The FRC problem should be fixed now. I asked Emil to simply replace the package so you can simply re-download.
There's no need to change version number because it has zero impact on regular chess,
meaning people that downloaded the original package can continue testing (unless they want Silvian's new logo of course
mar wrote:The FRC problem should be fixed now. I asked Emil to simply replace the package so you can simply re-download.
There's no need to change version number because it has zero impact on regular chess,
meaning people that downloaded the original package can continue testing (unless they want Silvian's new logo of course
Not statistically significant but I ran a little match with the new cheng against bright 0.5c. According to CCRL, the previous cheng and bright were very close in rating. In my little sample cheng won 7 games, drew 4 and lost none! That is about 200 ELO difference between the two engines. Of course in reality, that is too small a sample but I'm guessing a gain of at least 100 ELO over the previous version.
Very nice work, well done Martin Sedlak! My machine is an i7 with 4 cores that likes to hyper thread. Engines that pick their own thread numbers volunteer "8 CPU" when I load them. So for these games the readout for both engines was "8 CPU". The games were fast; 3 mins + 2" bonus per move.
Results look good so far and it seems that the real improvement may be around 60 elo. Still too early though. I measured nearly 100 elo in bullet (selfplay), of course this doesn't mean much.
cheng defaults to 1 thread which I consider a good practice. I would expected that the GUI tells the engine how many cores to use (as instructed by the user of course).