Anyone have result with Ryzen and Hexxcoin ? ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif) . I think it would be nice if some could compile new version incl. new ryzen cpu. I am trying to get it to work, but somehow ryzen is not so good as intel cpus for mining. Try cpuminer-aes-avx2.exe, that should work. If not try cpuminer-aes-avx.exe.
|
|
|
cpuminer-opt-3.5.13 is released, see OP.
There is a smalll optimization but more importantly for those interested in solo mining there is a major fix to getwork.
I am unable to test the fix but I encourage users to try with their favorite coin and report their results:
1. Basic info on test environment: CPU, OS, algo, coin,
2. Information for each miner used: version, compiled from source or binaries, link if not well known
2. Does solo mining using the wallet miner work?
3. Does cpuminer-opt work solo mining? Provide details including console session showing command line options.
4. Is there another miner available that works solo mining? Provide details.
5. Any technical info available about the algo/coin that affects solo mining, ie similarities/differences with other algos that produce different test results.
|
|
|
Anyone have result with Ryzen and Hexxcoin ? ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif) . There are a couple of threads discussing Ryzen but the only results I've seen are for cryptonight which are in line with Intel based on L3 cache size. I'd like to see a test with lyra2z330 as a real memory stress test regardless of cache. I'd also like to see a test with deep to give those 16 threads a good workout. They represent the extremes of I/O bound and compute bound algos.
|
|
|
I'm looking at updating the errata for old AMD CPUs and would like more precision on which CPUs work and which don't.
Athlon 64 x2 and Phenom II x4 are confirmed not to work. It appears anything less than Bulldozer, which introduced AVX, will not work with all algos.
If no one has data that says otherwise I'll update the errata in the OP.
These CPUs are working fine with legacy versions on zr5 algo. I haven't tested other algos, since I have no interest in them. m7m also works if I recall corectly - will test that on a weekend. Thanks, that helps localize the incompatible code, it's not in code used by zr5 or m7m. Let me know if something is broken using an AVX enabled AMD CPU, that would be unexpected. If that can help you identify (and hopefully fix) broken algos, I will do a full test tomorrow on all algos supported on zpool.ca. That includes Linux and Windows builds of legacy versions (since they are faster on these CPUs). I appreciate the effort but It's not something that I can fix easilly, so the work would only be for information purposes. My main focus is on the newer architectures so I want to make sure all algos work on AVX enabled AMD CPUs. Any problems with them would need to be fixed.
|
|
|
I'm looking at updating the errata for old AMD CPUs and would like more precision on which CPUs work and which don't.
Athlon 64 x2 and Phenom II x4 are confirmed not to work. It appears anything less than Bulldozer, which introduced AVX, will not work with all algos.
If no one has data that says otherwise I'll update the errata in the OP.
These CPUs are working fine with legacy versions on zr5 algo. I haven't tested other algos, since I have no interest in them. m7m also works if I recall corectly - will test that on a weekend. Thanks, that helps localize the incompatible code, it's not in code used by zr5 or m7m. Let me know if something is broken using an AVX enabled AMD CPU, that would be unexpected.
|
|
|
Edit: I just found a bug that would break getwork, It will be fixed in the next release. Until then it's probably not worth it to try testing it.
I took a look at the getwork and gbt code and there are not a lot of algo hooks. It might be possible to fix it for some algos. But this will require a lot of help from user testing and good reporting of test results. Comparing to cpuminer-multi should make it easier.
For anyone interested in helping out please provide the following information.
1. Basic info on test environment: CPU, OS, algo, coin,
2. Information for each miner used: version, compiled from source or binaries, link if not well known
2. Does solo mining using the wallet miner work?
3. Does cpuminer-opt work solo mining? Provide details including console session showing command line options.
4. Is there another miner available that works solo mining? Provide details.
5. Any technical info available about the algo/coin that affects solo mining, ie similarities/differences with other algos that produce different test results
This will hopefuly identify the problem areas that need attention, Having a working model to compare with should make it easier to fix them.
|
|
|
no support for solo mining right? there was a coin with very very low hash, i was thinking to give it a try with cpu, since with gpu would be overkill
but the miner only support stratum connection...
