Introducing Igel chess engine

[Damir]

I was playing on Playchess ( Chessbase server) when the engine crashed and lost on time after 32 moves.

[AlexChess]

Hello Alex,

Please, compile it also for Windows 10 ARM64

Please check the ARM64 NEON build at https://github.com/vshcherbyna/igel/rel ... _3_0_5.exe. The network is already embedded inside, so nothing to configure.

...Sygyzy Tablebases up to 6 man and GMbook8mvs.abk like all the others not compatibles with cerebellum3merge.bin

Entering the Arena also with Igel 3.0.0 NNUE to see the differences. Playing now against a strong "alien" (No source code available to compile for other OSs)



Regards, Alex

[voffka]

I was playing on Playchess ( Chessbase server) when the engine crashed and lost on time after 32 moves.

Thanks Damir. This was a single cpu run or with multiple cpus? With ponder off or on? I may have some issues with ponder I am aware, just checking.

[Damir]

I was using 10 cores with 18 Threads with ponder on yes. It was multiple CPU.

[voffka]

I was using 10 cores with 18 Threads with ponder on yes. It was multiple CPU.

OK, that's good to know. I will debug to see if I can catch those ponder issues.

[voffka]

Damir,

Do you have by any chance a debug log from the Chess GUI tool?

[AlexChess]

Just tested Igel 3.0.5 NNUE _bmi2_avx2_3_0_5.exe on my brother's AMD Ryzen 9 3900x using 24 cores (12 virtuals) : on starting position it calculates 21 mN/S against 7 mN/s macOS BS M1 Native 4CPUs (Windows 10 ARM64 3.5 mN/s, Ubuntu ARM64 20.04 4 mN/s (M1 fan completely silent while PC seems an helicopter )

I'm happy with my Mac mini M1 (and with 4 more CPUs availables I can use all other apps in multitasking at normal speed..) Since all other engines also use only 4CPUs the test is perfectly valid.

Regards, AlexChess

[voffka]

Just tested Igel 3.0.5 NNUE _bmi2_avx2_3_0_5.exe on my brother's AMD Ryzen 9 3900x

For AMD I think you need to try the popcount build instead (igel-x64_popcnt_avx2_3_0_5.exe) because AMDs are well known to be a bit slower on BMI2.

[AlexChess]

Just tested Igel 3.0.5 NNUE _bmi2_avx2_3_0_5.exe on my brother's AMD Ryzen 9 3900x

For AMD I think you need to try the popcount build instead (igel-x64_popcnt_avx2_3_0_5.exe) because AMDs are well known to be a bit slower on BMI2.

AMD Ryzen 9 3900x ATI + AMD RADEON PRO WX7100 + 32GB BM2 24 CPUs


AMD Ryzen 9 3900x + AMD RADEON PRO WX7100 + 32 GB POP 24 CPUs


SILICON M1 8 CPUs


I have restarted PC and engines.. now AMD gives me less speed on starting position (+- 16 mN /S)

Mac mini Silicon M1 isn't bad with 8 CPUs for 700$ against a 2000$ PC

[AlexChess]

Just tested Igel 3.0.5 NNUE _bmi2_avx2_3_0_5.exe on my brother's AMD Ryzen 9 3900x

For AMD I think you need to try the popcount build instead (igel-x64_popcnt_avx2_3_0_5.exe) because AMDs are well known to be a bit slower on BMI2.

IGEL 3.0.0 and IGEL 3.0.5 NNUE Windows 10 NEON ARM64 against almost all best chess engines

Arena 3.5.1 Windows 10 4CPUs 3.5 mN/s under Windows Parallels M1