Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Vas, why not give us the best Rybka possible?
1 2 3 4 Previous Next  
Parent - - By Mark (****) Date 2010-06-13 10:14

> emergency jobs


Maybe I'm too laid back about chess, but I can't imagine what a "chess emergency" is.  Anyway, if there is such a thing as a chess emergency, then I predict that the rental Rybka will be a resounding success!!  :smile:
Parent - By ChessControl (*) Date 2010-06-13 14:04
It's all about time. I can imagine that if you are a club player and come Saturday back home from the chess club and your partner demand you to help with shopping or want you to do something active with the kids and you know you have only 10-15 minutes to analyze the game you just played, before your partner become nasty. An emergency job what's the same as a queued job but more costly ($) and directly executed can bring some peace.
Parent - - By Vasik Rajlich (Silver) Date 2010-06-13 11:10
These are great questions! :smile:

> 8 hours (33% utilization) a day is not impossible but when a new launched service is continuous growing you probably want to order new hardware and racks at the moment it reach 12 hours a day (50% utilization) and hope they are delivered, tested, installed, tested again and operational before 80% utilization is reached what's probably the maximum you want to allow the cluster to be used.


There will be times when some hardware is idle, and there will be other times when all hardware is used and we have to refuse a potential user. There is a tradeoff between these two points.

Please note that the hardware won't be under my direct control, so the decision of where on this point we sit won't be made only by me.

> That queues sounds interesting to fill idle time. Is it software wise possible to freeze a processing ex-queued job, store it somewhere and queue it again and later get it back and go further where is was, so live players or emergency jobs have always higher priority to use a system?


This will be handled by the guys providing the first line of service to the user (ie. ChessBase and Convekta), if they choose to. The details will be up to them.

The general idea is that users can submit positions for analysis, and these positions are analyzed at a moment when live demand for the hardware is low.

> Is it useful to give opening book and table base authors a way to upload their latest versions and renting customers can select an author and version and pay a small fee (where a percentage go to the author) to use it in the cluster but can't download it?


Ok, this is another huge topic. Of course, eventually opening books will be online as well.

There is an entire evolution of this, which is partly done.

Until a few years ago, "computer chess opening theory" was a really underutilized resource, only a tiny number of people knew about it. By computer chess opening theory, I mean the games played in the Playchess engine room with huge amounts of daily creativity poured into them, the computer chess book wars in tournaments, the SSDF, and other media, the private analysis done by the authors, and their access to top computer chess resources such as private engine versions and top hardware.

This situation has been improved quite a bit with Jeroen's and Jiri's books, and also with some of the new Aquarium tools for managing this data. The depth and quality of data in Jiri's latest book is really high. This wouldn't have been possible until very recently.

The main remaining issue is to make things more intuitive for users. This will probably require some format enhancements (there is only so much you can do with numbers and colors), the main one probably being text comments. The issue here is that this will require a massive amount of work, which in turn will require a better mechanism for everything (and more income).

Moving the books online will (I hope) be that better mechanism. Then, we will finally have books which combine the "computer chess opening theory" strengths of ultra-deep, ultra-accurate analysis, together with the friendly presentation (meaning: text comments) which casual users want.

Ok, that was a bit of a detour :smile: but .. yes, this will also hopefully happen and it's going to be really great!

> What if independent people or companies want to build a big hardware cluster and offer chess engine programmers to upload their latest multi-core chess engine or cluster chess engine and let renting customers select a chess engine and version they want and pay a small fee (where a percentage go to the author) to use it in the cluster but can't download it. Is Rybka cluster version available for this kind of initiatives?


Private Rybka software versions are under lock and key, I won't be offering those to real third parties. It's too risky - a security leak would kill everything. There will be only one source for those.

In a few years, we'll add an FPGA version of Rybka. People who buy this product can of course rent it to others, I couldn't prevent that if I wanted to. Third-party renting will probably become quite prevalent at that point. We'll still have an "official" Rybka renting source, where everything has my stamp of approval and where we can make some commitments and promises, but there will definitely be a "free market" situation at that point.

Anybody who wants to buy the FPGA version will consider one pro and one con. The con is that they could also rent the same thing with less hassle, less commitment, etc. The pro is that they can offset their investment by renting.

