|I won't rent it no matter what elo||110||60%|
|It has to play unbeatable chess first||1||1%|
|>250 elo than best public engine||15||8%|
|+250 elo than best public engine||4||2%|
|+200 elo than best public engine||11||6%|
|+180 elo than best public engine||2||1%|
|+150 elo than best public engine||8||4%|
|+130 elo than best public engine||6||3%|
|+100 elo than best public engine||7||4%|
|+75 elo than best public engine||7||4%|
|+50 elo than best public engine||5||3%|
|+25 elo than best public engine||6||3%|
|same elo as the best public engine||1||1%|
In theory, the Elo comparison should be between the remote entity as a whole (hardware + software) and whatever hardware the voter currently has at home. A price should also be specified, I would go with the Starter package (ie. 50 cents per hour or 750 Euro per year). Let's see how this looks.
Users will compare the aggregate strength of the remote package with the aggregate strength of whatever they can have at home.
Anyway, the results are more or less what I expected. 100 Elo is a sweet spot. There is probably a small sampling bias in favor of bigger Elo numbers here.
> Users will compare the aggregate strength of the remote package with the aggregate strength of whatever they can have at home.
On university in the 80s I got my computer-hours on university-mainframe.
Valuable hours. Not increasable. With strongly reglemented calculating-time-amounts.
And then I bought my c64 and the world got much more friendly to me.
I could program something, and I let him calculate over night.
Wonderfull, even if the C64 wasn't fast.
(BTW: Thanx Martin for 'let me think...'-Colossus!!!)
I surely don't want to go back to the 80s while calculating a chess problem on a charged mainframe.
(I'll shurely buy a full-strenght R4, and I'll look with interest at discussions about online-R4 becoming online-R5 after a while and step by step.
But I'll buy no weaker R4-version, and I'll not rent R4)
So have you shortlisted any providers for the remote server?
If not what is the procedure to be interested in that business:)
I have lined up the hardware for the launch, this will take us through the first few months (but hopefully not too much later than that :)).
After that, I will make further decisions.
> 100 Elo is a sweet spot
A question no one asked is who will verify that the Remote Rybka is 100, 200 or may be 300 elo strong. Currently independent testers like CCRL and CEGT is doing this job. Will these tester have free access to this online package and give an independent review of elostat? or this is still TBD or people should go by what you says?
Every couple of months, I'll update the proven version as well.
Going through Convekta or ChessBase will give you access to exactly the same engines, but there will be new GUI features.
I'm not so sure this is true. I still have vivid memories of the first day that R1 Demo appeared in the chessbase engine room. Only a few hours after it first arrived, it became the dominant engine. Nobody needed CEGT or CCRL confirmation of its obvious superiority. Now you will probably reply that R1 Demo was free, and indeed it was, but this leads to a different conclusion from the one you have drawn, i.e.; Before committing to a remote subscription, people will need to be convinced of its overwhelming superiority. This might be accomplished via CEGT or CCRL, or even more effectively by providing some kind of limited demo. The key thing is that people won't even consider paying for this as a service unless it is MUCH better than what they have at home. I think Vas realizes this and believes he can deliver the requisite performance.
I agree that 50 to 75 Elo won't sell this service. I think 150 Elo would be required. At this level, everyone will know who the big dog is, even without CEGT or CCRL results.
But seriously, if you assume 50 Elo per doubling of speed, 150 Elo probably makes a single core on a server play at about the same level as an i7 quad. Maybe you need 175 Elo to match an i7 Octal. But this would assume that you need to match the i7 octal with the single core server model. This is question about demand elasticity. Maybe a better comparison would be 2 server cores, or maybe the Elo jump will need to be higher than my guess. We'll find out soon enough.
As far as the ratings lists are concerned, you and I would like to see a scientifically derived Elo for the online service, but it's not clear this is a necessity. As a prime example, there are many people who believe that the mystery engine is better than R3 on a single core, but there are no official testing site results to back this up. Also, remember that most engine buyers are not even aware of the testing sites' existence. I know I had no idea there was such a thing when I bought F7 many years ago.
As far as the free short trial, I think this will be absolutely essential for getting people to pay to use remote Rybka. Nobody will want to sign up for a pig in a poke. A few minutes of free server time won't cost Vas an arm and a leg, and it will give prospective buyers a rational for ponying up for the service (assuming its good enough to warrant this level of interest).
They don't need to pay for it. For Rental Rybka you not only have to pay, you have to pay over and over again. People will be very careful and doubting.
- The hardware for the Starter version will need to be set so that the strength is sufficient.
- Having trials available for prospective users is quite important.
- I haven't decided yet about independent testing. There are a number of reasonable solutions to this, this is not a major issue.
>100 Elo is a sweet spot
So public non-rental engine will be weakened by 100 elo or delayed until another 100 elo is achieved, is this right?
(Local engine + local hardware) is a static value unless local hardware is upgraded, which is outside the scope of the comparison.
(Remote engine + remote hardware) is a moving target. The engine will be changing frequently, presumably opening the gap between local and remote. The hardware may also be changing faster than a typical private individual might upgrade ordinarily (roughly every three years) resulting in additional discontinuities.
In short, there is no way you could sustain a +100 balance if everything is changing on your end and everything is static on our end. Come time to release Rybka 5 you might be +200 better on the remote site, in which case buyers of Rybka 4 will feel like they are a full version behind and then some.
Rubbing in some salt I suppose, Sorry,
http://www.dailytech.com/Windows+7+Boasts+Better+Multicore+Support+for+up+to+256+Cores/article16479.htm Or "Why Vas already switched to Windows 7".htm
I mean, I'm just quoting from this thread the single programmer for the whole range of Rybka products, he is without any assistance responsible for the complete development of this groundbreaking project and he knows quite well that there is only time for:
I'm planning to spend at least the next 12 months on chess Elo. (Of course, with a few detours like fixing bugs. ) This is crucial right now for the logistics of the online service.
Later on, I do plan to spend more time on other things.
Realistically, where in this schedule will there be time for another regular Rybka 4. I just do not see this.
It is ok to release Rybka 4, but please cripple it so that it is no stronger than R3. This will reduce the incentive for people to reverse engineer this product. Take a long vacation and then release the non-crippled R4 as R5, but not until Lukas is running R6 on the cluster.
> No, no, no. Vas is following the brilliant campaign that I laid out earlier:
> It is ok to release Rybka 4, but please cripple it so that it is no stronger than R3. This will reduce the incentive for people to reverse engineer this product. Take a long vacation and then release the non-crippled R4 as R5, but not until Lukas is running R6 on the cluster.
Well, I think Vas is just taking your strategy to a logical conclusion and refinement Alan; if there is zero Rybka 4, there is zero incentive for people to reverse engineer it. There is anyway, as I see it, zero time for Vas alone to develop stand alone Rybka 4 because this would be a second project, a "version branch", and Vas would only be equipped to take this on with at least one extra programmer. Vas own words, I remember them well because it was very instructive of how this had to be done in the software branch, and made an impression on me at the time.
Actually Vas already made all this quite clear just after Rybka 3 and in fact in 20/20 hindsight it is clear that Rybka 4 would have to be reverse engineering proof and was planned as such already then. So I have my doubts that a new release of stand alone Rybka is realistically possible, unless the revenues double, so the programmer team could double and with the proviso that the stand alone version Rybka 4 is no threat to the server version, both in terms of product competition and in risk of reverse engineering the top line product.
For the quotes list:
Topic Rybka 3 FAQ By Vasik Rajlich Date 2008-08-12 18:34
For normal piracy, wait until Rybka 4 - we've got this all planned out!
Topic bugs in Persistent Hash By Vasik Rajlich Date 2008-08-10 07:51
This will take a version branch. I have some plans to add another programmer after Rybka 4 (just wait and see - Rybka 4 will be quite different) and then we can do such things. We slowly build everything up, just give us time ..
The main advantage of selling old, rather than branched stuff, is that it only requires a single baseline (and Vas clearly can't handle multiple baselines). This would be a problem if he had to release bug fixes, but after R3, I doubt anyone will expect this.
So in a way, it all makes sense as long as Vas is good enough to sell competitive engines that are 2-3 years old, and the cloud version is strong enough so that at least a small number of people want to pay for it.
> Horses are noble animals. What have you got against horses, Vas?
Nothing at all - just don't expect car-like performance from your horse. :)
With just 1000 customers a year. You will gross $1,000,000. Just off the yearly fees, not counting any hourly fees you may charge. With this kind of money potential someone will produce a product to compete with you. As you did to the chess programmers that came before you.
You ask too much IMO, and I will support the next programmer.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill