On-line engine blitz tourney September

[Joost Buijs]

It is no problem to have "Dumb" on a RPI joining the tourney, if you wish I can make an account for it.

I will be pleased to have an account for it with the name rpiDumb.

I've added an account for rpiDumb and sent the initial password to your edax mail address.

[Joost Buijs]

Well, if an odd number of participants was a problem I'd have withdrawn Zappa.

Well, it was no problem at all, we shouldn't have added POS and Sargon. The point is that we need more engines in the middle of the pack to make the tournament interesting for everyone. At the moment the gap between the top and the weaker engines is too big.

I can add Micah if you like?

Hi Folkert, that would be fine.

You already have 2 accounts for it, micahgb and micahTR, if you dont' know the passwords anymore I have to send you new ones. Although the server is always up there is no need to let Mica run on it permanently, putting it online at the day of the tournament would be good enough.

Can't remember them yes. Please send me the micahgb one. Thanks in advance.

Oh never mind, found it!
It now runs on an old i7 (3d gen), 1 thread.

I just saw Micah is online.

I'm thinking about adding some old engines from the nineties too, the problem is that these old engines don't have UCI or CECP, so it will be quite a lot of work to get them running.

[Bjoern]

Well, my engine feligres just played against micagb and it is not that much better than POS. Micagb sacrifices a piece every 2nd move.

[flok]

Well, my engine feligres just played against micagb and it is not that much better than POS. Micagb sacrifices a piece every 2nd move.

Ah, then it could not find tune.dat.

Let's see if it works better now.

[dangi12012]

I wonder how RTX 4090 will perform with lc0.
Also Zen4 has some good upgrades relevant to stockfish

But joining with existing engines is frowned upon. Even tho it's fun to test hardware.
Adjacent engines like lc0+stockfish make double participating actually a really good hardware test.

[Ras]

But joining with existing engines is frowned upon. Even tho it's fun to test hardware.

That's because there is no effort involved in buying some hardware, and you can easily test that at home. Just determine the NPS rate of SF or LC0 on the new hardware and test that with asymmetric thinking time or node budget relative to the NPS of the hardware you'd like to compare to. The strength increase is simply the NPS scaling of the engine, after all. It's not exciting, it's just this year's hardware being faster than last year's, as always.

[dangi12012]

But joining with existing engines is frowned upon. Even tho it's fun to test hardware.

That's because there is no effort involved in buying some hardware, and you can easily test that at home. Just determine the NPS rate of SF or LC0 on the new hardware and test that with asymmetric thinking time or node budget relative to the NPS of the hardware you'd like to compare to. The strength increase is simply the NPS scaling of the engine, after all. It's not exciting, it's just this year's hardware being faster than last year's, as always.

There is much more going on than that. Did performance per dollar increase. Did performance per watt increase (maybe not with RTX 4080)
New instructions being avalable. Are they used by SF? Can the system hold up under stress.
What about DDR5 it has 2x32bit memory fetch. Can the reduced latency increase TT lookup performance etc..

For me its much more exciting than for you obviously but there is objectively going on much more than "hardware getting faster"
Also by knowin trends in the industry today - one can anticipate and write code that will be competitive tomorrow. IE bfloat16.

[Ras]

There is much more going on than that. Did performance per dollar increase. Did performance per watt increase (maybe not with RTX 4080)
New instructions being avalable. Are they used by SF? Can the system hold up under stress.
What about DDR5 it has 2x32bit memory fetch. Can the reduced latency increase TT lookup performance etc..

You don't need the monthly tournament to figure out any of that. By contrast, simply buying hardware and then running a top engine where you had no part in writing is exactly what your track record in the monthly would suggest: the wish to end up first without investing work. That's frowned upon for sure.

[dangi12012]

There is much more going on than that. Did performance per dollar increase. Did performance per watt increase (maybe not with RTX 4080)
New instructions being avalable. Are they used by SF? Can the system hold up under stress.
What about DDR5 it has 2x32bit memory fetch. Can the reduced latency increase TT lookup performance etc..

You don't need the monthly tournament to figure out any of that. By contrast, simply buying hardware and then running a top engine where you had no part in writing is exactly what your track record in the monthly would suggest: the wish to end up first without investing work. That's frowned upon for sure.

Why is it allowed then? For me its an interesting Hardware test. Failed 2 times - because like printers the hardware knows when it needs to perform and then it fails.
So tournament conditions are whats needed to test hardware.
As I have suggested MULTIPLE times now - just take SF and lc0 people out of the leaderboard.