If one went to the deep thought level using FPGA, it would still be deterministic if it did not do a parallel search. And parallel search on a single FPGA is not something anyone would want to think of, since an FPGA is basically a finite-state machine... But since you can do all of the search and stuff in the hardware there is a speed gain. There is also a significant speed loss since one can't do SEE, nor things like "killer moves" and "hash move" or even a hash table for that matter.
I think that this is a solution without a viable problem today. The only way to make an FPGA useful is to use a bunch of them so that their raw speed dominates the significant efficiency losses they have to deal with...
Hsu's dissertation gives lots of insight into the problem(s).
The real gains come from multiple FPGA cards, ala Hsu and hydra, where you take those very inefficient searches done in the FPGA, and run 'em in parallel from a more usual software search, except the software search stops at some remaining depth (8, 10, would vary by program) and hands that position off to one of the many FPGA cards and says "give me a score for this branch". There is a lot lost in terms of efficiency. But not as much as a pure FPGA solution because the software part of the search can still do all the move ordering tricks. And we all know that move ordering near the root is _much_ more important than move ordering out near the tips where the FPGAs would be searching.
It is pretty easy to be impressed until you get over the "what kind of speed can I see" and get to the "what do I have to give up to get that kind of speed?" And when you start listing killer moves, SEE, hash tables you begin to get a sinking feeling. Somewhat akin to a new Indy car showing up that has a motor that turns 30K rpm. But so little torque that you have to gear it to the point it has a 150mph top speed. Then that 30K doesn't look so hot. Sounds impressive when it goes by, like a Honda S2000 winding out, but it doesn't go anywhere.
That's my take. If I had one per CPU, I would certainly take the make/unmake/evaluate/generate/etc operations and move 'em out to the FPGA as I would gain a ton in terms of speed, without giving up hashing and the good things that make us efficient today. That's no good for multiple CPUs if you only have one FPGA, as you would need one FPGA per cpu so that each could have its own board.
Also, what are your estimations of the latest FPGA cards and chess re:speedup over the best CPUs out there atm?
> isn't one of the issues with IDeA the need to have the same, reproducible, analysis each time?
>> isn't one of the issues with IDeA the need to have the same, reproducible, analysis each time?
Well there are SOME advantages to it Vempele. I had to re-do some analysis recently due to the endless power cuts I get at my home and it was nice to see things progressing pretty much like they did the time before.
For comparison, my bimonthly electric bill is around 180 pesos (11 Euro) and includes everything. It's going to be interesting to see if the Cluster flies or flops, but it feels as if it wasn't an offer directed to the Rybka Forum population.
Some deep pocket fan would DIE to please his candidate world chess champion.
So, he will offer 1 full year of total Rybka rental for his champ to prepare against his opponent for $1M USD.
And the perverse effect is that in doing so, he is actually HIJACKING the system.
Due to the early reservation scheme, the opponent team is fully prevented to use the same facility
595/100x296x365x1.3295= $854,653/yr for rental of 296 cores or another more marketable way to put it
just over 2 1/2 cents per second for a year.
> Maybe Majd.
He said he's not interested.
So it was speculated it's a service aimed to the top 10 GMs, or something.
> I'm wondering who is ever going to buy into that
We all are free not to spend money in ClusterRybka.
But nobody should try to convince others not to spend money in ClusterRybka!
Some Postings here look like such a try!
And I think, this is not fair!
Today I do not know whether I really should think so.
Maybe it's much more interesting to have a look at a large Engine-Cluster and at reports written about it and at discussions about experiences with it, than simply to have a R4 with maybe 30 ELO more.
Let's have a look at things which will come.
Spend money if you want to do it.
Or be an interested but provident spectator and maybe a discussion partner, when first reports are posted here.
(I hope, that there will no be a closed elitist (and fee required?) Forum for ClusterRybka users with verifyable at least 100 or 1000 ClusterRybka-CoreHours)
Let's have a look.
And enjoy changes in computer chess.
> stretched Ferrari Limo
> LoL, April Fools coming too early?!!!
That's not very nice.
Hydra had a similar approach and the project was abandoned. Cray Blitz and Deep Blue also were dedicated chess machines. Someday some UCI engine will emerge from nowhere and it will be stronger than any cluster around.
Ill just wait a few more years and buy my own "cluster strength" desktop then for $2000 or less. Good time to buy since we will be at the heels of the theoretical core efficiency of the cpu t i.e. adding more cores beyond this is traded off for heat dissipation. speedup = NIL Although superconducting cores could change all of this!
If in the future CPUs have more cores (which they surely will), then we'll remap everything. We'll also automatically remap existing (pre-paid) reservations in such a case, as mentioned in the announcement.
>A Rybka Cluster 100 consists of 100 physical cores and can be rented for 595 Euro (incl. VAT) per day, or 500 Euro (without VAT) per day for customers outside the EU or businesses outside Germany. The minimum continuous Rybka Cluster 100 rental time is 2 days.
How is this pricing divided? What percent goes into renting the engine and renting the real hardware itself? And I bet this is gonna be more expensive, what about internet bandwidth connecting to those servers, how much $$ you charge per GB a month on those servers not mentioning how reliable is the uptime? And what OS does those servers use? Of course you have to pay for license of those if it is Windows, even Amazon charges linux that is they don't let you install from iso for free.
You'll still need a client computer, a chess graphical user interface, electricity to run your client computer, and an internet connection.
For example, you need to buy one of this: http://www.newegg.com/Product/Product.aspx?Item=N82E16832116806&cm_re=Windows_server-_-32-116-806-_-Product
For each of your nodes, so if you have 50 nodes it can cost you 33k for the OS license alone. So what % of the renting scheme goes into this?
The cluster input and output is very sparse text, it's almost impossible to have a bandwidth problem on either end.
It has become obvious that Rybka is no longer intended to be used or enjoyed by the normal chess playing folk.
These prices are absolutely ridiculous, I wonder who will be renting your cluster.
Furthermore, there's no evidence the Rybka program running on the cluster is anything different from current version available, which by the way, has been relegated to the 2nd place in strength, substantially behind the 1st.
Good luck anyway.
this is insane vas. i sincerely think you have lost the plot entirely. you're relying on psychos & sheiks. it's a sad day for computer chess.
> The current rental options are just the starting point. We do realize that they are
> suitable for maybe 0.1% of the chess engine using population. I'll be happy with
> one good new user in the next month. Cheaper and more flexible options will be added later.
Yes. With a little luck you will soon have renting options for 0.5% of the chess engine
using population, while 99.5% of the chess engine using population are still waiting for a
"If one has Deep Rybka 4 and a single CPU PC then what is the advantage he has from rental Rybka with 1 core?
I mean is rental Rybka stronger from Rybka 4 and by how much?"
So the main question is, if one has a Quad or a 12core, what strength improvement would one be getting by renting 4 or 12 cores?
The current announcement is intentionally minimalistic. Felix is also putting together a web site and we'll add this information there, together with installation instructions, pictures, etc. Probably, this will be February-ish rather than January-ish.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill