asdlei99 / cpuminer-gr-avx2

Optimised Version of GR miner for RTM

cpuminer-gr is a fork of cpuminer-opt by Jay D Dee which is a fork of cpuminer-multi with optimizations imported from other miners developped by lucas Jones, djm34, Wolf0, pooler, Jeff garzik, ig0tik3d, elmad, palmd, and Optiminer, with additional optimizations by Jay D Dee.

All of the code is believed to be open and free. If anyone has a claim to any of it post your case in the cpuminer-gr by email.

Miner programs are often flagged as malware by antivirus programs. This is a false positive, they are flagged simply because they are cryptocurrency miners. The source code is open for anyone to inspect. If you don't trust the software, don't use it.

There is NO official bitcointalk thread about this miner. It is due to unjust ban after posting about first release about this miner.

See file RELEASE_NOTES for change log and INSTALL_LINUX or INSTALL_WINDOWS for compile instructions.

Requirements

1. A x86-64 architecture CPU with a minimum of SSE2 support. This includes Intel Core2 and newer and AMD equivalents. Further optimizations are available on some algoritms for CPUs with AES, AVX, AVX2, SHA, AVX512 and VAES.

ARM and Aarch64 CPUs are not supported, yet.

2. 64 bit Linux or Windows OS. Ubuntu and Fedora based distributions, including Mint and Centos, are known to work and have all dependencies in their repositories. Others may work but may require more effort. Older versions such as Centos 6 don't work due to missing features. 64 bit Windows OS is supported with mingw-w64 and msys or pre-built binaries.

MacOS, OSx and Android are not supported.

3. Stratum pool supporting stratum+tcp:// or stratum+ssl:// protocols or RPC getwork using http:// or https://. GBT is YMMV.

Supported Algorithms

                      gr            Ghost Rider (RTM)

Quick Setup

  To add or use options from the miner, use included config.json file.
  All options should be presented in JSON format like:
  "long-flag-name": "Some_value"

  Some examples:
  "tune-full": true
  "tune-config": "some_filename"
  "url": "stratum+tcp://YOUR_POOL_ADDRESS:PORT"
  "user": "YOUR_WALLET"

For full miner option list and other tips please read the readme.txt file.

Tuning

Tuning starts automaticaly with the start of the miner. If previous tuning file tune_config exists (or --tune-config=FILE flag is used), it is used instead. This behavior can be overridden by --no-tune or --force-tune. On non-AVX2 CPUs default tuning process takes ~69 minutes to finish. On AVX2 CPUs default tuning process takes ~155 minutes to finish.

A small explanation of what tuning does. The traditional way of hashing would be, take some input, hash it to generate output hash. That is what would be called normal hashing (aka 1way) as we are doing 1 hash at a time. What we can do is to hash 2 or 4 hashes at the same time! Due to different variants used in each block etc, the memory requirement changes and we would like to have as much of it in the cache as possible as RAM is SLOW! So 1way would require 128KiB, 256KiB, 256KiB, 512KiB, 1MiB or 2MiB to be stored somewhere if we wanna solve it. 2way would need 2x of that amount and 4way 4x. In some cases, like in 256KiB variants, that would increase the requirements to something like 1MiB or up to 2MiB for the 512KiB variants. In most cases, this amount of data can fit in the cache which can bring additional performance. Putting too much can decrease it imagine 8MiB for 4way Fast (2MiB variant) (in some cases it is still fastest if your CPU lacks cache, like i3 or some mobile CPUs). OK, so there are 6 variants, that make 20 possible "rotations". And we check all those 20 rotations to see if we use 1way, 2way, or 4way on each of them brings improvement or not. We cannot check it individually as to when they are hashing using all of them, it might not be accurate anymore so we have 8 scenarios per rotation for AVX (only 1way or 2way, 2^3, 2 ways of solving for 3 variants) and 27 for AVX2 (1way, 2way, 4way, 3^3, 3 ways of solving for 3 variants) -> this is --tune-full that checks everything. --tune-simple only checks 4way on Turtle and Turtlelite variants and default also check Dark and Darklite ones as those are the most likely to be used or benefit CPUs from all our testing. That way we can use the most amount of cache and use it as efficiently as possible. The question you might ask, how is doing 2way faster than doing just 1way 2 times? That is coz we can use some more parallelization and other tricks to notify the CPU of what is coming so it can prepare data faster.

Ghost Rider (GR)

Ghost Rider (GR) algorithm that is used to mine RTM consists of 15 "core" algorithms (same as in X16 without SHA) and 6 different variants of Cryptonight which only 3 are used for hashing. Each block (in part of the previous block in reality) dictates what should be the order those algorithms are calculated in and which of the Cryptonight variants should be used. It goes like this: 5 core algorithms, 1 Cryptonight, 5 core, 1 Cryptonight, 5 core, 1 Cryptonight. As you can see all core algorithms are always used but only 3 out of 6 variants of Cryptonight. Those Cryptonight parts are the slowest/"hardest" part of the whole hashing process. Core algorithms perform very well on pretty much every CPU but Cryptonight requires a specific amount of memory, if that memory can be fully or mostly stored in a cache, that will increase the performance significantly (the main reason why AMD Ryzens are so much faster than Intels as they have so much more L3 Cache). The variants of CN use 128KiB, 256KiB, 256KiB, 512KiB, 1MiB, or 2MiB or memory. In addition, the larger the memory needed, the more iterations through it (making it even slower :P). The performance of those algorithms can vary A LOT. You might be getting like 1500H/s if the blocks want the 3 slowest variants (The ones using most memory) and 10000+H/s if the blocks use the fastest ones (Using my 2x2698v3 as an example). This means 2 things. First, the hash rate like you notice is very volatile and changes almost always with each block. The miner should show you Cryptonight variants used in the current block, Turtlelite, Turtle, Darklite, Dark, Lite, Fast - those are variants in the same order as I mentioned memory, irony is that the Fast variant is the slowest one and Turtlelite is the fastest :). Second, some blocks (in most parts the ones using easier variants) are found faster and not in a very consistent manner. All should average to around 120s per block tho in the long term.

Bugs

Users are encouraged to post their bug reports using git issues or on official RTM Discord or opening an issue in git:

https://discord.gg/2T8xG7e

https://github.com/WyvernTKC/cpuminer-gr-avx2/issues

All problem reports must be accompanied by a proper problem definition. This should include how the problem occurred, the command line and output from the miner showing the startup messages and any errors. A history is also useful, ie did it work before.

Donations

Any kind but donations are accepted. Jay D Dee's BTC: 12tdvfF7KmAsihBXQXynT6E6th2c2pByTT

This fork introduces 1.75% donation on added Ghost Rider (GR) algorithm only.

If you wanna support us, any donations are welcome:

Ausminers:

RTM: RXq9v8WbMLZaGH79GmK2oEdc33CTYkvyoZ

Delgon:

RTM: RQKcAZBtsSacMUiGNnbk3h3KJAN94tstvt ETH: 0x6C1273b5f4D583bA00aeB2cE68f54825411D6E8c

Happy mining!