Skip to content

rieMiner 0.9
Compare
Choose a tag to compare
@Pttn Pttn released this 15 Dec 03:25
· 158 commits to master since this release
0.9
5b31951

rieMiner 0.9 is finally out :D !

Standalone Windows binaries provided (also for Linux, compiled on Debian 9; not totally static but should work on Debian or Ubuntu fresh installs). Also, a PDF manual. L versions are without assembly and precomputed data optimizations, but in 99% of the cases, you should use the Win64/Deb64 versions.

Updates from fastrie, the previous reference Riecoin miner:

  • Adds solo mining with a Riecoin Core wallet using the GetBlockTemplate protocol. Supports custom payout addresses (P2PKH and P2SH);
  • Adds a Benchmark Mode to allow much easier performance comparisons;
  • Some bug fixes like the long delay before Stratum mining starts or some random segmentation faults;
  • One important bug fix is the CPU Underuse: when running with 16 threads or more with fastrie, the CPU usage would not be maxed out at current difficulties. It could be worked around by launching multiple instances with each less threads, (not practical at all and wastes a lot of memory), or reducing the Prime Table Limit (significantly lower 6-tuples find rate). Note that the issue is actually still present (notably in Testnet), but negligible in practice;
  • Performance enhancements (threading improvements, assembly optimizations, precomputed data usage). From fastrie, you will get a gain of around 5-10% for the same settings and computer. You can thank Michael Bell for all his contributions;
  • The assembly optimizations include usage of SSE, and AVX, AVX2 or AVX512 when available. AVX512 support is experimental and not much tested, but a huge performance increase of ~20% over AVX2 has been reported! As a trade off, the CPU needs to be x64 with SSE to be able to run rieMiner. However, a "Light" branch is provided in order to support older CPUs and other architectures (you can for example mine Testnet blocks with a Pentium II computer or a RaspBerry Pi);
  • Now supports prime table limits (PTL, also called "Sieve Max") higher than 2^32, to get even better speeds by reducing the ratios between k and (k + 1)-tuples. Indeed, using a PTL of 2^33 instead of 2^31 (which is not possible on fastrie) for benchmarking with Difficulty 1600 increased the blocks/day metric a lot (about 25%)! However, in order to set such a high PTL, you would need at least 24 GB of RAM;
  • Support of prime constellations types other than the Riecoin one (rieMiner can be configured to find 7 or 8-tuples, for example);
  • Use of configuration files instead of command line options, much more parameters to customize;
  • Complete source code refactoring for a much more modern and developer friendly code;
  • Cleaner user interface, more interesting statistics.

For comparing performance with fastrie, you can use the rieMiner's Benchmark Mode and my fastrie fork (benchmark branch). Note that you need to test long enough to get precise numbers.
Examples of results (in primes per second, Standard Benchmark): 221.0 pps -> 233.4 pps for a 2700X @ 4 GHz, and 81.3 pps -> 89.3 pps for a 6700K @ 3 GHz (respectively 5.6% and 9.8% increases).
For the 2700X, the fastrie benchmarking had to be done with 2 instances of 8 threads each. Else, it would suffer a lot from the CPU Underuse Bug.

Happy Mining :D !

Assets 3
Loading