Page 1 of 1

Strange behavior when analyzing

Posted: Mon May 18, 2020 10:21 am
by ernst
Hi all, perhaps someone can reproduce the following behavior With latest Stockfish Development and Large Pages in Windows.

While analyzing the position below, I noticed that, after about 15 minutes, the nps drops signifcantly. At that same moment, CPU power consumption drops to half, but utilisation en clockspeed remains the same.

Anyone willing to give this a test?



Engine: Stockfish x64 popcnt (20480 MB)
by T. Romstad, M. Costalba, J. Kiiski, G.


62/99 10:40 -2.59 1.Nxf2 Rxc7 2.Nd1 h5 3.a4 Rd7 4.Ne3 Rd2
5.Bf3 Rf2 6.Nd5 Rf1 7.Ne3 Rc1 8.Be4+ Kf7
9.Bf3 h3 10.Bxh5+ Ke7 11.Nf5+ Kd7
12.Kg3 Rxc3+ 13.Kh2 Ke6 14.Ng3 (20.018.442.347) 31267
TB:179.596.992

63/51 11:17 -2.69-- 1.Nxf2 Rxc7 (21.200.940.226) 31277
TB:189.878.340

63/51 12:17 -2.79-- 1.Nxf2 Rxc7 (23.063.995.234) 31291
TB:207.247.666

63/51 12:38 -2.94-- 1.Nxf2 Rxc7 (23.725.912.398) 31296
TB:213.213.982

63/51 23:01 -3.15-- 1.Nxf2 Rxc7 (28.947.134.583) 20957
TB:245.141.050

63/96 31:34 -2.98++ 1.Nxf2 (30.091.633.400) 15879
TB:251.592.679

63/96 32:14 -3.15 1.Nxf2 Rxc7 2.Nd1 h5 3.a4 Rd7 4.Ne3 Rd2
5.Bf3 Rf2 6.Nd5 Rf1 7.Ne3 Rg1 8.Bg2 Ra1
9.Bc6 Kf7 10.Bd7 Ke7 11.Bb5 Ke6
12.Kf3 Rc1 13.Nc4 Rxc3+ 14.Kg2 (30.172.163.900) 15597
TB:251.993.972

64/39 32:46 -3.05++ 1.Nxf2 (30.238.749.709) 15373
TB:252.345.944

64/42 32:55 -2.95++ 1.Nxf2 (30.256.040.554) 15316
TB:252.419.032

64/84 33:09 -2.80++ 1.Nxf2 (30.284.338.589) 15225
TB:252.556.829

64/84 33:27 -3.12 1.Nxf2 Rxc7 2.Nd1 h5 3.a4 Rd7 4.Ne3 Rd2
5.Bf3 Rf2 6.Nd5 Rf1 7.Ne3 Rg1 8.Bg2 Ra1
9.Bc6 Kf7 10.Bd7 Ke7 11.Bb5 Ke6
12.Kf3 Rc1 13.Nc4 Rxc3+ 14.Kg2 (30.322.051.641) 15107
TB:252.752.225

65/91 52:32 -3.16 1.Nxf2 Rxc7 2.Nd1 h5 3.a4 Rd7 4.Ne3 Rd2
5.Bf3 Rf2 6.Nd5 Rf1 7.Ne3 Rg1 8.Bg2 Ra1
9.Bc6 Kf7 10.Bd7 Ke7 11.Bb5 Ke6
12.Kf3 Rc1 13.Nc4 Rxc3+ 14.Kg2 (32.775.465.448) 10395
TB:266.602.174

66/91 93:28 -3.16 1.Nxf2 Rxc7 2.Nd1 h5 3.a4 Rd7 4.Ne3 Rd2
5.Bf3 Rf2 6.Nd5 Rf1 7.Ne3 Rg1 8.Bg2 Ra1
9.Bc6 Kf7 10.Bd7 Ke7 11.Bb5 Ke6
12.Kf3 Rc1 13.Nc4 Rxc3+ 14.Kg2 (36.707.573.812) 6545
TB:295.057.746

67/100 97:11 -3.16 1.Nxf2 Rxc7 2.Nd1 h5 3.a4 Rd7 4.Ne3 Rd2
5.Bf3 Rf2 6.Nd5 Rf1 7.Ne3 Rg1 8.Bg2 Ra1
9.Bc6 Kf7 10.Bd7 Ke7 11.Bb5 Ke6
12.Kf3 Rc1 13.Nc4 Rxc3+ 14.Kg2 (37.206.905.230) 6380
TB:299.348.002

68/96 116:00 -3.16 1.Nxf2 Rxc7 2.Nd1 h5 3.a4 Rd7 4.Ne3 Rd2
5.Bf3 Rf2 6.Nd5 Rf1 7.Ne3 Rg1 8.Bg2 Ra1
9.Bc6 Kf7 10.Bd7 Ke7 11.Bb5 Ke6
12.Kf3 Rc1 13.Nc4 Rxc3+ 14.Kg2 (39.633.848.067) 5694
TB:320.906.704

Re: Strange behavior when analyzing

Posted: Mon May 18, 2020 11:59 am
by ernst
The issue only happpens when Large Pages are being used.One can check tis with i.e. RAMmap.

Re: Strange behavior when analyzing

Posted: Mon May 18, 2020 12:05 pm
by zullil
ernst wrote:
Mon May 18, 2020 10:21 am
Hi all, perhaps someone can reproduce the following behavior With latest Stockfish Development and Large Pages in Windows.

While analyzing the position below, I noticed that, after about 15 minutes, the nps drops signifcantly. At that same moment, CPU power consumption drops to half, but utilisation en clockspeed remains the same.

Anyone willing to give this a test?



Engine: Stockfish x64 popcnt (20480 MB)
by T. Romstad, M. Costalba, J. Kiiski, G.

How much RAM do you have?

Re: Strange behavior when analyzing

Posted: Mon May 18, 2020 12:28 pm
by ernst
32 GB, AMD 3900x