2008 WCCC - Unbalanced Hardware

[Ovyron]

Also, we all know that to date Zapper scales better than Rybka.

That changed with Rybka 3, she now scales as good as Zappa, or better.

[Zach Wegner]

That changed with Rybka 3, she now scales as good as Zappa, or better.

Proof?

[bob]

That changed with Rybka 3, she now scales as good as Zappa, or better.

Proof?

The author said so.

[Tony Thomas]

That changed with Rybka 3, she now scales as good as Zappa, or better.

Proof?

The author said so.

Bang! I programmed my program in a way that it only reports the nodes dependent on your hardware..

According to CCRL

Single CPU Zapper 64bit ----> 2955 (gained rating = 117)
4 CPU Zapper 64bit ----> 3072

Single CPU Fishka 64bit -----> 3163 (gained rating = 59 )
4 CPU Fishka 64bit -----> 3222


I am sorry bro, it just doesnt look good comparing pure rating..Then again there is probably diminishing returns as strength of the program increases.

[M ANSARI]

That changed with Rybka 3, she now scales as good as Zappa, or better.

Proof?

In my testing this most certainly seems to be true after hundreds of test games against Zappa Mexico II. On an Octa when you match Rybka 2.3.2a against Zappa, Zappa catches up dramatically with Rybka 2.3.2a and closes the gap where at 120 minutes per game using 8 cores at 4.8 Ghz it is almost at par with Rybka 2.3.2a. The same setup on a Quadcore using faster time controls shows that Rybka 2.3.2a has a large gap against ZM II. Offhand I think that Rybka 2.3.2a was scoring around 68% against ZM II at fast time controls ... and this gap dropped to around 51% with big hardware and LTC.

Now the same tests with Rybka 3 and ZM II show that the gap obtained with faster time controls and Quadcore hardware stays constant all the way to big hardware and LTC. As a matter of fact if I remember correctly in a 250 game semi LTC match on Octa (30min games), ZM II managed only 2 wins against R3. The score with R3 was around 85% at FTC and remarkably help up all the way to big hardware and LTC.

While not empirical proof, as far as I am concerned the data seems to show that while ZM II had a tremendous scaling advantage against R 2.3.2a ... that is not the case with R3.