versions soon.
In next version I will remove your blake.c from my sources.
2. I apologize, it's only my fault that I used this part of your code. I was not going to do it in public version.
3. I want to send you some compensation from my ZEC dev fee if you don't mind, just send me your ZEC address.
4. I'm not going to hide my fault in any way, I will post this message in my miner thread in first post.
Thank you for doing the right thing. I fetched your new tarball Claymore's ZCash AMD GPU Miner v1.1 Beta - LINUX - Catalyst 15.12.tar.gz and can confirm you included the MIT license as mandated in your "Readme!!!.txt" file, as quoted below. You are in no way obligated to "pay" me. But if you feel this can help re-establish some of your image and reputation, feel free to publicly send to my donation address t1cVviFvgJinQ4w3C2m2CfRxgP5DnHYaoFC You are now fully compliant with the MIT license. I hope my SILENTARMY code was useful to you ==========
This miner uses some code of SILENTARMY OpenCL GPU Zcash Equihash solver:
# License
The MIT License (MIT)
Copyright (c) 2016 Marc Bevand
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"),
to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense,
and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-Marc Bevand |
|
|
|
I went back and tried to install the official AMD drivers thinking that was the problem.
The following packages have unmet dependencies:
unity-control-center : Depends: libcheese-gtk23 (>= 3.4.0) but it is not going to be installed
Depends: libcheese7 (>= 3.0.1) but it is not going to be installed
E: Error, pkgProblemResolver::Resolve generated breaks, this may be caused by held packages.
mine@mine:~/silentarmy$
I am on 14.04 using 5x 280x. I was on 16.04 until I realized that they don't make drivers for the 280x for that version of Ubuntu so I downgraded. Doesn't matter to me which version I am on. This is strictly a mining machine. Thanks for your time.
If you are on 14.04 there is no need to install the official AMD drivers. Just "sudo apt-get install fglrx" as documented in silentarmy's README.md The fglrx package is the drivers nicely repackaged/redistributed by Ubuntu. |
|
|
|
It is wrong way from the start. You'll never achieve CUDA speeds with OpenCL on nVidia.
Not true. According to solardiz on zcash-miner-dev: "JtR -jumbo and Hashcat both went OpenCL for NVIDIA, obsoleting/dropping CUDA support" and performance was acceptable. The reality is that most CUDA developers don't write good OpenCL code so they blame the platform, but it's their code that is the problem. |
|
|
|
What about performance ? with eXtremal's SA version on nvidia i got 30 s/s on one card, and 22 per card on 6-gpu rig. The bottleneck was high cpu usage.
I don't know we will have to see. OpenCL code has a tendency to need to be tweaked to run well on a device. Current silentarmy runs really well on AMD, but it may not be as efficient on Nvidia. So I will have work on optimizing for Nvidia and this is going to be incremental improvements released over time. |
|
|
|
|
frozn0000: try "clinfo" - does it report your GPUs?
|
|
|
|
The author of the code can't know if it was infringed without seeing the source.
I am the author. And yes it was infringed. I can know with 100% confidence, without seeing claymore's full source. How can I know? Because part of his source leaked in his binary. We can see the content of the assert statements, the name of the variables in these asserts, he name of the functions, the order, type, and number of arguments. Everything matches perfectly my blake.c implementation. And I wrote it myself, from scratch, with no inspiration from anyone else. Now, I don't personally believe Claymore copied all of silentarmy. I think he just took blake.c which is only ~100 lines of code. That's why his miner is way different than mine. The authors solver is slower, uses more VRAM, uses a lot more CPU, and PCIe bandwidth, runs on fewer cards, and crashes a lot. Why would you want to copy that?
Actually Silentarmy is faster (by a hair) than Claymore's But I do use more CPU and more PCIe bandwidth :-( That said, I encourage everybody to CHILL OUT here. I didn't want this issue to be public in the first place (I messaged Claymore privately), but some started discussing here... Give Claymore time to respond. All he has to do is include a copyright notice in his miner, and he will be compliant with the MIT license. That's all. |
|
|
|
Is the NV support being done in OpenCL or porting kernel to cuda?
sure windows support is needed, but NV should have priority, tested linux fee miner it goes 50+ on 1070 so I guess a lot more optimization is possible
It's going to be in OpenCL. It's only going to be a <200-line patch to silentarmy. Way simpler than rewriting everything in CUDA. |
|
|
|
|
Show Posts
