LCZero vs Dedicated machine from 1990s

[Werewolf]

At this time, April 3rd 2018, I estimate the web version with 2000 rollouts to be 1800 elo and the version playing on my PC with 100,000 rollouts to be 2200 elo.

This is an incredible step forward from last month.

However, its tactics are still awful. Just enter in these moves:

[pgn] 1. e4 e5 2. Nf3 d6 3. Nc3 g6 4. Bc4 Bg4 [/pgn]

and it cannot find 5.Ne5! which almost any chess computer from the 1980's could find! (Master & Vancouver 68000 only take about 1 second)

[IanO]

Dedicated oldies are good measuring sticks for weak engines. Maybe get the German forum interested, there is a whole spectrum of even weaker oldies which could be use to measure this initial progress. Check the dedicated Activ list for possibilities.

I have about 10 dedicated machines and using the MESS emulator access to about 30 more.

There are weaker ones i could try, yes. But now I think The Master (Milano Pro) may really be running out of time. It won't keep winning for long.

I'm finding that that Morsch's tactical oldies fare better than most. Even the lowly Cosmos at default 5s hasn't lost a game yet to the 2000 playout level on the web. On the other hand, the more positional Genius from Lang keeps simplifying to draws and the Sapphire II tends to tangle up its pieces.

Looking forward to the new 10x128 nets on the website http://play.lczero.org/

[IanO]

Dedicated oldies are good measuring sticks for weak engines. Maybe get the German forum interested, there is a whole spectrum of even weaker oldies which could be use to measure this initial progress. Check the dedicated Activ list for possibilities.

I have about 10 dedicated machines and using the MESS emulator access to about 30 more.

There are weaker ones i could try, yes. But now I think The Master (Milano Pro) may really be running out of time. It won't keep winning for long.

I'm finding that that Morsch's tactical oldies fare better than most. Even the lowly Cosmos at default 5s hasn't lost a game yet to the 2000 playout level on the web. On the other hand, the more positional Genius from Lang keeps simplifying to draws and the Sapphire II tends to tangle up its pieces.

Looking forward to the new 10x128 nets on the website http://play.lczero.org/

Actually, I was playing the new nets; they must have gotten substantially weaker! I guess this is either because they are starting from scratch or 2000 playouts is not enough to keep up with even the basic tactical ability of the oldies.

I played against the weakest in my stable, an Elite Avant Garde, and it couldn't even manage a win against the lowest level! Even when it trapped two pieces (!) it eventually let them go and settled for a repetition draw.

Is there a minimum number of playouts for exhibiting this alleged +200 elo improvement?