Rybka Chess Community Forum
All of you Bolsheviks decrying Vas's prices, business model, sanity, etc. are the ones missing the plot.
Be honest now. Put yourself in Vas's shoes. Would you rather...
A) Collect $100 bucks a pop from the several thousand honest people out there who actually BUY the software, half of whom are always screaming at you to fix bugs, or do this, or do that, and will abandon you the second some other engine is +5 ELO stronger on an engine ranking list, and expose yourself to nameless criminals who think it is fun and morally right to decompile and disseminate private property to the general public, OR...
B) Be at the cutting edge of chess, a person of very great interest to the professional chess community, a magnet for the very best people in hardware, software and opening books, able to provide the undisputably strongest chess the world has ever seen to anyone who is willing to pay for it? With a chance at immortality, because as others point out chess may well be practically solved sooner than any of us think.
The closest analogy I can think of: would you rather serve in the infantry and fight a nasty jungle war during monsoon season, or would you rather be a fighter jet pilot flying the best machine ever invented with total air superiority?
When you put it like that, it makes perfect sense! However, it seems you forgot to include a selling point for me as a customer. I am not here to help Vasik achieve air superiority and realize his dream as a Jedi Knight
If that is why you are here, save yourself the time and send him a blank check
What I didn't mention is that, back in the day, I served in the infantry. But I was in the airborne infantry, so at least I got a brief thrill up-front.
> you forgot to include a selling point for me as a customer
I can only giggle at this. I am explaining it to you: Vas has opened a new line of business. He is now selling air superiority fighter jets with advanced avionics, fire-and-forget missiles, heads-up target display, stealth technology, the works. If you can't afford one of these babies you can join me in my foxhole. Here, I'll trade you my can of pork and beans for your can of pineapples. I'll throw in a couple of smokes.
I'll see you there. In a couple of years we'll get the hand-me-down in the form of Rybka 5. We can then again delight in our coolness and novelty
sure if you are rich you can do anything more or less.
Come on, do the math. Most people have quads with Deep Rybka 4 on it. Vas said that Rybka is about 1.7 times faster for each doubling of the number of cores. With 256 cores the cluster is about 10 times faster compared to people's machines. If you double the speed of a CPU you gain roughly 70 ELO points. Because a cluster has some drawbacks (slow interconnection) I would suggest the cluster has +200 ELO compared to Deep Rybka on a quad. Decide for yourself if this is worth the cash.
If you are a top 10 GM or a freestylist. These 200 ELOs might just come in handy!
If money doesn't matter, you surely can rent Anand as a chess coach.
Actually, I think Anand himself could be bought for $850k/year,I bet he would probably even fly out to your desired location annnnd if you're really lucky he might give you an hour or two of "free" instruction along with the personal appearance fees :)
i think these numbers are off :)
Chess, its 1.7 for the first doubling the efficiency decreases with the addition of cores. by the time you go from 128 to 256, readup on amdahl's law. The most you will get is a speed up of 5.7x with 6x being an infinite amount of cores.
Actaully if you stick to the 1.7 times for doubling the cores you get to 256 cores being 24 times faster than a quad core. You reach 5.7 times with only 39 cores.
I agree there will be a tail off, with a law of diminishing returns, but assuming the allocation of work between cores is sensible I am sure you will contimue to get good returns up to at least 128 cores. Just think of the number of sensbile nodes to examine even 5 plies deep and then hand those nodes out to different cores. Add more cores, simply do the allocation of work at a greater depth. Of course it a bit more complicated than that, but that gets the point across.... this is a highly parallised problem and Amdahl's law recognises that it is highly scaleable.
On the other hand the 1.7 rule was not for a cluster, it was for a single server. We should expect a cluster to give less than 1.7 because the overhead for communicating between cores and retrieving or not retrieving shared reuslts (from hash) will be much higherthan on a server. I wouldn't care to speculate how much but of course Lukas will know.
