If you have one dual socket based computer, a workstation setup is clearly the way to go. If you have 100 dual socket based computers, rack mount is clearly the way to go. The real question is where the proper break over point is located, not whether this point exists or not.
The advantages of a rack based system for larger problems are:
1) Processing power in a given volume is significantly improved. Putting 14 6U blades in a row is currently about the greatest processing density you can achieve, making cooling a significant issue. There is currently a cost premium for blades, so this makes sense only if there are offsetting benefits to packing a lot of processing into a small volume.
2) Rack based systems are much more conducive to the use of PDUs which can offer higher efficiency. Supermicro claims 93% for their most efficient unit. This would be valid only at one load point, but its about 8% higher than what you would expect from a high end case power supply. The higher efficiency reduces energy cost for both processing and cooling.
3) Many rack based systems are designed to provide redundancy through automatic fail-over. This is a huge savings, since otherwise each software application has to be designed to work in a degraded mode. I'm sure you are familiar with the difficulties of this task.
4) A rack is generally considered a fixture which is expected to last for a long time, along with environmental controls and the PDU. The motherboards or blades are expected to be swapped out 3-5X more frequently. One could say the same about a case and its power supply, but a case is more vulnerable to future modifications in the physical characteristics of motherboards or blades. Since the great majority of Intel and AMD multi-socket solutions are sold for rack based applications, they and their system solution vendors are very sensitive about maintaining backward compatibility. I really doubt the question of what changes will mean for case based systems would even come up because the workstation market doesn't focus on upgrades. When your workstation becomes obsolete, you throw it out and start from scratch.
My initial comment was partially tongue in cheek, and I'm not sure that 12 motherboards in a rack configuration could be shown to be more cost effective than in cases, but it would certainly be close enough to justify a short trade study. Go to 24 motherboards and things become much clearer.
#2 is surprising, I always thought that electricity is electricity. :)
Re. #4 - all but two of my computers are single-socket. That's still where the best price vs performance tends to lie. This issue in general should favor desktops - when you get an incompatibility, you only have to replace one case.
BTW, there is one more issue. Every now and then, I retire my development machine and replace it with one of my test computers. (I just did it last week.) In theory, I could also sell my older desktops (although I have never done this). Desktops are just more flexible.
Blades are out of the question for any dedicated computer chess computing. They are just way too expensive for not nearly enough benefit. This might change if you're trying to run out of a nuclear-powered sub. :)
Racks vs cases are pretty much inconsequential if space is not a major issue. With a rack, you save a little bit of space, spend a little bit more money, and are a little bit more cramped in your cooling options. The main advantage that I can see is the ability to share power and cooling. My summary here: small plus, small minus.
My father used to tell me that if his first car had been a Rolls Royce, he would have saved money in the long run. I'm not sure this was true, but the argument for racks is pretty much the same. You pay more up front for are a long term solution to hosting your electronics, along with a high capacity, high efficiency power distribution network and sometimes a large cooling system, rather than a bunch of smaller ones. Your up front costs are higher, but you pay less down the road. The rapid decrease in computer costs has been working against this concept, but I think we've reached the end of that road. I'm also pretty sure that as the number of processors grows, rack solutions becomes more appealing unless you live in a warehouse.
As a fellow room fan cooling guy who is trying, semi-successfully, to hit multiple cases with one fan, I think a rack would actually be easier for this approach. One large stand mounted room fan could hit all the processors in a small rack in an optimal manner.
> You pay more up front for are a long term solution to hosting your electronics, along with a high capacity, high efficiency power distribution network and sometimes a large cooling system, rather than a bunch of smaller ones. Your up front costs are higher, but you pay less down the road.
I suspect that the long-term costs of a rack are still a little bit higher. Fans and power supplies are not expensive. A rack is a little bit tidier, though. This definitely is a viable solution for anyone with many computers.
FWIW - for me, space is really not an issue. I have a dedicated computer room and would have no problem tripling the # of computers in there. I'd hit the electricity supply limits of our flat long before I would hit the floor space limits. It's hard to believe that any home user would have a different situation.
Some doubts cross my mind.
2) In a chilled water cooler for either 1U motherboard or blade server format, the water is used to cool air from a front plenum which is then used to cool the processors. There is rarely enough real estate available for a 2U solution in a motherboard format.
3) Although chilled water is almost always available on ships, there are significant logistics issues associated with getting it where you want it, so it is actually common for a system to start its life without chilled water, under the assumption that ambient temperatures will not be at extreme specifications. The Dutch require extended warranties on the equipment they buy, so running the equipment hot is not a very desirable outcome. For the US, which requires a 1-year warranty which is covered by the equipment manufacturer, only software stability is an issue.
4) The Dutch submarine system that I referred to earlier is being delivered later this year and uses Xeon 5400 series processors in a 1U motherboard configuration. A follow on system that is currently in for bid uses Xeon 5500 series based blades. For numerous reasons, in the commercial world, hardware upgrades end at the final approval point, generally CDR for this type of system.
5) Radiated noise is a significant issue for submarines, so this is tightly specified. The server systems are not whisper quiet, but they are significantly quieter than what would be normal for a data processing center.
Although my role was very limited this time, I enjoyed being on the team, following the games 'live'. At last we had some decent WCCC coverage and it once again showed me how much fun such a tournament is.
Before the last round I was 100% sure we would clinch it and when I saw a Sveshnikov being played, I reported to Nick that 'we never can lose this'. But somehow Junior always finds out some weaknesses in Rybka's eval and suddenly Nick, Lukas and I (who were communicating the whole weekend through MSN) became quite nervous. With Deep Junior running on my Quad, showing small advantages for black and the seemlingly dangerous position on the board, we actually feared that we might end up getting into trouble. Then Junior played 42... Qc7? spoiling a well played game and the relief was enormous!
Even the regular mr. Murphy - who especially likes to vomit on us in the blitz championships :-) - could not spoil it, as Rybka won the WC blitz, as well as the Olympiad, with unrestricted hardware. The win against Shredder with cluster Rybka was tremendous: I was doubtful about Nick's 5.Qe2 vs the solid Berlin Wall, but Nick assured me that he saw only quick wins in his testing. Well, a win in 28 moves it was, never saw something like that in a game between two top programs! A fantastic game. Together with Junior-Hiarcs IMO the two best games in Pamplona.
At last I want to congratulate Deep Sjeng, Shredder and Junior for a well deserved shared 2nd place and an excellent performance. I must say I absolutely enjoy the fact that Deep Sjeng is a cluster program now, too. I really hope there will remain tournaments where we can use this formidable power to show the best chess there is. Hiarcs ended 5th, but they really were a bit unlucky. The last game could have gone either way, with a win Hiarcs would have been unshared 2nd.
See you next WCCC!
congratulation for the win!
Against 5. Qe2 we were out of book and the reply 5... Ng5? made this move extremely strong. Rybka played really well and won both times against us and Deep Sjeng which I think was out of book too after this move.
Well, we will have to improve further to make it harder for you guys next time! :-)
I really enjoyed the tournament this year as was well organized and covered in Internet too.
Thanks a lot, very much appreciated!
Congrats on the fine result by Shredder, like Rybka the only program without a loss in the WCCC. Give my best regards to Stefan and I am sure we will meet again in the next WCCC!
Kind regards, Jeroen
And welcome to our forum :)
well, many engines like 5... Ng5 and I do not. To me the moves should be part of a plan... maybe I am just an old fashion player! :-)
It is true that Sjeng nearly got equality, but I think such move makes it easier the game for white, however chess is nice because sometimes single moves can change the evaluation of a variation completely. Even some of the strongest players had such problem.
I just wanted to point out that the played moves by Shredder were not book moves.
This forum is really interesting and people do share their thoughts without offending anybody and getting offended either!
Of course the moves were no book moves, noone expects the spanish inqu... eh Qe2 line :) I don't think we're going to play this line again :)
well, I think that the variation with 5... Nd6 do allows some possibilities to surprise the opponent and the moves for black are not forced in my opinion. Maybe one day we will have another fight on this variation... who knows.
5... Ng5 may be playable too, but I do not like it. It is a matter of taste, but do not trust me to much what I say... sometimes I change my mind! :-))
Congratulations to you also for the good job on the book.
This forum is really interesting and people do share their thoughts without offending anybody and getting offended either!
Welcome to the Dark Side ;-) Don't tell anyone that SMK supports the new WCCC rules :)
Best Wishes and well done in Pamplona,
>Welcome to the Dark Side Don't tell anyone that SMK supports the new WCCC rules
Only because you used The Force to convince him that way. :-P
>You seem to keen on stirring things up.
Get to know his sense of humor and i don't think you would say this. :-P
OK, I will keep this secret between me and you :-)
Nick decided not to find out what improvement against 5. Qe2 you prepared for our game in the 8-core tournament. Hence the 1. d4.
Actually I advised Nick to go for 1.d4, where Nick was more in favour of 1.e4. With hindsight that might have been the better choice, but that is easy to say after the game :-).
well, I sure did have something... you know me, but to find it out you will have to play this line again :-)
Well, I do have a lot of respect for Rybka team as they all are professional and well prepared people, so this makes the fight more interesting and fun.
It is not easy to win against Rybka as your program play really strong and with many cores even more, but to be able to win is what it makes it fun for me! So far I did not succeed lately...
I am curious to see the new engine, so I hope it will be commercially available soon.
