Fritz 15 is unsuitable for analysis.

[Guenther]

Fritz 15 is not able to under promote with a bishop so it's unsuitable for analysis.

Your prejudiced opinion about Vasik Rajlich / Rybka disqualifies you from uttering such statements !

Ernest, it is completely useless to try to have a conversation with that Distel.
He is on thousand years crusade against Rajlich and has very strange views
about everything else anyways...
He just used this lack of underpromotion as a trojan horse for his bashing and some uninformed souls here even fell for it.
I am quite sure he will be on a forced hiatus soon with no chance to spread his poison again for a while as it happened already multiple times.

[BBauer]

@Allan:
For anaysis purposes I have modified SF a little.
If the mobility is less/equal a number I check for it.
For example if I set PieceMob_GT=12
NullMove and LMR are controlled by PieceMob_GT.
This way I get:

Code: Select all

FEN: 8/1P4K1/p7/k2p4/Pp1P4/1P6/6p1/6Bq w - - 0 1

St-mod-15-12-20:
 1/1	00:00	 37	12k	-2,36	b7-b8Q Qh1xg1
 2/2	00:00	 166	55k	+10,22	b7-b8Q Qh1-h7+ Kg7xh7
 3/3	00:00	 410	137k	+10,22	b7-b8Q Qh1-h7+ Kg7xh7
 4/5	00:00	 781	260k	+0,45	b7-b8Q Qh1-h7+ Kg7-f6 Qh7-h4+ Kf6-g6
 5/7	00:00	 2k	571k	+0,20	b7-b8Q Qh1-h7+ Kg7-f6 Qh7-h4+ Kf6-f5 Qh4-h3+ Kf5-f4
 6/8	00:00	 2k	561k	+0,20	b7-b8Q Qh1-h7+ Kg7-f6 Qh7-h4+ Kf6-f5 Qh4-h3+ Kf5-f4 Qh3xb3
 7/10	00:00	 7k	1.166k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 8/10	00:00	 8k	1.281k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 9/10	00:00	 11k	1.608k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 10/10	00:00	 13k	1.489k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 11/10	00:00	 18k	1.632k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 12/10	00:00	 26k	1.850k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 13/10	00:00	 26k	1.754k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 14/10	00:00	 31k	1.808k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 15/10	00:00	 48k	1.779k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 16/10	00:00	 79k	1.707k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 17/10	00:00	 140k	1.708k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 18/10	00:00	 140k	1.710k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 19/10	00:00	 231k	1.894k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 20/10	00:00	 293k	1.979k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 21/10	00:00	 500k	2.214k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 22/10	00:00	 653k	2.275k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 23/10	00:00	 688k	2.250k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 24/10	00:00	 2.444k	2.497k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 25/10	00:01	 3.153k	2.507k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 26/10	00:02	 6.289k	2.496k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 27/10	00:03	 8.082k	2.507k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 28/10	00:04	 11.856k	2.463k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 29/10	00:07	 19.350k	2.452k	 0,00	b7-b8Q Qh1-h6+ Kg7xh6
 30/29+	00:12	 30.898k	2.440k	+0,07	b7-b8B
 30/29+	00:12	 30.925k	2.441k	+0,14	b7-b8B
 30/42+	00:28	 81.293k	2.835k	+74,36	b7-b8B
 30/42-	00:29	 83.412k	2.859k	+61,69	b7-b8B Ka5-b6
 30/42	00:29	 84.219k	2.868k	+60,91	b7-b8B Ka5-b6 Bb8-h2 Kb6-c6 Kg7-f7 Kc6-d7 Kf7-f6 Kd7-e8 Kf6-e6 Ke8-d8 Ke6xd5 Kd8-d7 Kd5-c5 Kd7-e6 d4-d5+ Ke6-f5 d5-d6 Kf5-e6 Kc5-b6 Qh1xg1+ Bh2xg1 Ke6-f7 d6-d7 Kf7-e7 Kb6-c7 Ke7-f6 d7-d8Q+ Kf6-g6 Qd8-g8+ Kg6-f6 Qg8xg2 a6-a5 Kc7-c6
 31/42+	00:29	 84.245k	2.868k	+60,98	b7-b8B
 31/42+	00:29	 84.295k	2.868k	+61,05	b7-b8B
 31/42+	00:29	 85.908k	2.879k	+91,37	b7-b8B
 31/42	00:29	 85.993k	2.879k	+92,15	b7-b8B Ka5-b6 Bb8-h2 Kb6-b7 Kg7-f7 Kb7-c8 Kf7-e8 Kc8-b7 Ke8-d7 a6-a5 Kd7-e6 Kb7-c6 Ke6-e7 Kc6-b6 Ke7-d6 Kb6-b7 Kd6xd5 Kb7-c8 Kd5-d6 Kc8-d8 Kd6-c6 Qh1xh2 Bg1xh2 Kd8-e7 d4-d5 Ke7-f7 d5-d6 Kf7-g6 Bh2-g1 Kg6-f6 d6-d7
 32/42+	00:29	 86.181k	2.879k	+92,22	b7-b8B
 32/42+	00:29	 86.211k	2.879k	+92,62	b7-b8B
 32/42+	00:40	 126.095k	3.132k	+121,98	b7-b8B
