LCzero network

[Laskos]

For now 31970 is the best

31970- 11248: +9 -5 =186
Hardware used: 2080Ti and 6men TBs
Time control: 2min + 1sec
Book: Private (about 30 = positions)

31970- 11248: +39 - 27 =139
GPU 2080Ti and 6men TBs
5000 nodes/move

https://groups.google.com/forum/#!forum/lczero

Have they been tried aganst regular engines? Last time I checked, 2 days ago, test30 was still some 40 Elo points behind test10 and advancing very, very slowly, on the average of less than 0.2 Elo points per net on 700 nets span. Against SF8.
Also, don't use fixed nodes when testing strength.

It seems you did not read the thread? Test B is discredited - games for test A are not available...

Yes, on my phone, just tried to reply to that a bit strange results and some not very correct methodology. Just saw now that the second test is bogus. I have doubts about the first one too, theu should check against some regular engines.

[mehmet karaman]

But how about the other 98 games. I don't think the games are scam

[mehmet karaman]

31954 vs 11248: +5 =93 -2

GPU 2080Ti and used 6men TBs. Time control was 2min + 1sec
Private opening book.

And this is his old test result

[mehmet karaman]

"There is a debate in discord about these results. On long TC (or high nodes count) test30 seems to be better. However, most of the tests are done at very short TC (or low nodes count) where it seems test10 is superior. So it seems test30 scales better… However, most of the elo tests are performed in bullet conditions."

From the Lc0 forum

[Guenther]

But how about the other 98 games. I don't think the games are scam

1. the rest is now 96
2. there is no doubt about the rest, otherwise he had noticed his error(?) and restarted the whole match
a quick check of the next 4 game pairs confirms this...

you should believe facts

[Laskos]

"There is a debate in discord about these results. On long TC (or high nodes count) test30 seems to be better. However, most of the tests are done at very short TC (or low nodes count) where it seems test10 is superior. So it seems test30 scales better… However, most of the elo tests are performed in bullet conditions."

From the Lc0 forum

How long TC on what hardware?

[jp]

Also, don't use fixed nodes when testing strength.

What does fixed nodes do to the testing?

[mehmet karaman]

"There is a debate in discord about these results. On long TC (or high nodes count) test30 seems to be better. However, most of the tests are done at very short TC (or low nodes count) where it seems test10 is superior. So it seems test30 scales better… However, most of the elo tests are performed in bullet conditions."

From the Lc0 forum

How long TC on what hardware?

Unfortunately there isn't much information about this

[Laskos]

Also, don't use fixed nodes when testing strength.

What does fixed nodes do to the testing?

Nets vary in speed, even among test30 nets there is quite a variarion. Strength is measured at same time control, not same number of nodes. Same time control could mean different number of nodes.

[jp]

Nets vary in speed, even among test30 nets there is quite a variarion. Strength is measured at same time control, not same number of nodes. Same time control could mean different number of nodes.

Interesting. Do you know why? I guess most assume same-size NN = same speed.