You deserved to win. I have no problems at all to admit this.
But I appreciate reading your post.
I remember the first time I did read your name in computer chess was when I first had MChess. What a beautifull time it was when human where closer to the solution to chess than machines.... :-)
I was very impressed when MChess won against US GM Larry Christiansen
Well, computer chess and chess has been my hobby and sport for many years and they are still even if I do not play chess myself anymore except with some friends when I meet them. I get fun doing what I do and this is the most important thing.
One can lose or win, but the important thing is that I always do my best and that I am able to learn something everytime to get better. If the opponent was better or is stronger I must admit it and congratulate with him! At the same time I must think how can do better to be able to win next time; if I have to improve it is my problem and not his. This is how I am.
MChess time was very special and will remain in my memories forever. MChess was the first program to win a match against a GM at long time controls.
I am going to meet Martin Hirsch in a little more than a week in San Francisco. He is like a brother to me and a great programmer, so we kept the friendship. This is very important to me. It could tell me that one day he may be back to chess... who knows... if he will than it will add something to this special world of computer chess fans. :-)
[Site "Den Haag"]
[Black "Christiansen, Larry"]
1. c4 e5 2. Nc3 Nf6 3. Nf3 Nc6 4. g3 Nd4 5. Nxe5 Qe7 6. f4 d6 7. Nd3 Bf5 8. Nb4
c6 9. d3 d5 10. a3 a5 11. Nc2 Nxc2+ 12. Qxc2 dxc4 13. e4 O-O-O 14. Be3 cxd3 15.
Bxd3 Bxe4 16. Nxe4 Nxe4 17. O-O-O f5 18. Bb6 Rd6 19. Rhe1 Qd7 20. Bxe4 fxe4 21.
Qxe4 Be7 22. Qe5 c5 23. Qxg7 Bf6 24. Qxd7+ Rxd7 25. Bxa5 h5 26. Kc2 Rg8 27. Bc3
Bd4 28. Bxd4 cxd4 29. Kd3 h4 30. gxh4 Rg4 31. Re8+ Kc7 32. Rc1+ Kb6 33. Re6+
Ka7 34. f5 Rxh4 35. Rc2 Rf4 36. Rf6 Rh7 37. Rd2 Rh3+ 38. Kc4 d3+ 39. Kc3 Rf1
40. b4 Ra1 41. Kb2 Rh1 42. Rd6 R1xh2 43. R6xd3 Rxd3 44. Rxh2 Rf3 45. Rh5 Ka6
46. a4 b5 47. a5 Rf4 48. Kc3 Rc4+ 49. Kb3 Rf4 50. Rh6+ Ka7 51. Rf6 Rf3+ 52. Kc2
Rf4 53. Kc3 Rc4+ 54. Kb3 Rf4 55. Rf8 Rf3+ 56. Kc2 Rf4 57. Kc3 Rc4+ 58. Kb3 Rf4
59. f6 Kb7 60. f7 Ka7 61. Kc3 Kb7 62. Kd3 Rf1 63. Ke4 Re1+ 64. Kf5 Rf1+ 65. Ke6
Re1+ 66. Kd6 Rf1 67. a6+ Ka7 68. Kc5 Rf5+ 69. Kc6 Rf6+ 70. Kxb5 Rb6+ 71. Kc4
Rf6 72. b5 Rf4+ 73. Kc5 Rf5+ 74. Kc6 Rc5+ 75. Kd6 Rxb5 76. Ra8+ Kxa8 77. f8=Q+
Ka7 78. Kc7 Rb1 79. Qc5+ Kxa6 80. Kc6 1-0
Hardware was a 486/25 MHz.
Sandro Necchi and MChess ... those were the days!
It is great to have you here Sandro, and to hear that Marty is ok. If only we could get SMK and Johann de Koning to join our forum, THAT would be fun! Or perhaps even Richard Lang himself! :-)
> Although my role was very limited this time, I enjoyed being on the team, following the games 'live'. At last we had some decent WCCC coverage and it once again showed me how much fun such a tournament is.
+1 - the internet coverage this time was very good.
Hans may be in trouble in Germany for Red Bull.
German Bundestag wasn´t closed! :-(
My name is Alberto Mena and I was in Pamplona during the 17th World Computer Chess Championship.
Junior plays very good games but Rybka did it too, and it was more precise, in my opinnion of course.
I took photos and videos that you can see here:
And in my blog:
I hope that this material was happy for you. The post is in spanish but for phots and movies that isn't important.
(Note: Hans Van der Zijden was a perfect embassador. Excellent his attitude during games with the audience.)
Bienvenido a nuestro foro.
Os seguiré de cerca y aportaré lo que esté en mi mano.
At 1:00 Hans just lays the smack down! :)
You can use it without problems whenever you want and wherever you want.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill