SSDF Rating List 2007-11-03

[Tony Thomas]

Also, is there any way I could obtain the names of those 500 people?

Thanks, Kirill

What would be the point though Kirill?

May be to get some of them to help with CCRL? Or may be to see how many of them are CCC members.

[Spock]

Also, is there any way I could obtain the names of those 500 people?

Thanks, Kirill

What would be the point though Kirill?

May be to get some of them to help with CCRL? Or may be to see how many of them are CCC members.

Kirill did say

for a review page of computer chess testing studies that I am slowly making

[S.Taylor]

Hi Thoralf,

thanks for posting the latest SSDF list.

Really looking forward to results on your new hardware.

Well done to all SSDF testers.

Regards, Graham.

New hardware, not the most urgent thing if not all programs tested. We might never be able to compare to those that are not tested again with the new hardware.

[ArmoredGuns]

Ponder on only makes sense if engines play 2 threads vs 2 threads in a quad.

[icander]

Ponder on only makes sense if engines play 2 threads vs 2 threads in a quad.

We always test with two computers against each other. Then ponder off makes no sence.

[Spock]

Ponder on only makes sense if engines play 2 threads vs 2 threads in a quad.

We always test with two computers against each other. Then ponder off makes no sence.

Absolutely right - the SSDF way is the very best way to run testing, if you have the resource to do it. Each engine runs ponder on and has the entire machine to itself. So SSDF can do 4CPU vs 4CPU ponder on, if a single tester has two identical quad machines

[ozziejoe]

can anybody estimate how many rating points are gained by turning ponder on in SSDF (standard) time controls?

If doubling speed (or time to think) increases elo by 50, then ponder would presumably increase elo by something less than this (since you probably arn't quite doubling time, and there are the occasions where you do not get a ponder hit)

J

[Dann Corbit]

can anybody estimate how many rating points are gained by turning ponder on in SSDF (standard) time controls?

If doubling speed (or time to think) increases elo by 50, then ponder would presumably increase elo by something less than this (since you probably arn't quite doubling time, and there are the occasions where you do not get a ponder hit)

J

I have seen indications that you get better than the expected return because of jumping over a ply (I believe Dr. Hyatt has tested this extensively). I think that the ponder hit rate is pretty high among the strongest engines. The CCRL group has ponder hit statistics.