part of the fun is building these damn things.
But I own rather good hardware. And I assume this will be similar in future.
I don't want to pay for the hardware of other people while my hardware only gives the Client.
But other people might think different.
i have some secret rybka3 settings and it seems rybka4 is only +17 elo when i tested it over 800 games at 5/0. maybe people will figure out how to better configure Rybka4 but i am personally very disappointed.
i think, just by the sales of Rybka4 and how poorly it is doing on market Vas and his rybka team are starting to see what i've been saying all along. Quite simply put you've destroyed your reputation and your customers confidence and you have done little to nothing to restore it. its as if your customers are still waiting for the real Rybka4/Cloud to show up before making a commitment. even Rybka4 as it stands today was so under tested, when delivered you didn't even know what product you were giving to your customers. this entire situation for the last 2 years has really been poor and to keep things going you need to give people more value not less, you have to give them extra incentive. if anything this version of Rybka4 as it stands is only adding more confusion. i can help you all find a solution. we as forum members can help you out of this dilemma but you all cut off communication with us.
it seems that somewhere someone has given you really bad advice... i think we all know exactly who this business person is; though he is certainly wealthy and a fantastic chess player he is just one person. you have ignored the opinions of many for the opinions of one. you did this in order to protect your product and your potential profit yet you can see now, how counter-productive it all has been.
you must clearly see what is going on. the question is what will you do about it? they say there are two kinds of people those that stand up and those that look away.
>i haven't purchased rybka4 and will not purchase it.
>i have some secret rybka3 settings and it seems rybka4 is only +17 elo when i tested it over 800 games at 5/0.
Thanks for these informations. They tell me you use a pirated Rybka 4. Could you please tell me your real name and address so we can prosecute you?
> but i am personally very disappointed.
Surely with Rybka4 you get one of the best and maybe the verybest chess engine ever created.
And this for a fee of only 65 Euro.
Surely no reason to be impatient.
if you own R3 you might say, strength increase is not large enough to give 65 Euro. I can understand this. I haven't bought each Fritz-Version too.
You could say that you better will use other engines which are very good too, and maybe freeware like stockfish. That's reasonable.
And You might have expected, that Vas would present a larger increase. And maybe you are dissapointed because fo your misjudgement.
But the R4-engine surely is a very good engine.
> The thing about the future is that it always comes.
That's another good one!
I didn't understand this outrageous pricing until recently. The wholesale price for electricity from a large coal fired plant in the western US (i.e. with access to the plentiful coal we have here) is about $0.04 per kWH. I end up paying about $0.06 a kWH during the 8 non-summer months and about $0.08 a kWH during the 4 summer months off-peak ($0.24 a kWH from 1 PM to 7 PM, but only the refrigerators and my computers run then).
But a little while back, the company that is building the windmill farm off of Cape Cod sold half of their electrical output for $0.207 per kWH!!! This is the wholesale price! And this is probably the most consistently windy places in the US! I am dead set against paying five times more for wind power. Solar power is even more expensive. Definitely not ready for prime time...
We've already been down this road in the 70s, when we wasted tens of billions on synfuels, and more recently we've been wasting billions on ethanol.
I can't believe they are actually building _new_ coal plants :/
I support building new nuclear power plants, and storing the waste here in Nevada, but we don't have anywhere near a 200 year supply of Uranium...
The French have good nuclear technology, but I suspect the Japanese have better.
A decent 2 x 6 core 3.33GHz Nahalem server with 250GB harddisk, 6GB memory ECC/registered, redundant power supply cost 4445 euro (power use 1.37 Ampere 307 Watt)
Spread over 3 years that's 123.47 euro a month
Full rack at decent ISP cost 449 euro a month (including 10 Ampere 2300 Watt and 25Mbit bandwidth)
This mean max 7 servers in one rack to use max Ampere.
That's 64.14 euro hosting cost a month for each server
Then come OS cost round 6 euro a month for Microsoft SPLA license each server.
Total cost each server are 123.47 + 64.14 + 6.00 = 193.61 euro a month
Let say 20% tax then cost price without Rybka cluster software cost, without big SSD RAID storage for endgame, without system administrator cost, without billing costs, without helpdesk etc. is 232.33 euro a month
That's 7.74 euro a day when 24 hours busy
And 0.32 euro an hour when 24 hours busy
When each server is only sold 4 hours a day average cost price is 1393,99 euro a month is 46,47 euro a day and to make a decent profit margin this amount must be multiplied with 2 or 3 so 92.94 euro to 139.41 euro a day or 3.87 euro to 5.81 euro an hour. This is still excluded all that cost listed above so I do not expect anybody offer remote chess cluster access cheaper then this.
but i have a 2-cpu for office use and a dedicated 4-cpu already, so i only pay electricity now and what is my hw/strength gain for the price difference?
