Glaurung SMP 64 : A New King is born

[Daniel Charles]

Xeon 5365
Glaurung 2.01 SMP 64 bits
perfect 12.33.ctg
F10 GUI
512 Mo
3-4-5 egtb
Time 3 mn + 2 s ( >5 mn)
Gauntlets
200 games





I tested the New Glaurung SMP 64 on my Xeon vs 5 commercial and vs 5 top Free UCI engines

Thanks to Tord for the new Glaurung 64 bits (up to 8 cpu and maybe more ??)
It was was very fast (over 8000 kn/s on some games)
Of course this is not a Rybka 2.3.2a smp Killer . it lost vs DS11, Zappa Mexico,Naum ( the top commercial engines behind Rybka ) and beat Fritz 10 ( a good commercial engine )

Now begin the interesting test vs Top Free UCI engines, Glaurung won over all top Free UCI in this test (Including Free Rybka, Fruit 2.3.3J, Toga, Belka and Spike

At least on my big hardware (8 CPU ) A new king for Free UCI engine is born

[Oscar L]

Thanks for the testing on your big hardware


I have a question, SMP engines are supposed to use the 8 CPUs? In the case of Fritz, is it Fritz 10 or Deep Fritz 10.1?


Regards.

[Tord Romstad]

Hello Daniel,

Thanks for running this test! Frankly, I don't think narrowly beating other free programs with the help of a 8x advantage in speed is a particularly impressive achievement, but it is nice to see that my program runs well and is stable on 8 CPUs (I've never tested it on more than 2 myself).

By the way, it is possible that the performance of Glaurung with 8 threads could be improved somewhat by tuning the two UCI parameters "Minimum Split Depth" and "Maximum Number of Threads per Split Point". The default settings work well on 2 CPUs, but I am not sure they are optimal for 8.

The parameter "Minimum Split Depth" controls how far from the leaves the program is allowed to split the search tree between multiple CPUs. Splitting the search tree is an expensive operation, because a lot of information has to be copied to all threads that participate in the work. It is probably more expensive with a bigger number of CPUs; it is therefore possible that the minimum split depth should be slightly bigger on a computer with 8 CPUs.

"Maximum Number of Threads per Split Point" controls how many threads are allowed to participate in sharing the work at a single split point. It is possible that a maximum of 4 would be better than a maximum of 8 (especially in the late endgame, where the average number of moves is low), but it is difficult to know without testing.

Tord

[Rubinus]

Hi Tord.
Glaurung 2.0.1/64bit, run in Fritz? When will download avalible? Cca 15.12. leaved my new Super Engine tournament, 90+30 (Rybka,Zappa, Fritz, Shredder, Hiarcs, Junior, Sjeng, Loop, Spike, Fruit, Naum - and probably new Glaurung/else Toga).
Best wishes Pavel Háse

[Daniel Charles]

Jim did a 64 bits compile 2 days ago (for windows )

[Rubinus]

Jim did a 64 bits compile 2 days ago (for windows )

Thanks You.
Best regards Pavel Háse

[Tord Romstad]

Hi Pavel!

Glaurung 2.0.1/64bit, run in Fritz?

There appears to be a bug in recent versions of the Fritz GUI, which prevents it from working with engines with long "about" strings. Glaurung's string is about 180 characters long (I'm too lazy to check the exact length), which is apparently too much for Fritz. A simple workaround is to compile a binary with this string removed (Jim and others have done so, I think).

When will download avalible?

I don't know. I'm facing unexpected problems compiling a binary which will work with anything older than the most recent version of Mac OS X, and I have no idea how to solve the problems.

Tord

[Alessandro Scotti]

I don't know. I'm facing unexpected problems compiling a binary which will work with anything older than the most recent version of Mac OS X, and I have no idea how to solve the problems.

I have 10.4.11 and I'm not upgrading, I can compile the code if needed.

[Tord Romstad]

I don't know. I'm facing unexpected problems compiling a binary which will work with anything older than the most recent version of Mac OS X, and I have no idea how to solve the problems.

I have 10.4.11 and I'm not upgrading, I can compile the code if needed.

Thanks for the offer -- but do you have the Intel compiler? Compiling with GCC slows down my engine by more than 50% on Intel Macs, which I consider unacceptable.

I'm afraid I'll just have to buy an external disk and install 10.4 on it.

Tord

[Alessandro Scotti]

Thanks for the offer -- but do you have the Intel compiler? Compiling with GCC slows down my engine by more than 50% on Intel Macs, which I consider unacceptable.

Nope.

A 50% slowdown is really terrible though, I would be surprised if some of the compiler gurus out there can't improve that!