Thanks for your explanation, Gary. Looks like this is the case, it has some kind of search that allows itself to search super fast in the first ten seconds and slowly drag down and take time and still stick with the same move it seearched and found in the first ten seconds at long time controls.gladius wrote:Well, it takes a real hit from using 32 bit, which a lot of tournaments seem to do. Second, it's buggyswami wrote:I just don't understand the whole deal with Garbochess, in one tournament it plays like a 2700 and in other tournaments it plays like a 2300. What's up with it?
maybe a good blitzer but not so consistent at long time controls?. Not buggy in that it will make incorrect moves or anything, but the evaluation features did not do exactly what they were supposed to do, and there were search bugs as well. Since 2.20 I have fixed at least 15 eval. bugs, and a few search problems as well.
Finally, I only test at 1'1, so it's highly optimized for shorter time controls. Still, I'm not sure why the search strength shouldn't scale decently. That may be another bug lurking
I have seen similar cases with other engines, but gap in strength is very rare. means Garbo has an ability to play high level chess in short time controls. With few more fixes, it could play even better at longer TC.
Majority of engines play fairly consistent in both short and longer time controls,Some play better at longer TC, while others play better at blitz . But with the gap of only less than 100 elo.
Garbo is the only engine I have encountered that plays much better at blitz given its strength at longer TC, with bigger gap. Indicates there might be a bug in search.
