8.0.0 Release
Yes it has been a quite while and I haven't worked on Rice much. However, there's been some search improvements made about 8-9 months ago and recently Rice has got a new net hexadecane_512_v2.net which has new architecture of 16 * 768 -> 512x2 -> 1. Reducing down to 512x2 from previous 768x2 hidden layer, has given quite a speed boost along with the strength from more buckets.
Elo difference should be ~70 elo from previous version, I haven't got time to test it so yeah that's the estimate.
As usual, binaries built for windows are given.
rice-8.0.0-v2.exe for POPCNT, SSE3, SSE4.1, SSE4.2, SSSE3 CPUs.
rice-8.0.0-v3.exe for AVX, AVX2, BMI1, BMI2 CPUs.
The speeds of these binary tested on i7 14700KF are:
rice-8.0.0-v2.exe ~0.95 MNPS
rice-8.0.0-v3.exe ~3.21 MNPS
If you face any bugs, feel free to open an issue and I'll look into it.
Cheers!
What's Changed
- Try out idea from StockDory in Aspiration Window by @rafid-dev in #31
- Introduce cutnodes and use them in extensions by @rafid-dev in #32
- Perform IIR on cutnodes according to cj5716 by @rafid-dev in #34
- Fix UCI go infinite by @Sazgr in #33
- Refactor NNUE and fix bug on refreshing by @rafid-dev in #35
- Don't adjust mate score in tt by @rafid-dev in #36
- Mate distance pruning by @rafid-dev in #37
- Adjust mate score in TT by @rafid-dev in #38
- Reduce TT entry size by @rafid-dev in #39
- Tweak replacement strategy by @rafid-dev in #40
- Tweak with SEE values by @rafid-dev in #41
- Continuation Pruning by @rafid-dev in #42
- Introducing new net by @rafid-dev in #43
- Update net to hexadecane_512_v2 by @rafid-dev in #44
New Contributors
Full Changelog: 7.0.0...8.0.0
