HAL9000

HAL9000
"It just isn't conceivable that you can design a program strong enough to beat players like me."

July 5, 2014

Looking for more knps with Snapdragon 800: Mission fails!

Since the day i'd bought my Samsung Galaxy Note 2, i've always been satisfied with the Exynos 4412 cpu, until the moment news and releases about Snapdragon cpu over 2GHz started to rise.

Used in LG G2, Sony Xperia and some Samsung tablets, the Snapdragon 800 has 4 cores working at 2.26GHz speed. So, it's very attractive for higher chess performance on an Android device.

In 2014, we have recently seen SD801 which pushes the limit even higher, to 2.46GHz.

It was enough for me to be exicted because every new cpu put price pressure on the previous one and this could be an opportunity for me to upgrade to SD800 at least.

I started to search and swallow every bit of information on internet to learn more about this new booster and to discover whether a SD801 or SD800 can break 1 million nodes per second barrier on Android with Stockfish engine.

The verdict is highly disappointing. I've found many forum posts stating the LG G2 delivers LESS KNPS than Galaxy Note II (Galaxy Note III as well but latest Exynos' are another story!!!).

In theory it's ridiculous that 2.26GHz x 4 cores perform worse than 1.6GHz x 4 cores even if the cpu's are not the same brand. There's a gap between them anyway. How could this be true?

I've seen the bare truth with my own eyes during a quick manipulation i could practice in a store a couple of weeks ago, on a Samsung Galaxy Pro 8.4" tablet which had that SD800 cpu.

It was connected to internet and i was quick enough to download and install Droidfish app. After setting 4 threads and 64MB hash, i've started the analysis mode. At the end of a one minute run, there were only 635kNps.

Stockfish 5 on Android delivers less kNps than DD version on Android, in contrast to PC versions. But it's not the answer because i have 680kNps on Note II with SF5 already.

Finally i'm sure SD800 performs worse than Exynos 4412. This is valid for chess at least and i must keep waiting for 3GHz octacore cpu's.

Technical explanation of the disaster is the way manufacturers manage cpu performance vs consumption. They always advertise using peak values that you will very hardly obtain in real life. This commercial trend is getting nastier than ever.

LG G2 for example has an ultra compact design which would cause overheating of the cpu. The Samsung tablet i've been testing was also highly compact.

The cpu's come with more and more core every year with higher GHz but it's not easy to taste the fruit unless manufacturers support them with better working conditions and better batteries. Until then, they will be selling bits of utopia to the customer.  

Blitzoid chess engine ratings for Android: Update Round 15 of 50

In round 15, i had to rebuild divisions as 2 groups x 10 + 8 groups x 9 programs in order to keep the number of games played closest possible between programs and also to avoid unbalanced number of games per opening. I think it worked ok.

At 30% of the long run, Stockfish 5 leads again as expected.

There are two issues which remain to be fixed whenever possible:
1) Distortion caused by versions of the same program > I guess Stockfish 5 has +20 ELO thanks to the games played against other fishes.
2) Overall calibration vs other reference rating lists > I guess the list may need +150 ELO but i need to validate this with a few (!) manual games vs tabletop computers and also Windows programs.

BLITZOID RANKING / 05-JUL-2014
12188 GAMES PLAYED BY 92 PROGRAMS
COMPUTED WITH BAYESELO, OFFSET: 2117
Ra Name                    Elo  +  - Gam Sco Oppo Dra
 1 Stockfish 5            3015 39 38 268 78% 2803 31%
 2 Stockfish DD           2974 37 36 268 74% 2803 39%
 3 Stockfish 3            2903 33 33 296 60% 2831 37%
 4 Stockfish 4            2901 33 32 296 60% 2831 46%
 5 Stockfish 2.3.1        2869 33 32 294 55% 2822 45%
 6 Critter 1.4            2858 33 33 294 54% 2823 40%
 7 Critter 1.6a           2853 33 33 294 54% 2822 43%
 8 BlackMamba 2.0         2785 38 37 280 64% 2651 33%
 9 Critter 1.2            2785 36 36 290 58% 2708 33%
10 Stockfish 2.0          2780 35 35 292 50% 2772 32%
11 RobboLito 0.085e4l     2722 35 35 282 46% 2746 35%
12 Komodo32 2.03 JA       2703 35 35 288 47% 2725 31%
13 RobboLito 0.085g3l     2684 36 36 280 52% 2657 33%
14 Senpai 1.0             2668 35 36 284 43% 2716 30%
15 Komodo32 3 AB          2665 36 36 280 48% 2673 28%
16 Texel 1.04             2607 39 39 266 53% 2577 21%
17 Gaviota v1.0           2582 36 36 280 56% 2534 30%
18 Komodo32 1.3 JA        2579 37 38 278 46% 2608 26%
19 Texel 1.03             2558 39 39 264 49% 2563 23%
20 Toga II 3.0            2503 37 37 282 50% 2513 24%
21 IvanHoe 9.46b          2483 38 38 274 52% 2468 22%
22 Arasan 15.2 JA         2420 38 38 274 49% 2432 20%
23 Gaviota v0.86          2404 38 38 268 48% 2428 24%
24 Toga II 1.4.1SE        2388 37 37 276 43% 2451 27%
25 Toga II 2.0 JA         2375 36 36 272 49% 2393 26%
26 DiscoCheck 3.7.1       2361 37 37 272 47% 2391 24%
27 Texel 1.01             2352 37 37 280 44% 2406 23%
28 Arasan 13.4            2327 36 36 270 50% 2335 26%
29 DiscoCheck 4.0.1       2299 36 36 274 48% 2313 27%
30 Arasan 14.0.1          2286 36 36 276 44% 2330 28%
31 Crafty_23.4.JA_xb      2286 36 36 272 50% 2290 23%
32 GNU Chess 5.50-32      2276 35 36 272 47% 2302 29%
33 gaviota v0.84          2262 37 36 268 51% 2261 24%
34 Crafty_23.5.JA_xb      2254 37 37 272 46% 2297 19%
35 Alfil 12.10 w32        2246 36 36 266 51% 2238 25%
36 RedQueen 1.1.2         2244 37 37 274 45% 2285 20%
37 Rhetoric 1.4           2242 38 38 240 47% 2263 27%
38 Rodent 1.00            2235 36 36 264 52% 2215 31%
39 RedQueen 1.1.3 TCEC    2233 38 39 270 42% 2301 18%
40 Rotor 0.7a             2214 35 36 270 50% 2219 27%
41 Rodent 0.18.0          2210 36 36 264 56% 2166 30%
42 cheng3 1.07 JA         2182 37 37 264 48% 2191 20%
43 Rotor 0.8              2176 37 37 262 50% 2180 25%
44 GarboChess 3           2172 37 37 266 56% 2128 23%
45 Scorpio_2.7.JA_xb      2167 37 37 264 51% 2159 22%
46 Daydreamer 1.75 JA     2164 36 36 268 48% 2176 28%
47 gaviota v0.83          2140 37 37 262 47% 2165 22%
48 Sloppy_0.23.JA_xb      2127 36 36 264 48% 2140 27%
49 Pepito v1.59           2091 37 37 260 49% 2096 21%
50 Danasah_4.88.JA_xb     2079 37 37 262 47% 2098 24%
51 GNU Chess 6.0.2        2066 36 36 260 48% 2081 28%
52 DanasahZ_0.4.JA_xb     2064 36 37 258 48% 2075 26%
53 Tucano_1.04.AB_xb      2063 37 37 260 50% 2061 20%
54 DoubleCheck 2.7        2055 38 38 260 50% 2050 15%
55 DoubleCheck 2.6 JA     2051 38 38 260 49% 2057 17%
56 Danasah_5.06.JA_xb     2032 37 37 256 56% 1982 27%
57 Danasah_4.66.JA_xb     2013 38 38 258 52% 1990 20%
58 BetsabeII_1.30.JA_xb   2006 38 38 260 50% 2000 16%
59 Typhoon_1.0.r358.JA_xb 2002 37 37 260 51% 1995 20%
60 Diablo 0.5.1b JA       1994 37 37 260 53% 1966 24%
61 Olithink_5.3.2.JA_xb   1976 39 39 260 54% 1945 16%
62 GreKo_9.8.AB_uci       1962 37 37 260 49% 1965 23%
63 Greko_8.2_uci          1961 37 37 256 52% 1943 27%
64 GreKo_9.0.JA_uci       1957 37 37 260 53% 1932 20%
65 GreKo_10.0.JA_xb       1940 37 37 260 50% 1936 23%
66 Phalanx_XXIII.JA_xb    1939 39 39 260 47% 1959 13%
67 Sungorus 1.4 JA        1898 39 39 260 50% 1892 16%
68 BetsabeII_1.22.JA_xb   1852 40 39 256 55% 1805 14%
69 TJchess 1.1U           1844 39 39 260 51% 1822 22%
70 Myrddin_0.86.JA_xb     1835 39 39 260 55% 1788 16%
71 DoubleCheck 2.3        1816 40 40 260 51% 1801 16%
72 Natwarlal_0.14.JA_xb   1814 40 40 260 51% 1798 13%
73 KmtChess_1.21.JA_xb    1809 39 39 260 49% 1812 19%
74 Jazz 6.40 JA           1783 39 39 260 45% 1810 22%
75 Scidlet_2.61b2.JA_xb   1758 40 40 260 51% 1741 16%
76 Jazz v444 JA           1736 39 39 260 50% 1732 21%
77 Jazz v5.01 JA          1720 39 39 260 52% 1704 20%
78 Sjeng_1.12.JA_xb       1598 41 41 258 52% 1573 12%
79 BikJump v1.8           1571 41 41 258 51% 1555 14%
80 AdroitChess0.4 JA      1540 42 43 254 48% 1554 11%
81 Leonidas_r83.JA_xb     1497 42 42 254 50% 1484 19%
82 AdroitChess 0.3        1486 43 44 250 47% 1493 15%
83 ZCT-0.3.2500           1475 45 45 248 52% 1439  9%
84 Sjaak_4.68.JA_xb       1467 45 45 244 58% 1369 11%
85 BikJump v2.1P          1464 43 44 250 48% 1467 14%
86 Tscp_1.8.1.AB_xb       1372 45 45 244 50% 1355 11%
87 Zzzzzz_3.5.1.JA_xb     1339 44 44 242 48% 1351 18%
88 Rocinante 2.0 JA       1287 45 46 238 51% 1275 11%
89 VIRUTOR CHESS 1.1.4    1140 46 47 234 38% 1242 12%
90 VIRUTOR CHESS 1.1.1    1135 46 47 234 38% 1243 11%
91 Chess for Android       971 52 55 234 21% 1264  8%
92 Simplex 0.9.8           783 68 49 234  8% 1288  2%

Blitzoid test platform:
* Samsung Galaxy Note II @ 1.6 Ghz without downscaling
* 64MB hash tables where selectable
* 4 cpu threads where selectable
* Own books disabled and replaced by Silver Opening Suite positions (15 of 50 played so far)
* Opening positions played twice with different colors
* Tablebases and pondering off
* GUI: Aart Bik's Chess for Android
* Time control: 5 sec/move