How many of you have ordered the new Naum?

[Uri Blass]

You took your data from the ccrl but even in the ccrl list I see the following:

Zap!Chess Zanzibar 64-bit 4CPU 3051
Zap!Chess Zanzibar 64-bit 2CPU 3021
Zap!Chess Zanzibar 64-bit 2903

Zap!Chess Paderborn 64-bit 4CPU 2953
Zap!Chess Paderborn 64-bit 2CPU 2909
Zap!Chess Paderborn 64-bit 2800

Zap!Chess Zanzibar 32-bit 2CPU 2951
Zap!Chess Zanzibar 32-bit 2889

Zap!Chess Paderborn 32-bit 4CPU 2933
Zap!Chess Paderborn 32-bit 2781

For Hiarcs:

Hiarcs 11.1 4CPU 2985
Hiarcs 11.1 2CPU 2944
Hiarcs 11.1 2893

Hiarcs 11 4CPU 2967
Hiarcs 11 2CPU 2926
Hiarcs 11 2876

For Naum

Naum 2.1 64-bit 4CPU 2967
Naum 2.1 64-bit 2CPU 2930
Naum 2.1 32-bit 2845

For Loop

Loop M1-T 64-bit 2CPU 2948
Loop M1-T 64-bit 4CPU 2948
Loop M1-T 64-bit 2920

For Deep Fritz10

Deep Fritz 10 2CPU 2953 +57 −56 61.2% −71.9 45.7% 94
78.0%
1 Deep Fritz 10 4CPU 2928 +21 −21 49.1% +5.6 39.8% 736
99.9%
Fritz 10 2881

For old Glaurung

Glaurung 1.2.1 64-bit 4CPU 2858
Glaurung 1.2.1 64-bit 2CPU 2818
Glaurung 1.2.1 64-bit 2769

Results for most engines is better than rybka.

[naum]

Naum 2.2 seems like it has tremendous potential. I say that because I noticed that its MP implementation is not good. This would make it seem like there is a lot of room for improvement with simply improving MP code.

Hi,

I don't know how you came to that conclusion. I don't think you even have Naum 2.2, since you are not on my customer list.

Naum's MP implementation is as good as in other top engines. Only Zanzibar has it better.

I just had some tests done on the 8 CPU machine and it seems to scale pretty well.
For one test position I got:
quad Q6600 at 3.5GHz -> 6MNPS
octo Xeon 2.66GHz -> 8MNPS which translates to 10.5MNPS at 3.5GHz

This translates to 75% improvement going from quad to octo, but I am not sure how Q6600 and Xeons compare.

Regards,
Alex

[Uri Blass]

Naum 2.2 seems like it has tremendous potential. I say that because I noticed that its MP implementation is not good. This would make it seem like there is a lot of room for improvement with simply improving MP code.

Hi,

I don't know how you came to that conclusion. I don't think you even have Naum 2.2, since you are not on my customer list.

Naum's MP implementation is as good as in other top engines. Only Zanzibar has it better.

I just had some tests done on the 8 CPU machine and it seems to scale pretty well.
For one test position I got:
quad Q6600 at 3.5GHz -> 6MNPS
octo Xeon 2.66GHz -> 8MNPS which translates to 10.5MNPS at 3.5GHz

This translates to 75% improvement going from quad to octo, but I am not sure how Q6600 and Xeons compare.

Regards,
Alex

I claim nothing about naum MP implementation but you cannot get conclusions based on nodes per seconds because the question is if different processors search efficiently.

The only way to know is based on rating and it is possible to have an estimate based on speed improvement in getting fixed depth assuming the search algotithm is not significantly different.

Uri

[naum]

Hi Uri,

As far as I know the scaling is measured by NPS.
I don't claim octo version is 75% better then the quad.
I am just saying that NPS is around 75% higher, and as you can see from my post I also mentioned that CPUs are different, so this is just a rough estimate.

Regards,
Alex

[Guetti]

Hi Alex,

are there Mac or Linux versions of Naum available? If not, are there any plans for it?

[naum]

Hi Andreas,
Demand for Mac is zero, and for Linux version isn't much higher. Mac version will not happen, but if I get a lot of free time I may create a Linux version.
Alex

[Oscar L]

Hi Alex.

There should be something else besides NPS?

I remember to have asked W. Schüle (CETG) about this on F10. NPS were higher on 4CPU compared with 2CPU, but the results are worse with 4CPU. Or perhaps it is a specific problem with f10?

Regards

[Uri Blass]

Hi Uri,

As far as I know the scaling is measured by NPS.
I don't claim octo version is 75% better then the quad.
I am just saying that NPS is around 75% higher, and as you can see from my post I also mentioned that CPUs are different, so this is just a rough estimate.

Regards,
Alex

Maybe I simply did not understant your post:
I understood:
"This translates to 75% improvement going from quad to octo" as something that is equivalent to 75% speed improvement.

A program may search more nodes and play weaker and the only real test is simply elo rating and it seems based on rating list that naum play better with more processors(at least 4 seem to be better than 2 and 2 seem to be better than 1)

I do not know about serious test with 8 processors.

Uri

[Guetti]

Hi Andreas,
Demand for Mac is zero, and for Linux version isn't much higher. Mac version will not happen, but if I get a lot of free time I may create a Linux version.
Alex

Thanks for your reply.

Indeed there is not much to be earned on the Mac and Linux market. Even more since Linux users are themed by this "free software" doctrine and might be more reluctant to spend money on software.

On the other hand, since Rybka is absent it is easier to create the "strongest chess program" running natively on Mac and Linux.

[Tony Thomas]

I claim nothing about naum MP implementation but you cannot get conclusions based on nodes per seconds because the question is if different processors search efficiently.

The only way to know is based on rating and it is possible to have an estimate based on speed improvement in getting fixed depth assuming the search algotithm is not significantly different.

Uri

I dont think Alex was talking to you, he was talking to Ansari.