I just pushed a new version of Asymptote to GitHub. The binaries should appear shortly at https://github.com/malu/asymptote/releases/tag/v0.4. Please note as always that the files named "-popcnt" actually do not use the popcount instruction and are usable by older CPUs. The binaries using the popcnt instruction are inside the archives named "+popcnt".
Testing at short time control (10s+0.1s, 8 MB Hash) show a self-play Elo rating increase of approx. 100:
malu wrote: ↑Mon Jan 21, 2019 5:55 pm
I just pushed a new version of Asymptote to GitHub. The binaries should appear shortly at https://github.com/malu/asymptote/releases/tag/v0.4. Please note as always that the files named "-popcnt" actually do not use the popcount instruction and are usable by older CPUs. The binaries using the popcnt instruction are inside the archives named "+popcnt".
Testing at short time control (10s+0.1s, 8 MB Hash) show a self-play Elo rating increase of approx. 100:
malu wrote: ↑Mon Jan 21, 2019 5:55 pm
I just pushed a new version of Asymptote to GitHub. The binaries should appear shortly at https://github.com/malu/asymptote/releases/tag/v0.4. Please note as always that the files named "-popcnt" actually do not use the popcount instruction and are usable by older CPUs. The binaries using the popcnt instruction are inside the archives named "+popcnt".
Testing at short time control (10s+0.1s, 8 MB Hash) show a self-play Elo rating increase of approx. 100:
Score of v0.4 vs v0.3: 510 - 183 - 307 [0.663] 1000
Elo difference: 117.94 +/- 18.46, LOS: 100.00 %, DrawRatio: 30.7 %
For reference: v0.3 is at about 2500 on the CCRL 40/40 scale.
This probably won't mean a strength gain of 100 Elo point on the usual rating lists.
I sadly won't have time to test the release at longer time controls. But I expect the rating change to be about the same.
Let me say in advance that I am thankful for anyone testing my engine.
Greetings,
Max
Thanks Max. Unfortunately I haven't found any executables for Windows with +popcnt. Where is the +popcnt archive?
Hey Gabor, the windows archives are uploaded now. AppVeyor usually takes a bit longer to build the windows archives than Travis takes with everything else.
Edit: Oops, the windows files are actually named ".popcnt" instead of "+popcnt". So choose the ".popcnt" ones. I'll investigate the naming discrepancy.
malu wrote: ↑Mon Jan 21, 2019 6:22 pm
Hey Gabor, the windows archives are uploaded now. AppVeyor usually takes a bit longer to build the windows archives than Travis takes with everything else.
This release fixes a bug which made the detection of draws by insufficient material fail in many cases. This bug was introduced in v0.4.
Sorry for any inconveniences this may cause.
Looks like one version didn't build, which happens to be the fastest (the msvc x64 windows popcount one). Can you check that one?
Good catch and thanks for letting me know. I've released v0.4.2 which does include a 64bit MSVC popcnt build. There's no difference in playing strength between v0.4.1 and v0.4.2. Anyone using Linux or OS X probably doesn't even have to update.