Let say you paid 500 euro for the 2 core and 650 euro for the 4 core is total 1150 euro.
Every 3 year you need to replace this for let say the same amount so you need to save 31.94 euro a month for that.
Let say you analyze every weekend average 2.5 hours chess games so 130 hours a year with both 2 and 4 core at the same time.
When you use a remote cluster and rent 6 core for 2.91 euro an hour is 378.30 euro a year is 31.52 euro a month.
In this case a remote cluster is cheaper (even when local; energy cost, system administrator cost, insurance cost, security cost are not calculated) and the remote cluster is more powerful, with higher uptime, no noise, no room space and to access from any computer and mobile device etc.
So as per your assumptions above, my 4-cpu costs €18,05 per month excluding electricity, which is €4,51 per month per cpu. Assuming 50% idle time, this would cost me €9,03 per month on a rental or €2,26 per cpu. Now we have to add electricity which is how much per cpu? And Rybka rental, let's assume €10,- per month.
And I will have to rent dedicated hw storage, I assume Rybka rental does not foresee in 200GB to store my databases and AN.
Bottom line imo is that if rental model is a pay as you go hourly rate and service, it may be interesting.
> And I will have to rent dedicated hw storage, I assume Rybka rental does not foresee in 200GB to store my databases and AN.
Things like tablebases, public databases and public opening books can be shared (with varying degrees of new functionality required).
Time you shed some light on release date and pricing ;)
I mean what's the difference if Vas found some new idea a day _after_ he released R4 that adds another 100 ELO and he saves it for RaaS, or if he found it a day before and kept it out?
It sounds like some people here are looking for some moral reason to be unhappy with R4.
Personally I *hope* Vas has another 100 ELO he found and kept out, otherwise only gaining 40-50 ELO from R3 in his two years of work is worrying about the potential progress of chess-engines ;)
I can accept a rather small strength jump of the engine of a developer, who has createt much in former times.
"Not so good, this time. I'll support his work although, maybe next engine will be better again. Maybe he will give a Servicepack"
But if he gives me bread crumbs 'cause he simply wants to make more money, this would leave me discontent.
I want to know whether I better should be discontent!
I'm not talking about him finding something _after_. That wouldn't be "intentional."
I'm not looking for a reason to be unhappy with it. More a reason to be happy with it.
If you'd prefer him to have held back 100 ELO than to have released the product in the same manner as he did before (well tested and at least near max strength) that's your own issue.
But I'm not here for your opinion, which most people here disagree with anyways.
I am here to ask Vas a question directly and see if he can at least say "no comment."
>Vas, did you intentionally hold back strength from the UCI release of Rybka 4 to give the rental version an edge and/or to discourage the cloners?
>Seriously, my policy for a long time now is to not talk about performance or specific Elo figures. This is for testers to measure and others to judge.
Ok, thanks for responding. Of course, I cannot speak for Graviton. But my interest in seeing your response is that there seems to be 2 prevailing theories regarding the strength of Rybka 4:
One theory is that Rybka 3 is already so strong that it is absurd and unrealistic to expect groups like CCRL and CEGT to obtain increases of 100+ elo, as we saw between Rybka 2 and Rybka 3. Implied in this theory is that you have not made any intentional changes to the evaluation or search code of Rybka 4 UCI as compared to Rybka 4 Rental, other than to allow the rental version to work in a cloud environment. Basically, right out of the gate at least, the 2 programs are identical.
Another theory, as implied in Graviton's question, is that for business reasons (perfectly understandable, but disappointing to many including me) is that some evaluation or search functions which will be included in Rybka 4 Rental are purposely withheld in the uci version.
So I was more interested in "difference in code" issue rather than the "difference in strength" issue. Is this something you could comment on?
I'd like to address this question later. For one thing, I'd like to discuss it with the guys who are actually out there selling Rybka 4 (now) and the remote version (later). Sorry for the lame answer.
I agree with your views and just to strenghten your argument a bit more...
1) 7 Servers could turn out to be quite a modest number. Assuming that chess professionals are comfortable with privacy issues they could end up using 16.7% of 7x 12 core servers on their own, with the rest of us providing useful expansion, possibly way beyond that point.
2) Even with 7 servers and 4 hours a day utilisation your figures give a profits of 119k - 238k euro profit per year. That is certainly a decent living. Add more servers or utilisation and the profit could look very large, giving the opporunity (but not necessity of course) to reduce prices.
3) Later in the thread when comparing with the cost of ownership you make a key assumption - 2.5 hours use a week for a typical user. Lots of people who argue against Rybka rental are assuming that they would have the powerful hardware anyway and would have much higher usage, making the rental model unattractive to them for 2 reasons. I suspect the average Rybka user probably uses it in earnest less than 2.5 hours a week, many perhaps less than 2.5 hours per week in total and would stick with their exisitng processor if it were not for chess. So if people are open minded enough to try rental I think it could work well.
[I am a case in point. If have 2 modest dual core laptops and probably would not upgrade them for many years if it were not for wanting more chess processing power. Personally, I would happily use and old engine such as R3 for many tasks (eg quickly looking at a blitz game I have played on Playchess) but would rent R4 for serious work on opening preparation or serious post-game analysis. That way I would probably use the rental software less than 2.5 hours a week so renting could seem attractive to me. I am waiting to see what the rental version looks like. If it is all good then I will use it, otherwise a will buy/build a 6 core desktop.]
Now for the what seems to me to be the trickiest assumption - utilisation. My first reaction was that 4 hours a day was quite a modest target, particularly if the number of users and servers went up so that natural demand smoothing occurred. On the other hand there are a number of things that argue against this:
- usage for non-professionals will probably be much higher at the weekend than weekdays
- during weekdays usage will mainly be in the evenings for most non-professionals and some time-zones will show much higher usage than others
- there will be events that could give very spiky demand - e.g. world championship matches where people watching on ICC or Playchess may want to analyse in the background. There is an option not to even attempt to service this demand properly (then how to protect chess professionals, perhaps a booking system?) or perhaps Vas can rent a flexible usage server farm that will cover this sort of eventuality.
Perhaps all in all 4 hours a day utilisation is a reasonable guestimate.
All in all I think the success of the rental model will depend on how well it is started up because scale is important:
- the initial pricing needs to be low enough to attract people in
- the initial service experience needs to be good enough that people want to keep using it and recommend it
- whether the privacy and usability, including integration with Chessbase, is perceived to be good enough.
It's one of those tough business calls. Do it at low cost and high prices so you are not risking much and it is more likely to fail. Do it properly so it is more likely to suceed and you are risking a lot of money.
I have my fingers crossed for a huge success :)
The best is to build up a cluster like this is in 3 years so there is always 1/3 of hardware with the fastest components and 1/3 with 1-2 year old components en 1/3 with 2-3 year old components.
There can be then 3 different prices depending the component speed.
A way to get that 4 hours a day sold up, is to offer the idle time in an auction model or lower prices as a last minute travel trip.
For users where a time component is important a cluster is the most useful. Because the price for 1 x 12 core rent for 2 hours is the same as 12 x 12 core rent for 10 minutes while the calculation speed is 8-10 times faster. In this case I assume there is a per minute counter/credit system.
The numbers here are pretty accurate.
There are a couple of costs you left out. I personally won't run the hardware (we'll have more to say about this shortly). This is a service which requires a profit. I also won't personally take orders and provide the first line of service to users. That will be done by ChessBase and Convekta and that also requires a profit.
I'm hoping for a utilization of 33.3%, not 16.7% as you give. Hopefully, this isn't a case where projections are driven by wishes instead of reality. There will be some possibilities for offline processing, which would be entered into a queue and performed when CPU time is available. (Convekta is interested in this but ChessBase for the time being is not.) There may also be different rates for different times (although in the interest of simplicity this is not the current plan).
Down the road, the servers will be replaced by computers with FPGA accelerators. For this and other reasons, I don't like the idea of using generic hardware providers.
The main short-term issue is that everybody who uses this service must have a first-class experience, and this means starting slow. We've had several plans to launch already, which always involved accepting hundreds of users at once. I got cold feet at the last minute, and I'm glad that I did. There would have been too many problems and no time to satisfy them. The same issue applies to renting single logical cores.
Anyway, it's going to be fun and I have no idea what will happen. The upside is really nice. Worst case: we'll just have a handful of wealthy users.
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?
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?
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?
Powered by mwForum 2.21.3 © 1999-2010 Markus Wichitill