LazySMP wrote: ↑Fri Oct 25, 2024 2:32 amAlthough my engine is 1800 elo, but it is very fast. My engine speed is about 10 million NPS on a single thread. Have you ever seen an engine like this before?
The last version of my 2700ish engine with full move make/unmake and in-check detection in perft had about 35M NPS in perft, single threaded. That's a simple mailbox approach with ray scanner loops, and it does already include MVV-LVA and mobility detection in the move generator.
Speed in itself doesn't mean much. I think 1800 Elo is a good achievement for a first release (so, congrats!), but you'll get tons of Elo by trading speed for features. Your raw NPS will go down anyway once you implement features like mobility detection, history, hash tables, and once you debug your move generator which will most likely fail in weird positions (that's where perft is really helpful). Not to speak of a refined eval which will drag down the speed even further. Just to illustrate that point - my engine also runs on a microcontroller where it has only about 30 kNPS, but that's still in the 2200 Elo ballpark.
If there is nothing in the 2000 Elo engine, why the authors of the 3800 ELO engines (Clover, Rubi, etc) don't want my engine to be tested in the rating list?!
That's not what was said to my understanding, at least not in this thread. The point is rather that it is your job to properly test and debug your engine beforehand. Also, the testing effort on CCRL is not evenly spread; the top engines naturally get more attention. A new Stockfish release is likely to be tested with top priority, but a midrange engine isn't. One or two releases per year should be fine, and expect some weeks delay before they get around to test that. That also implies that you don't want to burn through the scarce testing slots and also annoy testers with buggy releases.
CCRL is like your engine having a date, and you wouldn't attend dates unkempt in a dirty shirt, right?
Modern Times wrote: ↑Fri Oct 25, 2024 5:29 ambut authors are fully entitled to choose if their engines are open source or closed source.
Sure. The original disagreement was not closed/open source, but public/private, and I think private engines shouldn't even be on CCRL unless they are somehow important. As of now, I can only see Torch in that role because of its commercial relevance on one of the largest online chess platforms, and people are surely curious whether its analysis can provide something that they can't get for free with Stockfish.
the clear implication here is that Graham is suspicious that it may be a clone or derivative
I don't think it would be that buggy in this case.