 32/42	00:49	 144.344k	2.925k	+123,34	b7-b8B Ka5-b6 Bb8-h2 Kb6-c6 Kg7-f6 Kc6-d7 Kf6-e5 Kd7-c6 Ke5-e6 a6-a5 Ke6-e5 Kc6-c7 Ke5xd5+ Kc7-d7 Kd5-c5 Kd7-e6 d4-d5+ Ke6-f6 Kc5-c6 Kf6-g7 d5-d6 Kg7-f7 d6-d7 Kf7-e7 Kc6-c7 Ke7-f6 d7-d8Q+ Kf6-f5 Kc7-c6 Qh1xh2 Bg1xh2 Kf5-g4 Qd8xa5 Kg4-f3 Qa5-f5+ Kf3-e3 Bh2-g1+ Ke3-d2 Qf5-d5+ Kd2-c3 Qd5-c4+ Kc3-b2 Qc4xb4
 33/42	00:53	 151.870k	2.865k	+123,36	b7-b8B Ka5-b6 Bb8-h2 Kb6-c6 Kg7-f6 Kc6-d7 Kf6-e5 Kd7-c6 Ke5-e6 a6-a5 Ke6-e5 Kc6-c7 Ke5xd5+ Kc7-d7 Kd5-c5 Kd7-e6 d4-d5+ Ke6-f6 Kc5-c6 Kf6-g7 d5-d6 Kg7-f7 d6-d7 Kf7-e7 Kc6-c7 Ke7-e6 d7-d8Q Qh1xh2+ Bg1xh2 Ke6-f7 Kc7-c6 g2-g1N Bh2xg1 Kf7-g6 Qd8xa5
 34/42	00:55	 158.514k	2.862k	+123,37	b7-b8B Ka5-b6 Bb8-h2 Kb6-c6 Kg7-f6 Kc6-d7 Kf6-e5 Kd7-c6 Ke5-e6 a6-a5 Ke6-e5 Kc6-d7 Ke5xd5 Kd7-e7 Kd5-c6 Ke7-f7 d4-d5 Kf7-g7 d5-d6 Kg7-f7 d6-d7 Kf7-e7 Kc6-c7 Ke7-e6 d7-d8Q Qh1xh2+ Bg1xh2 Ke6-f7 Kc7-c6 Kf7-g6 Qd8-g8+ Kg6-f5 Qg8xg2 Kf5-f6 Kc6-b6
 35/42	00:57	 166.738k	2.895k	+123,38	b7-b8B Ka5-b6 Bb8-h2 Kb6-c6 Kg7-f7 Kc6-d7 Kf7-f6 Kd7-e8 Kf6-e6 Ke8-d8 Ke6xd5 Kd8-e7 Kd5-c6 Ke7-f8 d4-d5 Kf8-g8 d5-d6 Kg8-f7 d6-d7 Kf7-e7 Kc6-c7 Ke7-f7 d7-d8Q Qh1xg1 Bh2xg1 Kf7-g6 Qd8-g8+ Kg6-f5 Qg8xg2 Kf5-f4 Qg2-f1+ Kf4-g3 Qf1xa6 Kg3-h3 Qa6-c8+ Kh3-h4 Qc8-g4+ Kh4xg4
 36/42	01:01	 177.448k	2.895k	+123,39	b7-b8B Ka5-b6 Bb8-h2 Kb6-c6 Kg7-f6 Kc6-d7 Kf6-e5 Kd7-e7 Ke5xd5 Ke7-f6 Kd5-c6 Kf6-f5 d4-d5 Kf5-g5 d5-d6 Kg5-f6 d6-d7 Kf6-e7 Kc6-c7 Ke7-f7 d7-d8Q Qh1xg1 Bh2xg1 Kf7-g6 Qd8-g8+ Kg6-f5 Qg8xg2 Kf5-f4 Qg2-f1+ Kf4-g3 Qf1xa6 Kg3-h3 Qa6-a5 Kh3-g2 Qa5xb4
 37/42	01:05	 187.815k	2.871k	+123,40	b7-b8B Ka5-b6 Bb8-h2 Kb6-c6 Kg7-f6 Kc6-d7 Kf6-e5 Kd7-e7 Ke5xd5 Ke7-f6 Kd5-c6 Kf6-f5 d4-d5 Kf5-g5 d5-d6 Kg5-f6 d6-d7 Kf6-e7 Kc6-c7 Ke7-f7 d7-d8Q Qh1xh2+ Bg1xh2 Kf7-g6 Qd8-g8+ Kg6-f5 Qg8xg2 Kf5-f6 Qg2-g6+ Kf6xg6 Bh2-d6 Kg6-f5 Bd6xb4
 38/42	01:21	 220.902k	2.720k	+123,41	b7-b8B Ka5-b6 Bb8-h2 Kb6-c6 Kg7-f6 Kc6-d7 Kf6-e5 Kd7-e7 Ke5xd5 Ke7-f6 Kd5-c6 Kf6-f5 d4-d5 Kf5-g5 d5-d6 Kg5-f6 d6-d7 Kf6-e7 Kc6-c7 Ke7-f7 d7-d8Q Qh1xh2+ Bg1xh2 g2-g1N Bh2xg1 Kf7-g6 Qd8-g5+ Kg6xg5 Bg1-c5 Kg5-f5 Bc5xb4
 39/44	01:28	 236.178k	2.665k	+123,41	b7-b8B Ka5-b6 Bb8-h2 Kb6-c6 Kg7-f6 Kc6-d7 Kf6-e5 Kd7-e7 Ke5xd5 Ke7-f6 Kd5-c6 Kf6-f5 d4-d5 Kf5-g4 d5-d6 Kg4-h3 d6-d7 Qh1xh2 Bg1xh2 Kh3xh2 d7-d8Q g2-g1Q Qd8-h4+ Kh2-g2 Qh4-g4+ Kg2-f2 Qg4xg1+ Kf2-e2 Qg1-g4+ Ke2-e3 Qg4xb4
 40/48+	01:51	 297.000k	2.660k	+123,48	b7-b8B
 40/48+	02:07	 338.772k	2.655k	+M62	b7-b8B
 40/48+	02:27	 407.553k	2.761k	+M30	b7-b8B
 40/50	02:44	 447.416k	2.719k	+M25	b7-b8B Ka5-b6 Bb8-h2 Kb6-c6 Kg7-f6 Kc6-d7 Kf6-f7 a6-a5 Kf7-f6 Kd7-e8 Kf6-e6 Ke8-f8 Ke6xd5 Kf8-f7 Kd5-c6 Kf7-f6 d4-d5 Kf6-f5 d5-d6 Kf5-g4 d6-d7 Kg4-f3 d7-d8Q Kf3-e2 Qd8-e7+ Ke2-d1 Qe7-d7+ Kd1-c2 Qd7-d5 Qh1xh2 Bg1xh2 Kc2-c1 Qd5xg2 Kc1-d1 Qg2-f1+ Kd1-c2 Kc6-d5 Kc2-d2 Bh2-f4+ Kd2-c2 Qf1-e2+ Kc2xb3 Qe2-d1+ Kb3-b2 Bf4-e5+ Kb2-a3 Kd5-c4
 41/50	02:53	 459.626k	2.657k	+M23	b7-b8B Ka5-b6 Bb8-h2 Kb6-c6 Kg7-f6 Kc6-d7 Kf6-f7 a6-a5 Kf7-f6 Kd7-e8 Kf6-e6 Ke8-f8 Ke6xd5 Kf8-f7 Kd5-c6 Kf7-f6 d4-d5 Kf6-f5 d5-d6 Kf5-g4 d6-d7 Kg4-f3 d7-d8Q Kf3-e2 Qd8-d5 Qh1xh2 Bg1xh2 g2-g1N Bh2xg1 Ke2-f1 Qd5-g5 Kf1-e2 Qg5-e3+ Ke2-d1 Qe3-d3+ Kd1-c1 Bg1-e3+ Kc1-b2 Qd3-d1 Kb2-a3 Be3-c1+ Ka3-a2 Qd1-c2+ Ka2-a1 Qc2-b2+
 42/50	03:00	 473.120k	2.628k	+M23	b7-b8B Ka5-b6 Bb8-h2 Kb6-c6 Kg7-f6 Kc6-d7 Kf6-f7 a6-a5 Kf7-f6 Qh1xh2 Bg1xh2 g2-g1N Bh2xg1 Kd7-d6 Kf6-f7 Kd6-d7 Bg1-h2 Kd7-d8 Kf7-e6 Kd8-e8 Bh2-d6 Ke8-d8 Ke6xd5 Kd8-d7 Kd5-e5 Kd7-c6 Ke5-e6 Kc6-b6 Bd6-e5 Kb6-b7 d4-d5 Kb7-c8 d5-d6 Kc8-d8 Be5-f6+ Kd8-c8 d6-d7+ Kc8-b8 d7-d8Q+ Kb8-a7 Qd8-c7+ Ka7-a8 Qc7-c8+ Ka8-a7 Bf6-d4+
 43/50	03:08	 491.896k	2.609k	+M22	b7-b8B Ka5-b6 Bb8-h2 Kb6-c6 Kg7-f6 Kc6-d7 Kf6-f7 a6-a5 Kf7-f6 Qh1xh2 Bg1xh2 g2-g1N Bh2xg1 Kd7-d6 Kf6-f7 Kd6-d7 Bg1-h2 Kd7-d8 Kf7-e6 Kd8-e8 Bh2-e5 Ke8-f8 Be5-f6 Kf8-e8 Ke6xd5 Ke8-f7 Kd5-e5 Kf7-f8 d4-d5 Kf8-g8 d5-d6 Kg8-f7 d6-d7 Kf7-g6 Ke5-e6 Kg6-h7 d7-d8Q Kh7-h6 Qd8-h8+ Kh6-g6 Qh8-g7+ Kg6-h5 Qg7-g5+
 44/50	04:06	 626.976k	2.541k	+M21	b7-b8B Ka5-b6 Bb8-h2 Kb6-c6 Kg7-f6 Kc6-d7 Kf6-f7 a6-a5 Kf7-f6 Kd7-e8 Kf6-e6 Ke8-f8 Ke6xd5 Kf8-f7 Kd5-c6 Kf7-f6 d4-d5 Kf6-f5 d5-d6 Kf5-g4 d6-d7 Qh1xh2 Bg1xh2 g2-g1N Bh2xg1 Kg4-f4 d7-d8Q Kf4-f3 Qd8-h4 Kf3-e2 Qh4-c4+ Ke2-f3 Bg1-c5 Kf3-g2 Qc4-f4 Kg2-h3 Qf4-f3+ Kh3-h2 Qf3-g4 Kh2-h1 Qg4-h3+

By the way, I think SF developers should do something like kommodo and give some controll to the user. Especialy for NullMove and LMR.
Kind regards
Bernhard

[h1a8]

I agree with the assessment:
Fritz 15 is utterly unsuitable for analysis.

Great for game play, perhaps, but an engine that does not under-promote should not be used for analysis because it is defective.

As far as I know, all (top) engines fail on some positions (so they are all unsuitable for analysis?). At least with the underpromotion problem the user knows what (not) to expect.

For analysis, if the engine is designed to fail, then I do not want to use it.
Accidental failures are another matter.

Other people may feel that use for analysis is fine.
I only stated my personal opinion.
I do not want to use an engine that excludes legal moves for analysis.

Now I understand. You must use at least 10 different engines for analysis. Otherwise, why would you purposely want to use a vastly inferior engine for analysis even if it did have bishop promotions? I just use the 2 strongest engines.

If this is the case then your point is sound. I'm looking to buy Fritz 15 for the features. I will not be using the included engine for analysis, even if it didn't have this problem. I might use it for friend mode though. But that's it.


P.S. I think that not including bishop promotions most likely significantly increases elo. It simply erases an entire branch of analysis (millions of positions) and uses those resources for further depth of the position. The only problem is that it fails when an actual bishop promotion is best. This is 1 out of millions chance or less. So this trick is primarily for elo gain in engine vs. engine games. Komodo and stockfish can do the same trick for increases to their playing strength but they are strong enough.

Someone should investigate this. Modify stockfish where bishop promotion is erased. See how much elo it gains. I'm guessing at least 20.

[Dann Corbit]

I agree with the assessment:
Fritz 15 is utterly unsuitable for analysis.

Great for game play, perhaps, but an engine that does not under-promote should not be used for analysis because it is defective.

As far as I know, all (top) engines fail on some positions (so they are all unsuitable for analysis?). At least with the underpromotion problem the user knows what (not) to expect.

For analysis, if the engine is designed to fail, then I do not want to use it.
Accidental failures are another matter.

Other people may feel that use for analysis is fine.
I only stated my personal opinion.
I do not want to use an engine that excludes legal moves for analysis.

Now I understand. You must use at least 10 different engines for analysis. Otherwise, why would you purposely want to use a vastly inferior engine for analysis even if it did have bishop promotions? I just use the 2 strongest engines.

If this is the case then your point is sound. I'm looking to buy Fritz 15 for the features. I will not be using the included engine for analysis, even if it didn't have this problem. I might use it for friend mode though. But that's it.

I rarely use chess engines for game play.
I have a raft of machines analyzing chess positions around the clock.
I generally use:
Stockfish (bleeding edge)
Komodo 9

I often know what the answer is due to statistics. If neither engine gets the right plan, then I use other engines to see if one of them can find it.

I have most chess engines, commercial or free.

If Stockfish or Komodo dumped underpromotion, I would stop using that engine and try to find a substitute.

Analysis is very different from game play. I want a tree that is as close to perfection as possible.
I don't care if I have to analyze a position 9 times at an hour a pop or more.

[Dann Corbit]

{snip}
By the way, I think SF developers should do something like kommodo and give some controll to the user. Especialy for NullMove and LMR.
Kind regards
Bernhard

I agree with Bernhard.
Given a lot of control, there will be fans who band together to find good combinations. It is free testing for the engine writers. Consider the efforts put into testing combinations for TheKing engine of ChessMaster.

If you allow tweaking of some parameters, users will tweak them. Given enough games, they will find good settings for various types of usage.

Perhaps some settings are very good at blitz, while others are very good at long time control. The engine writers could use this to create a sliding scale by default that adjusts from one parameter value to the other.

Perhaps some settings are very good for tactical analysis. In that case, perhaps a setting like "Houdini Tactical" can be created.

[carldaman]

I agree with the assessment:
Fritz 15 is utterly unsuitable for analysis.

Great for game play, perhaps, but an engine that does not under-promote should not be used for analysis because it is defective.

As far as I know, all (top) engines fail on some positions (so they are all unsuitable for analysis?). At least with the underpromotion problem the user knows what (not) to expect.

For analysis, if the engine is designed to fail, then I do not want to use it.
Accidental failures are another matter.

Other people may feel that use for analysis is fine.
I only stated my personal opinion.
I do not want to use an engine that excludes legal moves for analysis.

Now I understand. You must use at least 10 different engines for analysis. Otherwise, why would you purposely want to use a vastly inferior engine for analysis even if it did have bishop promotions? I just use the 2 strongest engines.

If this is the case then your point is sound. I'm looking to buy Fritz 15 for the features. I will not be using the included engine for analysis, even if it didn't have this problem. I might use it for friend mode though. But that's it.


P.S. I think that not including bishop promotions most likely significantly increases elo. It simply erases an entire branch of analysis (millions of positions) and uses those resources for further depth of the position. The only problem is that it fails when an actual bishop promotion is best. This is 1 out of millions chance or less. So this trick is primarily for elo gain in engine vs. engine games. Komodo and stockfish can do the same trick for increases to their playing strength but they are strong enough.

Someone should investigate this. Modify stockfish where bishop promotion is erased. See how much elo it gains. I'm guessing at least 20.

I don't know if the Elo increase is that significant. Even if that were the case, a commercial engine author would do well to also provide an analysis version of the engine that allows all under-promotions. The playing version could still be Elo-optimized if the author insists on it

It is baffling to me that engine developers sometimes forget that many of us like to use chess engines for analysis. Good analysis features are a must for a commercial engine.

Regards,
CL

[A Distel]

Fritz 15 is not able to under promote with a bishop so it's unsuitable for analysis.

Your prejudiced opinion about Vasik Rajlich / Rybka disqualifies you from uttering such statements !

Ernest, it is completely useless to try to have a conversation with that Distel.
He is on thousand years crusade against Rajlich and has very strange views
about everything else anyways...
He just used this lack of underpromotion as a trojan horse for his bashing and some uninformed souls here even fell for it.
I am quite sure he will be on a forced hiatus soon with no chance to spread his poison again for a while as it happened already multiple times.

Hey dude, you'll never been very good at writing love ballads for Rajlich... why not growing a garden of heavy fake chessprograms?

[peter]

{snip}
By the way, I think SF developers should do something like kommodo and give some controll to the user. Especialy for NullMove and LMR.
Kind regards
Bernhard

I agree with Bernhard.
Given a lot of control, there will be fans who band together to find good combinations.

+1 from me too.
But fortunately there are many versions already around with many more parameters, eg. your MateFinder- versions with numeric nullmove, (thanks again for that, Dann!) the study- mode of late MateFinders modifies LMR.

The old parameters like King Safety, Space and so on still can be found again in late SugaR- Versions, and "Ipman" and "Bullet" once in a while bring up some modifications of their own again and again, each with certain tactical possibilities, I have one Ipman- Version from last year with almost all search- parameters tuneable and a version form Bullet from this August which is a real tactical monster too, just too especial for game playing, but who cares?
And we have Eelcos Rainbow Serpent and other dragon- and snake- fishes.

What I for my myself would appreciate most would be one more SF dev. combining all of the eval- and search- parameters in a single engine.
Of course it wouldn't be quite easy to tune it, but I always liked it most to have my own best engine for each and every possible single chess position.

And quite seriously: some corr. chess players I know use such experiments rather kindly too, they just don't talk about it.


Merry Christmas to everybody, btw

[Evert]

P.S. I think that not including bishop promotions most likely significantly increases elo. It simply erases an entire branch of analysis (millions of positions) and uses those resources for further depth of the position. The only problem is that it fails when an actual bishop promotion is best. This is 1 out of millions chance or less. So this trick is primarily for elo gain in engine vs. engine games. Komodo and stockfish can do the same trick for increases to their playing strength but they are strong enough.

Someone should investigate this. Modify stockfish where bishop promotion is erased. See how much elo it gains. I'm guessing at least 20.

I, on the other hand, would be astounded if there were any significant difference. Certainly nowhere near 20 elo.
Consider this: over 95% or so of cut-offs occur from the first three moves (I have the exact numbers somewhere), moves further down the move list are heavily reduced or pruned. Under promotions are sorted late in the list, you do not spend "millions" of positions searching those.
I think Marco once published the cost of considering obviously pointless moves (admittedly in the root) and it was insignificant.

Quite frankly the argument that omitting bishop under promotions is a way to increase strength is rather pointless when there are (many?) stronger engines that do include it. Features are more important than strength in that case.

[h1a8]

P.S. I think that not including bishop promotions most likely significantly increases elo. It simply erases an entire branch of analysis (millions of positions) and uses those resources for further depth of the position. The only problem is that it fails when an actual bishop promotion is best. This is 1 out of millions chance or less. So this trick is primarily for elo gain in engine vs. engine games. Komodo and stockfish can do the same trick for increases to their playing strength but they are strong enough.

Someone should investigate this. Modify stockfish where bishop promotion is erased. See how much elo it gains. I'm guessing at least 20.

I, on the other hand, would be astounded if there were any significant difference. Certainly nowhere near 20 elo.
Consider this: over 95% or so of cut-offs occur from the first three moves (I have the exact numbers somewhere), moves further down the move list are heavily reduced or pruned. Under promotions are sorted late in the list, you do not spend "millions" of positions searching those.
I think Marco once published the cost of considering obviously pointless moves (admittedly in the root) and it was insignificant.

Quite frankly the argument that omitting bishop under promotions is a way to increase strength is rather pointless when there are (many?) stronger engines that do include it. Features are more important than strength in that case.

makes sense. I would like to see a test done though to be sure.