It's not really by intent, I just don't feel like installing wallets for every coin for testing. I have been told some algos work but I have no idea which ones.
|
|
|
I'm looking at updating the errata for old AMD CPUs and would like more precision on which CPUs work and which don't.
Athlon 64 x2 and Phenom II x4 are confirmed not to work. It appears anything less than Bulldozer, which introduced AVX, will not work with all algos.
If no one has data that says otherwise I'll update the errata in the OP.
|
|
|
joblo, it seems that XEVAN is broken on all non-AES AMD CPU's
CPU: Athlon 64 X2 Dual Core Processor 6000+ (MMX, SSE, SSE2, SSE3) OS: Ubuntu 14.04 x64 miner versions: v3.4.12-legacy / -march=core2 - After miner and stratum start > "Illegal instruction (core dumped)" v3.4.12-legacy / -march=btver1 - After miner and stratum start > "Illegal instruction (core dumped)" v3.5.9.1-legacy / -march=core2 - After miner and stratum start > "Illegal instruction (core dumped)" v3.5.9.1-legacy / -march=btver1 - After miner and stratum start > "Illegal instruction (core dumped)" v3.5.12 / -march=core2 - After miner and stratum start > "Illegal instruction (core dumped)" v3.5.12 / -march=btver1 - After miner and stratum start > "Illegal instruction (core dumped)"
CPU: AMD Phenom II X4 965 (MMX (+), 3DNow! (+), SSE, SSE2, SSE3, SSE4A, x86-64, AMD-V) OS: Win 7 Ultimate x64 SP1 miner versions: v3.4.12-legacy / -march=core2 - After miner and stratum start > APPCRASH v3.4.12-legacy / -march=btver1 - After miner and stratum start > APPCRASH v3.5.9.1-legacy / -march=core2 - After miner and stratum start > APPCRASH v3.5.9.1-legacy / -march=btver1 - After miner and stratum start > APPCRASH v3.5.12 / -march=core2 - After miner and stratum start > APPCRASH v3.5.12 / -march=btver1 - After miner and stratum start > APPCRASH
CPU: AMD FX-7600P (MMX (+), SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, SSE4A, x86-64, AMD-V, AES, AVX, XOP, FMA3, FMA4) OS: Win 7 Ultimate x64 SP1 miner versions: v3.4.12-legacy / -march=bdver3 - XEVAN Works v3.4.12-legacy / -march=native - XEVAN Works v3.5.9.1-legacy / -march=bdver3 - XEVAN Works v3.5.9.1-legacy / -march=native - XEVAN Works v3.5.12 / -march=bdver3 - XEVAN Works v3.5.12 / -march=native - XEVAN Works
Illegal instruction pretty much tells the story. With a lot of help I can probably figure out which instruction AMD is tripping over but I won't be able to fix. The instruction isn't implemented in the CPU, it's as simple as that. This kind of imcompatibility is probably worked around by the compiler but that doesn't work for intrinsic functions or assembly code. The only alternative is to use a miner that is free of SSE2 code such as cpuminer-multi.
|
|
|
Hi took me a while to get the miner running on xmr but mainly down to me not knowing what i was doing with bat files etc running a xeon E5 2670 v2 10cores 20 threads was hashing at 340h/s using 12 threads What im looking for is a ZEC cpu miner are you doing the algo so that can be mined
many thanks for your work
cpuminer-opt doesn't support zcash mining. There are a few miners available for zcash, refer to the pool's instructions.
|
|
|
cpuminer-opt v3.5.12 is released with support for sha256t, 29% faster than ocminer version.
joblo, can u tell us the optimal -t (thread number) on this algo? Depending on CPU cache size(s), core count, thread (hyper-threading) count. Tested on i5-4570 (Haswell, 4 cores, 4 threads, 6MB L3), seems that -t 4 is the fastest setting in time, while -t 5 or -t 6 gives maximum peaks (first 3-5 minutes). P.S. The avg speed on my CPU is 4850... 5200 kH/s (sha256t) Thanks. Try different combos and tell everyone what works best. For compute intensive algo it's usuallty best to run N threads. For memory bound algo you'll reach a peak and adding more threads won't help. Cryptonight is special because it is cache bound. There is a sweet spot when everything fits nicely in the cache. Add another thread and speed drops dramatically.
|
|
|
HexxCoin also moved to use lyra2z330 as algo name. This is more easy as its algo name.
