On-line engine blitz tourney December

[Modern Times]

I would like to join this tournament as my engine is still new and I would love to get an elo estimate of it.

You will not get a decent Elo estimate from 9 games.

[GyatsoYT]

I would like to join this tournament as my engine is still new and I would love to get an elo estimate of it.

You will not get a decent Elo estimate from 9 games.

True but I would get to know how it performs against other developing engines who might be in the similar range or elo.

[Joost Buijs]

I would like to join this tournament as my engine is still new and I would love to get an elo estimate of it. But the timing is not possible for me it's 12:30 AM + I won't be in my home on that day + I have an exam the day after so there are a lot of constraints.

You also need an account, I can add an account for GyatsoYT, and I still have some spare cores, if you wish I can add GyatsoYT tomorrow morning and let it play some games to check if it works without problems and to give it a more established rating before entering the tournament. I can PM you the details of the account later.

Edit: I see your engine is called Gyatso, so I will make an account for that.
Please PM me your email address, then I can add you to the mailing list.

[Joost Buijs]

I've already added Gyatso, and it's currently playing at the server.
At the moment it plays with 4 threads and avx2, during the tournament I can let it play with 12 threads and avx-512.
It all seems to work fine.

[GyatsoYT]

I've already added Gyatso, and it's currently playing at the server.
At the moment it plays with 4 threads and avx2, during the tournament I can let it play with 12 threads and avx-512.
It all seems to work fine.

Really appreciate that mate, And you can check your PM for the email.

[Joost Buijs]

Gyatso played some games at the server, the games look reasonable, no strange things, but I have the impression that it overlooks mate threats, and that it sometimes looses equal positions because it didn't see checkmate coming.

After 10 games on the server it's Elo rating is 1943, I don't know how much Elo this would be at CCRL, probably over 2300 because the Elo ratings at CCRL are way to high compared to human ratings.

Tomorrow it will play the tournament on better hardware, we'll see what happens then.

[GyatsoYT]

Gyatso played some games at the server, the games look reasonable, no strange things, but I have the impression that it overlooks mate threats, and that it sometimes looses equal positions because it didn't see checkmate coming.

After 10 games on the server it's Elo rating is 1943, I don't know how much Elo this would be at CCRL, probably over 2300 because the Elo ratings at CCRL are way to high compared to human ratings.

Tomorrow it will play the tournament on better hardware, we'll see what happens then.

Thanks for the useful insights I'll use this to find issues in the code.

Tbh let's also wait for CCRL to test it to see it's rating and performance against different engines.

And yes let's see how it performs in the tournament with better hardware.

[Modern Times]

Juggernaut is now playing 100% reliably. For years I've used the -xreuse parameter when starting winboard on a recommendation from someone, can't recall who. The description for this is:

WinBoard kills the engine process after each game and starts a new one. This ensures a clean slate.

However initially for Juggernaut I was ending up with multiple instances in memory eating CPU resource. It appears that Juggernaut wasn't in fact being killed successfully every time. After I removed that parameter, all is well.

I've only ever though that this parameter could either potentially be useful, or of no benefit, but never harmful as in this case.
Perhaps I won't use it going forward. I'll ask on the ICS to see if anyone else uses it.

[Joost Buijs]

Thanks for the useful insights I'll use this to find issues in the code.

Tbh let's also wait for CCRL to test it to see it's rating and performance against different engines.

And yes let's see how it performs in the tournament with better hardware.

Gyatso is running on an AMD with 4 Zen5 cores and 8 Zen5C cores. I tried both AVX2 and AVX-512, AVX2 is a little bit faster, this is what I see with my own engines as well.

When I run Gyatso from the command-line with 1 thread or 12 threads I see a nice speedup of a factor 11,4. But when it runs under icsdroneng the speedup is much lower, I don't know what causes this. The Windows thread-scheduler has difficulties with AMD-SMT, probably you have to add a thread-affinity-mask to circumvent this. This is a well known problem, somehow Microsoft doesn't want to fix this.

[GyatsoYT]

Thanks for the useful insights I'll use this to find issues in the code.

Tbh let's also wait for CCRL to test it to see it's rating and performance against different engines.

And yes let's see how it performs in the tournament with better hardware.

Gyatso is running on an AMD with 4 Zen5 cores and 8 Zen5C cores. I tried both AVX2 and AVX-512, AVX2 is a little bit faster, this is what I see with my own engines as well.

When I run Gyatso from the command-line with 1 thread or 12 threads I see a nice speedup of a factor 11,4. But when it runs under icsdroneng the speedup is much lower, I don't know what causes this. The Windows thread-scheduler has difficulties with AMD-SMT, probably you have to add a thread-affinity-mask to circumvent this. This is a well known problem, somehow Microsoft doesn't want to fix this.

Hey there thanks for letting me know about the issue with threading.


Also if you don't mind what was the nps you got on single thread while searching like a depth of 20 vs multithreaded running on 12 cores / 24 threads if your cpu supports it?