I get 4.864x after 32 cores.
efficiency drops to around 50% at 16 cores. I wouldnt see it feasible to go beyond this on my limited budget
Do you realize what Aquarium does? It eliminates this overhead to a great extent. You can split cores to individual lines of analysis and combine results later. Yes, the overhead will be there during an actual game where all CPU power is spent on that 1 specific game, but the overhead loss will be reduced greatly during an Idea analysis (assuming no crashes :-) )
I was thinking about this when dreaming about Lukas combining two remote servers, ya.
Actually, A is really fun too, I have no plans to stop that!
The thing with software is - you just have to keep improving. Everything else takes care of itself ..
>The closest analogy I can think of: would you rather serve in the infantry and fight a nasty jungle war during monsoon season, or would you rather be a fighter jet pilot flying the best machine ever invented with total air superiority?
But I could think of an even better scenario: sitting at a computer monitor, sending cruise missiles and UAVs - and (that's for the near future) robot figthers instead of soldiers
Btw. I also served in the infantry - as a captain medical
i think those persons buying at $100,- a pop have been providing Vas with an income over the last 6 years.
that client base does not seem to be happy with the product and service they have received over the past two years.
let's see if the cluster will allow Vas to be at the cutting edge of computer chess AND put bread on the table.
Good summary of the situation :-)
Thanks Felix for the graphical stuff
, and everyone else for the feedback!
The Warranty is the weakest part of the Rybka Cluster Rental Program
Problems with the cluster can and will strike no matter what precautions procedures were done (Murphy's Law).
If I rent the cluster for a few days and the cluster goes down for 2 hours, I Must be credited for the downtime, either 2 free hours or its cost. (a log in the cluster should indicate the downtime, not the user)
If I rent the cluster then disater strikes (earthquake, fire, flood..etc), god forbids, and the cluster cannot be fixed for several days/weeks, then I must be credited with the lost time or the cost, whichever I choose
If you buy (or even rent) a luxury car, you get full warranty for at least 3 years
Anyone paying these astronomical prices should get much more respect, so they would comeback again
With no warranty, the Cluster Rental Program does lose large credibility on its reliability
So I suggest to have at least some kind of limited-warranty
You are right, we should (and will) guarantee something like 95% uptime - but we'll have to discuss this. Of course I don't want dissatisfied customers. Btw. there's a hot spare computer available - and my scripts discover crashes and hardware failures. In such a case the cluster will be restarted in ~ 2 minutes and in case of a hardware failure the hot spare will be used without any interaction from my side. But it also takes 2 minutes to restart the cluster if you change engines.
>If I rent the cluster then disater strikes (earthquake, fire, flood..etc), god forbids, and the cluster cannot be fixed for several days/weeks, then I must be credited with the lost time or the cost, whichever I choose
Yes, of course
95% availablity is simply awful. As an ex-CIO I would have expected to lose my job if I even suggested an availability any near as poor as that.
Imagine analysing with the cluster overnight, then there's a crash and it's all lost...ooohh that would hurt.
I suspect if there is an earthquake, followed by a fire, followed by a flood at your house, you really won't give a shit about about the guy who isn't getting his money's worth from the cluster!
Sorry, 95% uptime is a joke, isn't it? But was a good one!
Idle question. If all cluster-connected machines at your place were powered down, how long would it take to get the entire beast back up and running at full power?
Are you considering this as a tactic for the next freestyle?
There maybe a future freestyle tournament here if you are interested.
I'm always game for a freestyle event if I'm at home, but many of the top people won't show up without a small prize fund.
Yes! A GPS-guided, tactical EMP weapon! Knocks out all power for about a city block.
If a transformer near Lukas' house blows up during the event, team cato will be a suspect!
It takes 4 minutes. Most time it takes to boot BMCs (baseboard management controllers - they are required for remote operation).
That is amazing. At work it takes my laptop ten minutes to get properly booted up!
That's because your office laptop is stuffed with all sorts of %#^@Q* software which slows down your system startup
>At work it takes my laptop ten minutes to get properly booted up!
Yeah this kind of thing drives people mad, my good m8 Becks put his fist through his Sony Vaio screen and then went out and bought an expensive MAC laptop .... When I found out about the booting time horror problem which had been going on for a long time I told him to bring to me at the weekend and I would sort it out for him but with only two days to go it finally got the better of him
I've also got a Sony Vaio - and I know the pain
But I found a cure
After trying to install Windows 7 pro (which failed because some drivers were missing) I started the recovery from the recovery partition. This first installs a clean Windows 7 Home Premium with all drivers - and then shows a short message after which it wants to install all that garbage which made me format the c: partition. That short message has a cancel button - if you click it, you get a clean Vaio
But nonetheless I will never ever again buy a Sony Vaio laptop.
Mmh, I know that IBM was the best, my brother still uses an ibm laptop. Don't know if Lenovo is as good as IBM was, but I heard good things about those laptops. The disadvantage is that they are expensive. But when I installed Ubuntu on the Acer laptop of my mother, I realised why paying more can be a good idea - the bios was buggy (fan control was not working therefore, the laptop got too hot, needed a bios update) and I had a crash losing all data I collected during an experiment at university. installing Ubuntu on the ibm laptop was no problem.
> Don't know if Lenovo is as good as IBM was, but I heard good things about those laptops.
My office laptop is Lenovo and it is terribly slow when it comes to number crunching. My Dell laptop (same configuration as far as processor goes) is at least 3 times faster! Of course there is a chance that all the garbage that is installed in my Lenovo has slowed it down. But here is an example of speed difference: I ran a Java based simulation in both the laptops - Dell took only 15 min to complete whereas, Lenovo took around 40 minutes
It's almost impossible that there's such a big difference with the same processor just because of the hardware I'd think. Difficult to say what's wrong :)
I am not saying its totally due to hardware. Software is the bigger culprit I believe in my case. However, I have known technology companies to have made switch from Lenovo to Dell just because Lenovo had higher hard disk crash rate!
I have known technology companies to have made switch from Lenovo to Dell just because Lenovo had higher hard disk crash rate!
Your statement is suspect. Neither Lenovo nor Dell make hard disks, and it is hard to believe that Lenovo uses cheaper components than Dell.
It is a fact. Since I am neither a Dell, nor a Lenovo representative, I have no reason to make that up. Hard disk crash can happen due to multiple reasons. And the reason why that company made a switch is because Lenovo was no longer reliable. Official laptops contain a lot of sensitive data and if they crash - it is a big loss for the company.
i was using an IBM thinkpad and have moved to a lenovo t410i. put an ocz ssd drive in, installed ubuntu, added some module to control the fan, and it's silent and fast.
it has a fingerprint reader which i thought was creepy. especially so as, after i disabled the fingerprint reader from within bios, while the machine was 'sleeping' i pushed on the finger print reader and the machine woke up
My office laptop is also a Lenovo (T61) and there are days when my laptop overheats and the fan starts making a sound like a propeller plane
BTW that fingerprint reader is a terrible nuisance. Invariably the palm of my hand will slide over the reader and a balloon will float up saying it can't recognize the print. Thankfully my model does not come with it.
>But I found a cure
But alas once the Sony Vaio had my m8s fist through the screen it no longer worked and went to laptop heaven or more likely laptop hell
He had Vista on the laptop where made it doubly as bad lol ....
The makers of the laptops are to blame filling them with crap and maybe with trial versions of Norton etc .... In fact you can save money by not agreeing with the EULA at least it used to be where windows is concerned and install Linux yourself etc
Some people will surely see that a bug fix was promised for R3 and did not happen. A bug fix was promised for R4 and has not happened in a timely fashion. Although there is an obligation on any developer to supply a free fix for major bugs to customers. Why should any customer believe a Cluster bug will be fixed?
Powered by mwForum 2.25.1 © 1999-2011 Markus Wichitill