Discussion about chess-playing software (engines, hosts, opening books, platforms, etc...)
-
Martin Thoresen
- Posts: 386
- Joined: Thu Jun 10, 2010 5:27 am
Post
by Martin Thoresen » Tue Mar 01, 2011 4:29 pm
Very nice, thank you.
Expect Gull 1.2 to play in TCEC C which starts in ~12 days.
edit: It says that it supports hash up to 2048 MB but when this is selected, the engine crashes (64-bit version).
Best,
Martin
-
ThinkingALot
- Posts: 144
- Joined: Sun Jun 13, 2010 7:32 am
-
Contact:
Post
by ThinkingALot » Tue Mar 01, 2011 4:39 pm
Martin Thoresen wrote:edit: It says that it supports hash up to 2048 MB but when this is selected, the engine crashes (64-bit version).
Thanks. I'll load a fixed version to sourceforge soon. Thought malloc's gonna work with 2Gbs
.
Edit: uploaded.
-
ThinkingALot
- Posts: 144
- Joined: Sun Jun 13, 2010 7:32 am
-
Contact:
Post
by ThinkingALot » Tue Mar 01, 2011 5:02 pm
This compile works only under Windows Vista/7. I'll upload a fixed version soon.
-
Martin Thoresen
- Posts: 386
- Joined: Thu Jun 10, 2010 5:27 am
Post
by Martin Thoresen » Tue Mar 01, 2011 5:18 pm
ThinkingALot wrote:This compile works only under Windows Vista/7. I'll upload a fixed version soon.
Ah okay, so you removed the 2 GB option. Was a bit confused.
Thanks.
-
Matthias Gemuh
- Posts: 295
- Joined: Wed Jun 09, 2010 2:48 pm
-
Contact:
Post
by Matthias Gemuh » Thu Mar 03, 2011 12:45 am
ThinkingALot,
it is a bad idea to automatically load LargePages on the computers of people who do not want them.
Matthias.
-
ThinkingALot
- Posts: 144
- Joined: Sun Jun 13, 2010 7:32 am
-
Contact:
Post
by ThinkingALot » Thu Mar 03, 2011 7:23 am
Matthias Gemuh wrote:
ThinkingALot,
it is a bad idea to automatically load LargePages on the computers of people who do not want them.
Matthias.
Why would one want Gull not to load large pages? This feature just provides tiny performance boost, nothing more. The same thing with automatic detection of hardware popcnt.
-
ThinkingALot
- Posts: 144
- Joined: Sun Jun 13, 2010 7:32 am
-
Contact:
Post
by ThinkingALot » Thu Mar 03, 2011 8:19 am
Of course if you'd like to have a non large page version for testing, just contact me via e-mail. Or you can just compile it yourself.
But from the user's point of view it's much better to have every useful feature turned on by default.
-
Odeus37
- Posts: 43
- Joined: Mon Jun 14, 2010 5:38 pm
Post
by Odeus37 » Thu Mar 03, 2011 9:49 am
I can't see too why it should be a probleme to automatically load LargePages.
When I want those, I start my GUI with admin rights, and when I don't want those, I start my GUI without admin rights... (I have UAC on).
-
ThinkingALot
- Posts: 144
- Joined: Sun Jun 13, 2010 7:32 am
-
Contact:
Post
by ThinkingALot » Fri Mar 04, 2011 11:41 am
Gull 1.2 appears to be much weaker than I expected:(. Stronger than 1.1 but only by a small margin.
That's quite surprising since I carried out some tests on ultra short "time controls" (with fixed number of nodes per move) and obtained a substantial gain:
Uploaded with
ImageShack.us
So there's probably a bug hidden somewhere.
-
IWB
- Posts: 195
- Joined: Thu Jun 10, 2010 4:10 pm
Post
by IWB » Fri Mar 04, 2011 1:00 pm
Hello
ThinkingALot wrote:Gull 1.2 appears to be much weaker than I expected:(. Stronger than 1.1 but only by a small margin.
That's quite surprising since I carried out some tests on ultra short "time controls" (with fixed number of nodes per move) and obtained a substantial gain:
Uploaded with
ImageShack.us
So there's probably a bug hidden somewhere.
But even your Elos are going down with more Nodes ...
I am still not sure if this ultra short testing is working. You/Gull 1.2 seems to be an example for overestimating the outcome and I remember the Stockfish team once to expect nearly nothing and it was a big jump in reality (1.6 to 1.7). Somehow the results seem not to be reliable with this Nodes/move method.
BYe
Ingo
