Testing Houidini 1.03

[Tomcass]

TESTING HOUDINI 1.03
240 games

TEST 1
Quad 2.33
Gui: Fritz 12
Ponder: Off
No Triple or Robbo Bases
Book: HS Masterbook 2
Time Control: 5 min+ 0 sec.
4 cores
120 games

2010-07-IvanHoe57WMJ-3 2010
1 2 3
1 Ivanhoe-B57d_11_w32(x4)Bill 22.5 - 17.5 21.5 - 18.5 44.0/80
2 Houdini 1.03 w32 4_CPU 17.5 - 22.5 23.0 - 17.0 40.5/80
3 Deep Rybka 4 w32 18.5 - 21.5 17.0 - 23.0 35.5/80


TEST 2.
I7 975
Gui: Fritz 12
Ponder: Off
No Triple or Robbo Bases
Book: HS Masterbook 2
Time Control: 5 min+ 0 sec.
4 cores and 8 threads
120 games

201007IvanWMH05-2 2010

1 2 3
1 Houdini 1.03 x64 8_CPU 21.0 - 19.0 21.0 - 19.0 42.0/80
2 Deep Rybka 4 x64 19.0 - 21.0 21.0 - 19.0 40.0/80
3 Ivanhoe-B 57d_whm05ppSEE_w64 19.0 - 21.0 19.0 - 21.0 38.0/80


GLOBAL

1 Houdini 1.03 4_CPU 38,5 - 41.5 44.0 - 36.0 82.5/160
1 Ivanhoe-B57d_WHM 41.5 - 38.5 40.5 - 39.5 82.0/160
3 Deep Rybka 4 36,0 – 44,0 39,5 – 40,5 75.5/160

COMMENTS

Very close performance of Houdini 1.03 and Ivan-B57d-WHM. Both clearly better than Deep Rybka 4 after 240 games in this test. Great job, Robert, Bill.

Regards,
Tom.

[Tomcass]

TESTING HOUDINI 1.03a
240 games

TEST 1
Quad 2.33
Gui: Fritz 12
Ponder: Off
No Triple or Robbo Bases
Book: HS Masterbook 2
Time Control: 5 min+ 0 sec.
4 cores for Houdini, Ivan B57D whm06 and DRybka4. Only ONE CORE (my mistake, sorry Ahmed!) for Ivan T53.
120 games
2010-07Houdini1.03a 2010

1 2 3 4
1 Houdini 1.03a w32 4_CPU 10.5 - 9.5 12.0 - 8.0 14.5 - 5.5** 37.0/60
2 Ivanhoe-B57d_whm06_w32(x4) 9.5 - 10.5 11.5 - 8.5 15.0 - 5.0 ** 36.0/60
3 Deep Rybka 4 w32 8.0 - 12.0 8.5 - 11.5 13.5 - 6.5 ** 30.0/60
4 IvanHoe T53 5.5 - 14.5 5.0 - 15.0 6.5 - 13.5 ** 17.0/60

Test 2.
I7 975
Gui: Fritz 12
Ponder: Off
No Triple or Robbo Bases
Book: HS Masterbook 2
Time Control: 5 min+ 0 sec.
4 cores and 8 threads
120 games
201007Houdini1.03ax8-1 2010

1 2 3 4
1 Houdini 1.03a x64 8_CPU 10.0 - 10.0 10.5 - 9.5 12.0 - 8.0** 32.5/60
2 IvanHoe T53(x8) 10.0 - 10.0 10.5 - 9.510.0 - 10.0 ** 30.5/60
3 Ivanhoe-B57d_whm06_w64(x8) 9.5 - 10.5 9.5 - 10.5 11.5 - 8.5 ** 30.5/60
4 Deep Rybka 4 x64 8.0 - 12.0 10.0 - 10.0 8.5 - 11.5 ** 26.5/60



GLOBAL (Not including Ivan T53 because of my mistake in first test)
1 Houdini 1.03 4_CPU 21,0 - 19.0 24,0 - 16.0 45,0/80
1 Ivanhoe-B57d_WHM 19,0 - 21.0 23,0 – 17,0 42.0/80
3 Deep Rybka 4 16,0 – 24,0 17,0 – 23,0 33,0/80

COMMENTS
Houdini 1.03a confirm its leadership. Excellent performance of Ivan-B57d-whm06 . Again both clearly better than Deep Rybka 4 after 120 games. (Plus 240 previous games at 5 min + 0). Ivan T53 seems also very strong. More test are needed to confirm this point.

Regards,
Tom.

[TPJR]

Test 2:

Shouldn't you use Houdini 1.03a x64 4_CPU instead of Houdini 1.03a x64 8_CPU? Somewhere I read that it is better to use the number of real cores.

[Tomcass]

Hello,

I use 4 real cores + 4 virtual cores hyperthread. Some people who knows well this issue recommended me to use 4 cores. I have checked my computer and the speed is very similar with 4 and 8 threads. The difference is less than 2% (up or down) in speed. Not only in NPS but also in real time to solve positions. Perhaps one day I will go to 4 cores.

Regards,

Tom.