Anyway, let's see how much of all of this actually happens. :smile:

Vas
Parent - - By ChessControl (*) Date 2010-06-13 14:35 Edited 2010-06-13 14:45
When I talked about 80% I mean 80% average use. By 80% average use already a lot of the time there is no free core or instance available, even by 50% average use this can already happen during peaks. It must be a balance between profitable and user satisfaction.

>This will be handled by the guys providing the first line of service to the user (ie. ChessBase and Convekta)


I was more asking if a chess engine like Rybka can be programmed to receive an external interrupt and force to stop what's doing at that moment and dump all memory data to disk and later with an other command this data can be loaded again in memory to start where it was interrupted?
Because if this is possible every little idle time can be used for scheduled jobs where the running time is unknown without delaying higher priority jobs who can suddenly be requested.

It's understandable to keep Cluster Rybka for own projects. It's almost impossible to prevent it leak but with contracts it can be made costly and there are some ways to detect and proof somebody use an unauthorized version. For example by letting Rybka cluster return very strange results with some very specials inputs only known to the author. In this case when somebody use an unauthorized Rybka cluster version in a commercial or free online cluster it's easy to detect by inputting some of this inputs and see the outcome. For example telecom companies used this to proof in court somebody copied their phone book directory because they added 100+ fake records who where then found in the copied version.

Maybe it's also possible to let programmers develop utilities for the cluster who can be rented after they are approved like the Ipad/Iphone apps store or Android apps store application who can be bought.
Parent - - By Vasik Rajlich (Silver) Date 2010-06-14 05:56

> When I talked about 80% I mean 80% average use. By 80% average use already a lot of the time there is no free core or instance available, even by 50% average use this can already happen during peaks. It must be a balance between profitable and user satisfaction.


Yes. We'll definitely do our best to add hardware before we reach 80% utilization, the numbers clearly work for that.

I'd also like to add that for at least the initial iteration, the plan is that Lukas will be providing our hardware. He's expanded his operation quite a bit. As long as Lukas provides the hardware, it will be really awesome and up-to-date and also very chess-specific when it comes to things like hard drives for tablebases, RAM needed for large pages, etc.

> I was more asking if a chess engine like Rybka can be programmed to receive an external interrupt and force to stop what's doing at that moment and dump all memory data to disk and later with an other command this data can be loaded again in memory to start where it was interrupted?


Because if this is possible every little idle time can be used for scheduled jobs where the running time is unknown without delaying higher priority jobs who can suddenly be requested.

In theory, yes. Rybka can already store her hash contents to disk. Saving the full state and restarting later isn't difficult at all.

> It's understandable to keep Cluster Rybka for own projects. It's almost impossible to prevent it leak but with contracts it can be made costly and there are some ways to detect and proof somebody use an unauthorized version. For example by letting Rybka cluster return very strange results with some very specials inputs only known to the author. In this case when somebody use an unauthorized Rybka cluster version in a commercial or free online cluster it's easy to detect by inputting some of this inputs and see the outcome.


This would be a piece of cake for any decent hacker to clean up.

> Maybe it's also possible to let programmers develop utilities for the cluster who can be rented after they are approved like the Ipad/Iphone apps store or Android apps store application who can be bought.


This sounds a little bit too sophisticated, we're at a much more primitive state right now. In theory, though, sure, why not?

Vas
Parent - - By mikal (**) Date 2010-06-14 09:34

>I'd also like to add that for at least the initial iteration, the plan is that Lukas will be providing our hardware. He's expanded his operation quite a bit. As long as Lukas provides the hardware, it will be really awesome and up-to-date and also very chess-specific when it comes to things like hard drives for tablebases, RAM needed for large pages, etc.


I thought you have some company to do the hosting like chessbase or convekta or rent some real data centers out there. This just made your renting plan with more issues, unsound and IMO unprofessional.

-First how many customers do you expect 10,100 or 1000. Can your 128 cores and his internet connectivity (which I guess is a home internet and not a datacenter class connectivity) handle all of that?
-How about necessity, how much cores and elo will each customer get? Can this beat Rybka 4 on 6x corei7 which I expect to be affordable in the near future? And how will testers test it? I don't think customers will like your statement "Elo is for testers and customers to judge".
-What about customer service, can lukas handle all of the problems and issues that might arise in the hosting? Looking at his profile I don't think he has a professional degree at some IT field.
-And what about your private hardware necessity, will customers lose access in the hardware when you have your own tournament to play or testing,  orwill you buy another 128 cores for that?
-And looking at the legal aspects, how do you handle taxes, contracts or warranties? Will this be an authoritarian model which you'll control everything and customers will get very little rights?

>In theory, yes. Rybka can already store her hash contents to disk.


So you'll also be using the customers generated data in for your own advantage that is other products like your online opening reference or tournament books etc.?
Parent - By Vasik Rajlich (Silver) Date 2010-06-15 06:09

> I thought you have some company to do the hosting like chessbase or convekta or rent some real data centers out there. This just made your renting plan with more issues, unsound and IMO unprofessional.


Actually Lukas is more professional than professionals. :smile:

> -First how many customers do you expect 10,100 or 1000. Can your 128 cores and his internet connectivity (which I guess is a home internet and not a datacenter class connectivity) handle all of that?


We will only accept as many users as we can serve. There will of course be rules about this so that for example before a freestyle tournament, users can decide for themselves how to position themselves to make sure they will have access.

> -How about necessity, how much cores and elo will each customer get?


After some time, there will be different possibilities (at first there will be just one possibility).

> Can this beat Rybka 4 on 6x corei7 which I expect to be affordable in the near future?


In the beginning, our plan is to only host things which are extremely strong.

After some time, we may open things up so that even relatively weak engines, running on fractions of cores, are hosted. Users may want to use such engines to have access to the features which are available or to connect from tiny hardware.

> And how will testers test it? I don't think customers will like your statement "Elo is for testers and customers to judge".


This is still TBD. For cheap versions, we'll probably provide some free access to testers. For the cluster, you won't need many games to get the picture.

> -What about customer service, can lukas handle all of the problems and issues that might arise in the hosting? Looking at his profile I don't think he has a professional degree at some IT field.


Customer service will be done by ChessBase and Convekta. Lukas will just log his computers into their servers.

We will also have some private users, which we'll handle directly. We've already done this and it was no problem.

> -And what about your private hardware necessity, will customers lose access in the hardware when you have your own tournament to play or testing,  orwill you buy another 128 cores for that?


We'll reserve certain time chunks like the WCCC.

> -And looking at the legal aspects, how do you handle taxes, contracts or warranties?


The same as with with any other purchase. Instead of buying a DVD with Rybka 4, you'll buy time. 99% of the transactions will be handled by ChessBase and Convekta. Special private users will get invoices directly from us.

> Will this be an authoritarian model which you'll control everything and customers will get very little rights?


The customer is always right!

>In theory, yes. Rybka can already store her hash contents to disk.
> So you'll also be using the customers generated data in for your own advantage that is other products like your online opening reference or tournament books etc.?


Only if a customer explicitly volunteers his data. We may provide some incentives.

Vas
Parent - - By Banned for Life (Gold) Date 2010-06-17 19:35
this was an interesting exchange (between Regularuser and ChessControl)

Interesting, maybe, but not very enlightening. Rule number one when trying to be realistic about offering a new service is that you shouldn't assume that your new service will dramatically increase the money flowing into the business area. Assuming this is the case here (and it almost certainly is), your sales of remote time will be based on cannibalizing sales of chess engines and possibly hardware they run on and electricity. Whatever this dollar volume is today is not likely to experience a tectonic shift upwards because of the advent of Remote Rybka. If you accept this premise, your success will be based on the following:

1) Cannibalizing sales of other engines - I have no idea what the market is for these, but I assume that you could estimate this fairly accurately with the possible exception of Chessmaster. In any event, I assume you are already the leader in non-Chessmaster engine sales, so the advantage is limited to getting a bigger cut of the gross receipts,
2) Redirecting money that would go into hardware and electricity for computer chess - I doubt this will be very significant because most people don't buy computers just for chess and many enthusiasts who buy big hardware for chess are just as excited about building and overclocking computers as they are about using them for chess.
3) Getting money from well heeled chess players looking for an advantage in their OTB or CC games. There are three problems with this; First, chess isn't polo and it's not clear that there is a lot of money sloshing around. Second, it may be hard to make the case that using Remote Rybka will improve chances in OTB games (it's probably an easier sell for CC, but there appears to be very little money in CC). Third, the fact that anyone can use this resource, means that it can't provide an advantage for anyone without the means to make an exclusive deal with Team Rybka. Apparently Topalov did this for the WCC, but at some point you will need to be able to demonstrate a correlation between using a stronger chess engine and better OTB results.

I'd be willing to bet that the money going into computer chess won't be dramatically different next year from what it was last year. There may still be room for a profitable Remote Rybka in this space, but assuming that this service will grow the sector significantly seems like wishful thinking to me.
Parent - - By Vasik Rajlich (Silver) Date 2010-06-18 08:17

> Interesting, maybe, but not very enlightening. Rule number one when trying to be realistic about offering a new service is that you shouldn't assume that your new service will dramatically increase the money flowing into the business area. Assuming this is the case here (and it almost certainly is), your sales of remote time will be based on cannibalizing sales of chess engines and possibly hardware they run on and electricity. Whatever this dollar volume is today is not likely to experience a tectonic shift upwards because of the advent of Remote Rybka. If you accept this premise


This is probably roughly true, but there are some exceptions. There are people who have money to spend and currently no good way to spend it.

> 1) Cannibalizing sales of other engines


This is negligible.

> 2) Redirecting money that would go into hardware and electricity for computer chess - I doubt this will be very significant because most people don't buy computers just for chess and many enthusiasts who buy big hardware for chess are just as excited about building and overclocking computers as they are about using them for chess.


This won't happen immediately, but people do spend substantial money on hardware and electricity. It's true that some of these users have fun doing this, but people also used to have fun investing in their personal horses and horse carriages. Times change! :smile:

> 3) Getting money from well heeled chess players looking for an advantage in their OTB or CC games. There are three problems with this; First, chess isn't polo and it's not clear that there is a lot of money sloshing around.


It's not just prize money, it's also hobby money.

> Second, it may be hard to make the case that using Remote Rybka will improve chances in OTB games (it's probably an easier sell for CC, but there appears to be very little money in CC). Third, the fact that anyone can use this resource, means that it can't provide an advantage for anyone without the means to make an exclusive deal with Team Rybka. Apparently Topalov did this for the WCC, but at some point you will need to be able to demonstrate a correlation between using a stronger chess engine and better OTB results.


Ouch. :smile: Better opening preparation gives better OTB results, I think that doesn't need illustration.

Anyway, at this point we're just going to proceed and see what happens. Business model discussions are fun but the main thing is to have a great product.

Vas
Parent - - By Banned for Life (Gold) Date 2010-06-18 18:06
There are people who have money to spend and currently no good way to spend it.

I am looking for these same people, but in my case they are spending other people's money. :lol:

    Apparently Topalov did this for the WCC, but at some point you will need to be able to demonstrate a correlation between using a stronger chess engine and better OTB results.

Ouch. :smile: Better opening preparation gives better OTB results, I think that doesn't need illustration.


No criticism was intended. As Jiri has pointed out, it's a long journey from being able to generate a better opening book, with hundreds of millions of positions, and being able to distill this down to a tiny subset which can be memorized for a match. In this process, selecting a good subset to memorize is almost certainly a lot more important than generating the book with a somewhat stronger engine.
Parent - By Vasik Rajlich (Silver) Date 2010-06-19 07:38

> I am looking for these same people, but in my case they are spending other people's money.


That's the best kind of customer. :smile:

> In this process, selecting a good subset to memorize is almost certainly a lot more important than generating the book with a somewhat stronger engine.


I think that both are pretty important. Chess players always like to start with the truth and make sure that what they are playing is sound.

Vas
Parent - By Lukas Cimiotti (Bronze) Date 2010-06-11 09:37
Thanks :smile:
It's good to see that there are not only people who say "hardware and electricity are cheap, so rental should be free", but also people who know how to really calculate such things.
Parent - - By yanquis1972 (****) Date 2010-06-10 21:16
i am certainly not, routinely. i would imagine plenty are, if the analysis is good enough.

my hope is for a package that would be close to that, but on a yearly basis, and/or an extremely convenient a la carte option. i of course have no idea if either is feasible.  i believe vas did mention an idea of a starter package at 750 euros/year, which right now is about $900. whether or not i'd go for that would depend heavily on how competitive it is with other software and hardware options.
Parent - - By dragon49 (****) Date 2010-06-10 21:25
Many of us, myself included, use the board to point out bugs and ask for fixes.

Some of these threads do however, get a little out of hand, considering the fact that Rybka 4 is undisputedly the strongest engine.  My testing has her as 59 points higher than Rybka 3 and CEGT (see http://www.husvankempen.de/nunn/40_40%20Rating%20List/40_40%20All%20Versions/rangliste.html) has her 42 Points higher than Rybka 3.

I am happy with Rybka 4.
Parent - - By yanquis1972 (****) Date 2010-06-10 21:33
what keeps me from being defining myself as happy thsu far is mainly that i was anticipating a much faster engine; something tactically along the lines of spark but with rybka's knowledge and strength. that's essentially what r3 was to r2 & from vas's words i figured we could expect another leap. so far i don't see it. i also have problems with some endgame evaluations, but i wasn't really expecting anything major there.
Parent - - By M ANSARI (*****) Date 2010-06-11 04:34
The leap from R2 to R3 will never again be repeated ... never EVER.  You have to realize that before Rybka, a 15 ELO jump per year was considered a dramatic jump and some had thought we are coming close to a wall in engine strength and that this figure would drop to maybe 5 ELO advance per year.  R3 jump was over 100 ELO and most likely that was due to a lot of "slack" in engines that Vas managed to find and improve on.  I guess if we go back a few years the best solution was for Vas to give 25 ELO more for each release ... this way he would have had 12 years of advancement where he would have made dramatically more money due to a release every year and everyone would have been happy due to progressive predictable chess engine ELO strength improvement.  If you really want to blame Vas for making you unhappy, you should blame him for spoiling us with R3.
Parent - - By yanquis1972 (****) Date 2010-06-11 04:46
i wouldn't say i am unhappy, nor do i blame anyone for anything.  of course R3 'spoiled' us, but i am quite confident (i think we all are) that R4 could have been a similar leap forward, at least elo-wise. as things are R4 is exactly as advertised & i can't say i am not 'satisfied' or regretful of having purchased it.

the idea that an individual chess engine author should behave like a company manufacturing shaving razors is quite silly to me.  i think it is human nature to take pride in ones work & want to give ones customers the best experience possible.  if vas decided this was not possible with R4 UCI i am hopeful he had good reasons & intentions for it. i don't think him greedy or disingenuous so i am comfortable assuming as much. we'll find out soon enough.
Parent - By M ANSARI (*****) Date 2010-06-11 06:25
Yes but with chess engines it is a little different.  Vas has been burned twice with de-compilation and release of his hard work.  This creates an ironic situation where he is then competing against an engine that has strength that he created himself.  Hopefully there will be a solution that will be good for all and that might come from a hardware solution.  This way full strength Rybka can be released and Vas keeps his code much more secure.  I know a lot of people are thinking that the Rybka Remote is Rybka 5, but really that is a solution for a completely different market of which very few here will make use of.
Parent - By KT (**) Date 2010-06-11 10:23
+1
Parent - - By rookie (**) Date 2010-06-11 10:39
That would not have worked given Stockfish or, for that matter, Shredder (unless the recent gains in those engines is due to Rybka 3 code). 25 elo a year would have left Rybka as just another strong UCI engine. I know nothing about programming, but I would think progress should be slowing down in engine strength improvements. The things is, though, every time someone says that, it never seems to be a programmer. I can't remember ever hearing that from a programmer.

  I'm really sorry that things have come to the point where the top engine has become a target of cloners, and where one of the most respected names in chess software development says that decompiling the strongest chess engine is a good way to 'level the playing field'. Under these circumstances I can't blame Vas for sitting on his elo.
Parent - - By Vasik Rajlich (Silver) Date 2010-06-12 07:44
There are two curves for a computer chess programmer.

When his program is weak, it's easier to improve.

As he spends more time, he learns more about how to improve things and how to allocate his time, he improves his testing procedures and hardware, he learns more about topics like how to work from home without getting distracted, etc.

These curves seem to be complementary.

Vas
Parent - - By rookie (**) Date 2010-06-14 13:10
Interesting! Unless corrected, I interpret that to mean your progress up to the present has proceeded more or less in a linear manner.
Parent - By Vasik Rajlich (Silver) Date 2010-06-15 06:11
The line is very jagged but it seems to be roughly level.

Vas
Parent - - By George Tsavdaris (****) Date 2010-06-11 11:59

>You have to realize that before Rybka, a 15 ELO jump per year was considered a dramatic jump and some had thought we are coming close to a wall in engine strength and that this figure would drop to maybe 5 ELO advance per year. 


No 15 ELO wasn't considered a dramatic jump so please do not try to distort facts to make your point that Rybka brought the evolution to big ELO leaps annually.
It brought the evolution of a huge initial ELO leap and it's true that it had from 2 to 3 another big ELO leap, but Hiarcs for example was having with every single version, big ELO leaps also.

There was a short period of time where the top commercial programs like Shredder 7 to S8 to S9 and Fritz 7 to DF7, to F8 to DF8, to F9 was having nearly zero ELO improvements, but generalizing this is not correct. So don't say "before Rybka this happened". There was a period of time that this happened but that's all.
Parent - - By M ANSARI (*****) Date 2010-06-11 14:56
What do you mean generalizing this is not correct ... it is the absolute truth.  I know so because I lived through that period and I was one of the people that started to think that we might have hit a wall in ELO strength.  Saying otherwise is ignoring history, and as you have mentioned the top engines had very little ELO gain year after year. When testing an engine we were ecstatic if there was a 20 ELO gain from version to version.  This all changed after Rybka with a major strength gain with the first commercial Rybka and then a tremendous jump after the release of R3.  I am not saying that R4 should not be stronger, it does feel like an engine that has been reduced in strength.  But I also fully appreciate and understand the reason behind that decision.
Parent - By George Tsavdaris (****) Date 2010-06-11 15:13

>Saying otherwise is ignoring history, and as you have mentioned the top engines had very little ELO gain year after year.


For some period of time and for some engines only. It was not a generally accepted phenomenon through years and for all engines.

>When testing an engine we were ecstatic if there was a 20 ELO gain from version to version.


Who was ecstatic????
I only remember critics, critics critics for the tiny ELO gains. In the CCC forum back then....

>This all changed after Rybka with a major strength gain.......


What about Fruit? Fruit was the first with the big 2-step ELO jump of +150 and +100 ELO i think(Fruit 1.5 to 2 and Fruit 2 to 2.1), after the dry period i described.
Then it was Rybka.

>I am not saying that R4 should not be stronger, it does feel like an engine that has been reduced in strength.  But I also fully appreciate and understand the reason behind that decision.


I don't appreciate it or understand it, that's why i didn't buy it.
Why do you understand it?
Parent - By George Tsavdaris (****) Date 2010-06-11 12:10

>I guess if we go back a few years the best solution was for Vas to give 25 ELO more for each release ... this way he would have had 12 years of advancement where he would have made dramatically more money due to a release every year and everyone would have been happy due to progressive predictable chess engine ELO strength improvement. 


For this to succeed 2 things should happen:
1)The other programmers should be sleeping all this time.
2)People would prefer the weaker Rybka than the stronger Stockfish for example(assuming 25 ELO per year for Rybka improvement).

You see to think that had Vas adopted the 25 ELO per year improvement, that the other programmers like authors of Stockfish or Hiarcs, Shredder or whatever, would not be able to improve their programs. Why do you believe that????

Had Rybka had 25 ELO/year improvement now it would surely be behind Stockfish, Hiarcs, Naum, Shredder Spark, Komodo and Fritz(perhaps others). Not to talk about 12 years later. Rybka's name would be forgotten for sure.
How this would be a good idea escapes my mind.:roll::roll:
Parent - By 8lrr8 (***) Date 2010-06-11 22:09
"The leap from R2 to R3 will never again be repeated ... never EVER."

seeing how r3 was ~60 elo stronger than r2 (at 40/400 on a quad), it should be no surprise to anyone r4 is only ~40 elo stronger than r3 (at ltc, no book restrictions, on an octal).

chess' draw death is here.  all hail the cluster!
Parent - - By Vasik Rajlich (Silver) Date 2010-06-11 09:13

> i believe vas did mention an idea of a starter package at 750 euros/year,


This probably won't be called the starter any more, but this will be the rough cost of having around-the-clock access to a single core for a full year.

Vas
Parent - - By Nelson Hernandez (Gold) Date 2010-06-12 15:47
Holy cow.  So a full-time octal might cost 6000 euros per year under your scheme.  At today's rate of about 1.20 to the dollar that's $7200 every year.  For that price I would surely buy the octal rather than rent.  Renting is not even remotely close to economic.

Which is fine, because Vas has made it clear that this offering is not really for us.  It's for chess professionals who can afford it. 

Renting one octal around the clock for a year probably exceeds the annual budget of a number of the less-developed FIDE nation-members!  We should just stop fixating on this product and let it go.  It's Vas's baby (well, it's his other baby).  It's not for us.
Parent - By DeletedAccount (**) Date 2010-06-12 17:42
I wonder for who this really is and who would want to use it. It makes no sense for amateurs who want to analyze their games, and it does not make sense for professionals either because they never know how secret the analysis really is. So maybe for rich computer chess enthusiasts who do not know how to set up their hardware but want to show they have the longer...ehm, stronger program.
Parent - By Vasik Rajlich (Silver) Date 2010-06-13 11:11
You probably won't rent a full octal, but you might rent one core. Just wait and see.

Vas
Parent - - By Geomusic (*****) Date 2010-06-14 20:35
what would be cool is if everyone registered into a cloud sharing our own cpus cranking positions 10x faster than Lukas on his cluster :)
Parent - By Vasik Rajlich (Silver) Date 2010-06-15 06:13
In theory that will be possible. In practice, 10x might be a bit optimistic. :smile:

Vas
Parent - - By Vasik Rajlich (Silver) Date 2010-06-11 08:37

> as long as he makes it feasible, i am fine with renting...i actually like the idea of scaling hardware & no resource time. for example if a package is economically consistent you could allocate 100% of your computers resources to IDeA with R4 & have rental rybka, even on one core if it's strong enough, doing IA.  the problem is i'm not sure how cheap he can make it. since i can rent a 12 core mega-server 24/7 for $600/month i would hope the cost would be significantly less than that. but the only way would be if he is holding back 100+ elo. we don't know that is the case.


I need to wrap up the Rybka 4 stuff, then I will have more to say about this.

One thing I understand now is that there needs to be a package which is cheap (let's say 100 Euro per year) and which doesn't use a meter (which is annoying). There will be a package like this, where the first X hours will be on a full single logical core and after that, the user will be transferred to a computer where he's getting only a share of a single logical core.

This will give people something which is very strong, fairly cheap, and where they don't have to watch the clock while still having an incentive to log off and let other people use the resources.

Please note that this will come later. We will launch with simpler, bigger things.

Vas
Parent - - By Carl Bicknell (*****) Date 2010-06-11 09:33
Vas, I have been a long time fan and I like your ideas.

Here's my concern. I am almost in the 'big hardware' category. I have a Core i7 machine I built myself which is now just under 4 GHz. What can you offer me which will compete or better that without breaking the bank?
Parent - - By Vasik Rajlich (Silver) Date 2010-06-12 07:49
First, you'll have to wait a bit. Your case isn't the one we'll address first. We're starting with the mega users, that's just the sexiest and coolest. :cool:

Down the road, you will be able to choose between a package which gives you unlimited 24/7 access, or a package which will work out to several hours per day (and after that you'll be on lesser hardware). The latter package shouldn't cost any more than Deep Rybka 4.

Vas
Parent - - By Nelson Hernandez (Gold) Date 2010-06-12 16:13

> that's just the sexiest and coolest.


Once again we see the things that most motivate you.  :grin:
Parent - - By Vasik Rajlich (Silver) Date 2010-06-13 11:12
What motivates me is the terror of five hundred people trying to log into a server at the same time! :grin:

Vas
Parent - - By Nelson Hernandez (Gold) Date 2010-06-13 11:29
From a technical standpoint that's very complicated.  But from an administrative one it's much more simple than you think!  You have a website that says something like

Current Core Capacity = xxx cores
Current Usage = xx.x%

The key thing is the fact that you are being transparent about the current online capacity.  If you hide that then people will get irate if they can't log on.  But if there are 300 cores and every one of them is filled then that's an entirely different matter from a public relations standpoint.
Parent - By Vasik Rajlich (Silver) Date 2010-06-14 05:59
Sure.

I was just referring to the fact that when you write some software and test it with (let's say) 10 users, and then jump to 500 users, you will get all sorts of unexpected new problems. Some of these problems will be scaling problems, which only become apparent with enough users (that was the problem InfinityChess ran into). Other problems will be with new users trying new things and hitting new bugs.

Vas
Parent - - By Carl Bicknell (*****) Date 2010-06-12 20:16 Edited 2010-06-12 20:32

> We're starting with the mega users, that's just the sexiest and coolest. :cool" alt=":cool::


Define mega user. I have a fast P.C and I use it 24/7. I want the fastest thing you can provide that doesn't break the bank. If you can give me 16 cores I want it.
Don't I qualify as a sexy mega user?

:grin:
Parent - By Vasik Rajlich (Silver) Date 2010-06-13 11:15

> Don't I qualify as a sexy mega user?


Well, I'm not sure about the "mega" part. :smile:

I expect that you won't want to increase the amount of money which you spend on your hobby. You'll simply redirect it from hardware and electricity to remote time. In this case, we're definitely months away from having something for you. Don't worry, the time will pass quickly. :smile:

Vas
Parent - - By KT (**) Date 2010-06-11 10:41 Edited 2010-06-11 10:53
I think all this is very interesting and exiting.

Will you also provide an option for the following scenario:
I have two hours left to prepare for the next round, and I have chosen an opening to analyze with IDeA. I decide to rent 50 cores for one hour and do some interactive analysis with IDeA, to get a grasp of the opening as fast as possible.

Would 50 cores for one hour be very expensive to rent?

EDIT: I'm talking about 50 single cores (which will be very effective for IDeA).
Parent - - By Vasik Rajlich (Silver) Date 2010-06-12 07:50
Assuming that 50 cores are currently available (on average they should be, but I am not sure exactly how availability will work out), this should be possible and should cost in the neighborhood of 25 Euro.

Vas
Parent - - By Bouddha (****) Date 2010-06-12 10:35

> this should be possible and should cost in the neighborhood of 25 Euro


An hour for 50 cores ?

regards
Parent - - By ChessControl (*) Date 2010-06-12 15:25
If looks like 0.50 euro/hour per core so 6 euro/hour for a 12 core server that's indeed close to my calculations.
Parent - - By Regularuser (***) Date 2010-06-12 16:13
Well its close to the top end of your calculations and leaves around 300% to 400% mark up for Vas and his partners if his 33% utilisation is right.  

I would probably just about use the service at that cost but sparingly.   That level of charging will probably put a lot of people off completely.

Always hard to know when a new style of product hits the market how the price/volume trade off will work.  But my gut feel is that if it was pitched lower the higher take up levels would more than compensate so that the overall profit would be much higher.

(On the other hand it is easier to start with high prices and drop them than start with low prices and raise them.)
Parent - By ChessControl (*) Date 2010-06-12 17:28 Edited 2010-06-12 17:59
The price sounds not bad to me. If the master servers has 5-6 x 512GB SSD disks (1310 euro each) in RAID5 (2-2.5TB nett) and a 8 ports fast RAID controller (510 euro) for table base storage then that server price go up with an extra 7060-8370 euro (included 20% tax).

For 50-60 cores somebody need to buy 5 x 12 core server to do it without renting. That's the choice of investing 5 x 5334 euro + 7060 euro = 33730 euro + software + energy + UPS + administrator cost etc. or rent it for 25-30 euro an hour when needed.

In my country phoning with a normal phone to a mobile phone cost 0.21 euro a minute that's 12.60 euro an hour and nobody is complaining.
Parent - By Vasik Rajlich (Silver) Date 2010-06-13 11:19

> Well its close to the top end of your calculations and leaves around 300% to 400% mark up for Vas and his partners if his 33% utilisation is right.  


The markup won't be quite that high - I would definitely prefer to go for more volume in that case.

Vas
Up Topic Rybka Support & Discussion / Rybka Discussion / Vas, why not give us the best Rybka possible?
1 2 3 4 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill