Arasan 23.0

[Dokterchen]

When you start arasan from the command line, uci -> go depth 5 -> it says the NNUE file is loaded and so I ran it without setting the NNUE option. When I saw the bad results (similar to the current) I added the NNUE option and went to bed. Still not good and so I aborted the tournament.

Not clear what is going on there but I notice when loading it as a UCI engine in Fritz the network name is blank in the UCI options - so w/o setting that it may not be loading the network.

It might be helpful when there would be message like "NNUE file xyz is loaded" at startup. At least under Areba I see no message.

[Rebel]

When you start arasan from the command line, uci -> go depth 5 -> it says the NNUE file is loaded and so I ran it without setting the NNUE option. When I saw the bad results (similar to the current) I added the NNUE option and went to bed. Still not good and so I aborted the tournament.

Not clear what is going on there but I notice when loading it as a UCI engine in Fritz the network name is blank in the UCI options - so w/o setting that it may not be loading the network.

It's loaded IMO, see my previous post - http://talkchess.com/forum3/viewtopic.p ... 11#p904611

[peter]

- The Windows installer that includes the Arasan GUI does not ship with a network and has NNUE disabled by default, because it is a 32-bit executable and NNUE is not likely to perform well in that environment.
- Only the AVX2 builds have SIMD support for the NNUE eval, so non-AVX2 builds will not perform nearly as well with NNUE enabled.

Thanks for the new version!
Because there's some doubt about loading of NNUE with other compiles by other posters in this thread:
Downloaded "engine only" Win64x compiles, avx2 installed in Arena, Deep Shredder 13 and Fritz 14 at Ryzen 5950 Win10.

All installations run without error- messages, NNUE seems to be loaded (regardless, if or if not name and or path to net is set additionally in UCI- option), just guessing by nodes- count, with NNUE- usaged checked there's about 1 third less n/s of unchecked pondering. (single threaded 650+ at starting position NNUE- using to 850+ kn/s NNUE-less).

In Shredder- GUI all NNUE- engines I've installed give some message about using certain NNUE- file, Arasan doesn't, seeing no difference with adding name of net neither, some feedback to GUI maybe would be fine to be sure, even if I for my personal pov yet wouldn't need it neither.

Having a UCI- path means you plan to update nets only, maybe?
Thanks again regards

[jdart]

I have just now pushed an update that fixes the issue with the default NNUE option values under UCI.

The console app does output a message on startup saying what net it is loading but it's not prefixed with "info" so may not be visible in the GUI. You should see it in debug output, though.

--Jon

[jdart]

Re the questions about the network:

The network is compatible with the Stockfish 13 network architecture.

The initial network was trained using 1B or so selfplay positions using nodchip's tuner.

The first gen network was then used to generate 2B selfplay games and those were used for another tuning run, generating the current network.

I still expect this network is not optimal but it does seem to perform well. Because it's external to the program the network is replaceable should I generate a better one.

[Rebel]

I have just now pushed an update that fixes the issue with the default NNUE option values under UCI.

The console app does output a message on startup saying what net it is loading but it's not prefixed with "info" so may not be visible in the GUI. You should see it in debug output, though.

--Jon

The problem for non AVX2 compiles remains the same, only 86K NPS. And the 32 bit version is even faster 101K NPS.

[jdart]

[The problem for non AVX2 compiles remains the same, only 86K NPS. And the 32 bit version is even faster 101K NPS.

That's a known issue: no SIMD support for non-AVX2 architectures. On the to-do list. For right now I just recommend disabling NNUE if not on AVX2.

[Rebel]

[The problem for non AVX2 compiles remains the same, only 86K NPS. And the 32 bit version is even faster 101K NPS.

That's a known issue: no SIMD support for non-AVX2 architectures. On the to-do list. For right now I just recommend disabling NNUE if not on AVX2.

An AVX compile will do. Most older computers support it.

[pohl4711]

I have just now pushed an update that fixes the issue with the default NNUE option values under UCI.

The console app does output a message on startup saying what net it is loading but it's not prefixed with "info" so may not be visible in the GUI. You should see it in debug output, though.

--Jon

The Engine (23.0.1) plays now with expected strength in cutechess-cli, but after each game, cutechess-cli prints a warning-message:
"EngineProcess: Process is destroyed while engine is still running".

So, there is definitly still an issue here!

[Gabor Szots]

Trying to compile it under Windows using the makefile I get this message: Makefile:41: *** missing separator. Stop.

I searched the web and found that this could be caused by using spaces in the makefile instead of tabs. However, whatever I do the message remains.