That was my intent. I prefer not to use the coin name as algo name.
|
|
|
I have a couple of suggestions for testing. Cryptonight is a good test for cache performance but doesn't stress the CPU in any way. Ryzen (16 MB cache) is equivalent to 2x i7-6700K (8 MB cache each). Very competitive in cache performance.
Testing raw compute performance is better done with the deep algo. It makes generous use of AES, AVX and AVX2 while not being bogged down by memory accesses. It's a good test of Ryzen's AVX/2 and hyperthreading performance.
Lyra2z330 (zoin) is a good test for a memory hard algo. It's extreme memory requirements blow away the cache and make it irrelevent. This tests the memory interface but is also dependetn on the memory itself. Ryzen went with a 2 channel controller so will be at a disadvantage over Intel E series 4 channel controller.
Lyra2rev2 has a good mix of compute intensive and memory intensive functions.
It's good to monitor CPU Usage while testing. As seen with cryptonight a memory or cache bottleneck means the CPU spends a lot of time idle. The CPU needs to be able to hit 100% to show its full compute potential.
|
|
|
For cryptonight the 1800x seems to have an edge over i7-6800K with a slightly larger L3 cache and 2 more hyperthreaded cores but also a higher price.
Cryptonight doesn't stress the cores, it would be nice to see performance on a well optimized compute intensive algo like deep.
|
|
|
It looks like git clone -b is what I needed all along. I also need to change my git push command to origin HEAD. That way I won't screw up and push to the wrong branch. git clone https://github.com/JayDDee/cpuminer-opt.git -b legacy Cloning into 'cpuminer-opt'... remote: Counting objects: 1039, done. remote: Compressing objects: 100% (148/148), done. remote: Total 1039 (delta 54), reused 0 (delta 0), pack-reused 886 Receiving objects: 100% (1039/1039), 2.90 MiB | 1.18 MiB/s, done. Resolving deltas: 100% (401/401), done. Checking connectivity... done. $ git branch * legacy
|
|
|
im currently trying to understand the structure of cpuminer-opt/cpuminer-multi: i want to implement a basic cryptonight hashing function in javascript and/or port the C/C++ part of cpuminer-opt to js with asm.js, though im not sure what the best starting point for this basic task is
is there some documentation for devs which are not familiar with how mining software works?
i suppose writing this in js with libs is fairly easy
Not that I'm aware of. The first step is to identify everything needed for cryptonight. If all you want is the bare hashing function look in algo/cryptonight/cryptonight-aesni.c. If you need to rely on SW AES look in algo/cryptonight/cryptonight.c. There's a lot more to it if you want to build a mining app: UI, stratum/networking, multithreading, algo interface, algo support SW, I'm probably missing a couple. If you want to build a frankenstein you need to find the line between the core SW and the algo SW. That line is primarilly scanhash, though it's a very blurred line. Almost everything above it is core SW, and everything below is algo specific. The intent of algo-gate was to try to better define that line, but you can still see when the line is crossed by the custom target functions the algo has to give to the core. multi and ccminer have a bunch of algo hooks in the core code. I'm not familiar with any other miner architectures but I presume there still exists that basic interface where you input a message and get back a hash.
|
|
|
fetch is only necessary if your local git repo doesnt have the new branch info from the remote(s) yet, after a clone there is no need for a fetch to get the legacy branch directly when cloning one can use: git clone <url> --branch <branch> Thanks. I'll include that in the instructions.
|
|
|
yes, you will need to execute the second command as well, see here: ![](https://ip.bitcointalk.org/?u=https%3A%2F%2Fi.imgur.com%2FTRYBhYS.png&t=663&c=fTQgKBgQxQFjAg) you can use to list all branches (not only the local ones), or for all remote branches You didn't do a fetch. ![Huh](https://bitcointalk.org/Smileys/default/huh.gif) Other that that it looks like what I want. Looking further down the road how does one clone the legacy branch? It seems like there is no option to select either the branch or the commit when cloning, you just get the repo. Not very useful for users who just want to compile the legacy to have to checkout to make the legacy visible.
|
|
|
|