[robbolito]

Hi Tom,
we have already discussed this question about real vs virtual cores.
On my i7 using 8 virtual instead 4 real cores increases nps by significant number but also increases the temperature of the CPUs close to 10 C.And it looks that the at the end the score is about the same.Is the stressing of your CPUs worth it if gain is insignificant.Just check yourself the CPU temp.before and after turning to 8 cores.
Anyway since I don't know much about this topic we are in the right place where knowledgeable people can give us correct answers.

[Tomcass]

Hi, Robbolito,

Probably you are right. I am considering the possibility of changing my testing rules from 8 virtual to 4 real cores. Anyway I have no problems at all with heating in my i7 975. It works days, weeks non-stop and I have not found any problem. Even now, with our hard summer in Barcelona, with temperatures not far from 35 degree.

Thank you very much for your advice!. Regards,

Tom.

[Tomcass]

TESTING HOUDINI 1.03a
240 games


TEST 1

Quad 2.33
Gui: Fritz 12
Ponder: Off
No Triple or Robbo Bases
Book: HS Masterbook 2
Time Control: 5 min+ 0 sec.
4 cores
120 games
2010-07Houdini1.03a2 2010

1 2 3 4
1 Houdini 1.03a w32 4_CPU 10.5 - 9.5 11.0 - 9.0 13.0 - 7.0 34.5/60
2 Ivanhoe-B57d_whm08_w32(x4) 9.5 - 10.5 11.0 - 9.0 9.0 - 11.0 29.5/60
3 IvanHoe T53(x4) 9.0 - 11.0 9.0 - 11.0 10.5 - 9.5 28.5/60
4 Deep Rybka 4 w32 7.0 - 13.0 11.0 - 9.0 9.5 - 10.5 27.5/60


TEST2.

I7 975
Gui: Fritz 12
Ponder: Off
No Triple or Robbo Bases
Book: HS Masterbook 2
Time Control: 5 min+ 0 sec.
4 cores and 8 threads
120 games
201007Houdini1.03ax8-2 2010

1 2 3 4
1 Ivanhoe-B57d_whm08_w64(x8) 11.0 - 9.0 9.0 - 11.0 12.5 - 7.5 32.5/60
2 Deep Rybka 4 x64 9.0 - 11.0 11.5 - 8.5 10.5 - 9.5 31.0/60
3 Houdini 1.03a x64 8_CPU 11.0 - 9.0 8.5 - 11.5 10.5 - 9.5 30.0/60
4 IvanHoe T53(x8) 7.5 - 12.5 9.5 - 10.5 9.5 - 10.5 26.5/60


GLOBAL

1 2 3 4
1 Houdini 1.03a w32 4_CPU 21,5-18,5 21,5-18.5 21,5-18,5 64,5/120
2 Ivanhoe-B57d_whm08_w32(x4) 18.5-21.5 20.0-20.0 23,5-16,5 62,0/120
3 Deep Rybka 4 w32 18,5-21,5 20.0-20.0 20,0-20.0 58,5/120
4 IvanHoe T53(x4) 18,5-21,5 16,5-23,5 20,0-20,0 55,0/120

FINAL GLOBAL CUMULATIVE TESTS


480 GAMES
1 2 3
Houdini 1.03a 81,0-79,0 86,5-73,5 167,5/320 52,34%
Ivanhoe-B 57d_whm 79,0-81,0 79,5-80,5 158,5/320 49,53%
Deep Rybka4 73,5-86,5 80,5-79,5 154,0/320 48,13%


COMMENTS

After 600 games at 5 min/0 control, and about 100 hours almost non stop I finished this series of tests. The final global accumulated includes only games played between Houdini, Ivanhoe-B57d_whm and Deep Rybka 4. That is 480 games. Ivanhoe T53 performance is not bad, but clearly below them.
At this time control it is quite clear for me that Houdini 1.03a is the number one.

My next test will double the time control: 10min/0. I will include Vlad0’s Ivanhoe 55mU together with the three mentioned engines. At 10 minutes/game I expect Deep Rybka to increase its percentage up to 50%. And Houdini 1.03a to maintain its clear leadership.

Regards,

Tom.

[robbolito]

Hi Tom,
even if you change your test rules from 8 to 4 real cores the scores should not be affected I think.
Jean Paul I think has done testing in both modes and the results were similar but at least one is not stressing the CPUs.
I wish that people who know about this stuff would give us definite answer.
In the future when engines are able to use MT than 8 cores will be helpful .

[Tomcass]

Hi Robbolito,

I have just started a new test, 720 games in three blocks of 240 each and about 240 hours almost non stop. The same engines: DR4, Houdini 1.03a and IvanhoeB57d-whm08, plus Ivanhoe 55mU. I have decided to keep going with 8 threads, since engines are a bit faster and I have no problems with stressing my computers or heating them. Perhaps one day I will change my mind...

Regards,

Tom.

[sirabc]

I wish that people who know about this stuff would give us definite answer.
In the future when engines are able to use MT than 8 cores will be helpful .

I'm sure I've read both Houdard and Vas writing it was better to use 4 cores. I'l try to find the links. Also, how does Houdini 4-core compile compare with the 8-core? With less overhead, could it be faster?