Deep Fritz 10 -- kn/s -- Skulltrail update :

[Nick C]

My participation would depend on a few things, not least that the organizers allow engines-on-auto back in. I don't subscribe to the argument that for the amount of prize money on offer one should at least turn up, that's just not true in some cases, free-weekend time is more important to me than Free-style! (So I am told).

Thanks for the offer though - I'll be in touch Will be an awesome rig.

[Dann Corbit]

Wow, maybe developments like these help put off the Hydra project.

Improvements in conventional hardwares is faster and cheaper than it would have cost Hydra to improve its hardware

Has Hydra finally lost all its head? Hope not

Hydra is custom hardware. Faster conventional hardware will have very little effect on Hydra. Rather, programs like Rybka which run on conventional CPUs and are multi-threading will be those which take the greatest advantage.

In ten years or less, your desktop will outrun the last model of Deep Blue, which cost many millions of dollars to develop.

At any point in time, a custom hardware solution can be created and it will outrun the current software based solution. But that hardware solution will age over time and eventually get overrun by commodity CPUs.

So I guess that my conclusion is this:
If someone wants to make the fastest chess engine in the world, then they will have to spend millions of dollars and create custom hardware. But if you wait ten to tweny years afterwards, you can get the same compute power for $1000.

[Nid Hogge]

Sorry for late response, been very busy lately..

To most of us here ... the more cores the better since most of us are interested in Chess engines which are one of the few applications so far that can benefit from as many cores as possible. Already Rybka and Hiarcs and ofcourse Zappa seem to be quite stable on 8 cores.

Yes, that's right, that was exactly my point. However, you could already build 8 Core machine in the past. that's what Hiarcs and Rybka used for theyre latest tournaments. Skulltrail will barely benefit them, _unless_ they OC..

Otherwise they better just build a normal dual socket Xeon machine (Harpertown) with 8 cores that would cost less and perfom in line..

I'm still holding on to my opinion that the price/performance is still bad. Yes, theyre the fastest, but that's a niche market, and you can still build a smarter price/performance that will have comperable performance. Ofcourse, if one can afford it to himself, than this is defiantly the best thing for absolute performance. This thing running at 4.5 GHz+ will be a killer at tournaments.. Maybe Hiarcs\Rybka or one of the other teams could get it going. that's going to be a nice demonstartion.

AS for FB-DIMMs, yes, I agree that they are probably the weak point in the overall system, But SkullTrail is still a great OC'er.. No need for pin mods or the like. http://www.xtremesystems.org/forums/sho ... skulltrail

They hit 4GHz+ on air pretty easily as well.. With a decent setup you could run it 4.5 GHz stable using common\pretty cheap cooling methods.
Indeed the 45nm reduced heat and wattage alot. Hope it goes well for you with the Vapo.. Guess it will have to go serious modding to cope with those beasts.. But a quality custom SS (Single Stage) or Cascade could handle this pretty well.

I am holding on my E6600 until Nehalem though.. Although those cheap Quads are very tempting!

[M ANSARI]

What I want to know is if you will be able to overclock older Xeons since Skulltrail will have multiplier unlocked. One possibility is that by simply using the FSB of 400 Mhz you could do a quick overclock of older much much cheaper legacy Xeons by simply going from 333Mhz to 400 Mhz with same multiplier.

With the case of my Vapo system ... it will definetely not be what you see in a normal Vapochill system .... this will be using huge rotary compressors which will be approximately 3X more powerful that stock Vapochills. Each socket will have its own compressor and evaporator.

[Nid Hogge]

I think the E5430's(2.66 GHz) will be a good option. 8 Multi so you could mod them and get to 3.2 GHz.costs about 500 $ each, which is way better than 1500 $ for couple of QX9775's.

Nice work with the Vapo.. Hope it goes right. Can't wait to see how fast Rybka\Hiarcs\Fritz searches with these beasts : P.