buying a new computer

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
Milos
Posts: 3387
Joined: Wed Nov 25, 2009 12:47 am

Re: buying a new computer

Post by Milos » Tue Jul 09, 2019 11:11 pm

smatovic wrote:
Tue Jul 09, 2019 2:21 pm
Milos wrote:
Tue Jul 09, 2019 12:36 pm
"Own power plant" for 100 extra W, seriously??? You can run it for a full month non-stop to accumulate the same amount of energy as one charge of Tesla Model 3 car.
"Own power plant"....it is not uncommon that people have solar plants on their roof, I know some gauchos who own windmills.

Let us assume 30 Euro-Cents per kWh, and running the hardware for 4 years 24/7, makes 1051.4 Euros for additional 100 watt.

But my point was the switch to a smaller fabrication process...

Code: Select all

Year   GPU          Fab          Power    TFLOPs    price
2014   GTX 980TI    TMSC-32nm    250 W    5.6       200 €  on eBay
2018   RTX 2080TI   TMSC-12nm    250 W    11.7      1200 € new
365 days * 24 hours * 0,250 kW * 0,30 Euros * 4 years makes 2628 Euros.

So which GPU would you prefer to buy to run 24/7 in a workstation or data-center?

Somehow I doubt that you would you buy six GTX 980TI for the price of one RTX 2080TI....

--
Srda
You are comparing apples and oranges. For GPU's power savings is huge with newer generations. So ofc one wouldn't use older generation GPUs if he plans to run 24/7 (btw. who the hell runs GPUs 24/7?)
For CPUs that's not the case. Take 2xE5-2689 (yes Sandy Bridge) that cost 100$ for both on ali/ebay. They are 16 cores it total and when run in full load they are around 220W both (at least in my system).
If you run them for chess they are almost as powerful as AMD Ryzen 3900X (non OC'ed). And they spend around 115W more (and cost just a fraction of cost of 3900X).
In addition, you have my condolences for paying 30Euro-c/kWh (you can thank your green idiots in Germany and Führer Merkel that closed most of your nuclear plants and made you pay the most expensive electricity in the world). I'm paying 12Euro-c/kWh.

Zenmastur
Posts: 515
Joined: Sat May 31, 2014 6:28 am

Re: buying a new computer

Post by Zenmastur » Tue Jul 09, 2019 11:43 pm

Laskos wrote:
Tue Jul 09, 2019 2:36 pm

I have never said that "it will be limited to its base clock". I just said that it will hardly match with the same cooling to the same temps the speeds of all cores average 3900X. Both will be boosted somewhat on all cores, but the plausible assumption would be that 3950X won't overcome the 300MHz gap towards 3900X. I even proposed that it will shrink the gap to 200MHz, having a better Boost numbers, but it will be still below 3900X. Don't misrepresent what I said. But let's see, who knows, maybe 3950X OC-es much better and you are right? I wouldn't bet on that, though.
Let me point out a couple pertinent points. The 3700X and the 3800X use the exact same chiplets as the 3950X will (except for their binning). Tests of these lesser SKU parts shows that the 8-core chiplets perform almost the same in all core clocking tests as the 6-core chiplets used on the 3600X and the 3900X. E.G. In a 10-minute run all 8-cores clocked to 4.125Ghz on Aida64's stress test when PBO 2 is used. This is when using the stock cooler supplied with the CPU and it was bumping the thermal limit. Using a non-stock cooler allowed 4.2 Ghz while dropping the CPU temperature 14 degrees C.
Again, can you read my posts properly? I said that all cores SF NPS will be higher on 3950X compared to 3900X by maybe 20-25%, and that would be higher than 12.5%. But I also said that Leela and the single core applications might be slower in the range of 5-10% on 3950X compared to 3900X (due to the cache size per core too), which might offset somewhat the SF advantage. All in all, it is not clear whether $250 are worth paying. For example, for video-gamers almost surely not. Then again, cooling might come $50 more expensive with 3950X, although you dismiss that. Let's see.
If you admit that the 3950X is likely to be 20-25% faster in nps then it is worth the 12.5% increased cost when building a system from scratch. The single core performance WILL NOT be slower on the 3950X it will be faster by 100Mhz. That is unless you configure it to NOT boost to the maximum low core count speeds. e.g. a locked in/hard overclock to a specific frequency that is less the the boost frequency. I wouldn't bother with an after market cooler unless you live in a very hot climate like a desert OR, you experienced thermal throttling problems. I live in a desert and I'm not planning on buying an aftermarket cooler. At least if I end up with a 3900X. Even if I get a 3950X I already have a cooler that will absorb 250 watts and will fit on the 3950X. So I'm good regardless of which CPU I buy.
I have never owned a Ryzen, so I don't know here. I have an Intel i7, I disabled the Turbo Boost and then OC-ed significantly all the cores. That's simple and stable, your solution seems to be too much "flexible", so I am not sure I would use such "dynamic solutions". I need stability and consistency, not some spikes of a half-GHz.
I'm currently using an I7 5820K. The base clock on this CPU is 3.3Ghz. I have run this CPU hard overclocked to 4.8 Ghz all core, 24 hours a day, 7 days a week, 365 days a year since November 2014. I can count the number of days this CPU has been UN-powered on my fingers. This system has been booted maybe 40 times in the last 5 years and half of those were due to Microsoft updates. So, I'm familiar with what it takes to get a stable and reliable overclock.

You can lock all cores to a specific frequency if needed. You will, very likely, give up some performance when doing this. Mostly you will give up the ability for 1-2 core to reach high boost clocks. They will be locked to what ever max frequency you have set. But you should realize that you will have the EXACT same problem on the 12 core part. So, I'm not sure your argument is a valid one.

Most people, myself included, don't need run to run clock stability so this isn't an issue. I just want to get the maximum NPS for the money I spend. Until it's been definitively proven that a 3900X is it, I'm planning on buying a 3950X.

Regards,

Zenmastur
Only 2 defining forces have ever offered to die for you.....Jesus Christ and the American Soldier. One died for your soul, the other for your freedom.

User avatar
Ovyron
Posts: 2829
Joined: Tue Jul 03, 2007 2:30 am

Re: buying a new computer

Post by Ovyron » Wed Jul 10, 2019 12:15 am

I think Uri's best option is to buy for the future. Mainly, if we expect that Leela is going to improve at the rate it has been improving, Uri's CPU is going to become irrelevant, and he'd want to buy the best GPU that he can afford, so his current computer would be fine, as long as it supports that GPU.
Great spirits have always encountered violent opposition from mediocre minds.

Zenmastur
Posts: 515
Joined: Sat May 31, 2014 6:28 am

Re: buying a new computer

Post by Zenmastur » Wed Jul 10, 2019 2:53 am

Ovyron wrote:
Wed Jul 10, 2019 12:15 am
I think Uri's best option is to buy for the future. Mainly, if we expect that Leela is going to improve at the rate it has been improving, Uri's CPU is going to become irrelevant, and he'd want to buy the best GPU that he can afford, so his current computer would be fine, as long as it supports that GPU.
Unlike many people on this board I don't think SF will become irrelevant any time soon. And for the amount of silicon used CPU's and AB engines still represent the best value for the money and the best value per watt of power used. I strongly suspect that the best engine will be a hybrid that uses both AB and NN's.

In any case he can add a strong GPU at anytime in the future with little effort.

Regards,

Zenmastur
Only 2 defining forces have ever offered to die for you.....Jesus Christ and the American Soldier. One died for your soul, the other for your freedom.

Dann Corbit
Posts: 10204
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: buying a new computer

Post by Dann Corbit » Wed Jul 10, 2019 3:30 am

Ovyron wrote:
Wed Jul 10, 2019 12:15 am
I think Uri's best option is to buy for the future. Mainly, if we expect that Leela is going to improve at the rate it has been improving, Uri's CPU is going to become irrelevant, and he'd want to buy the best GPU that he can afford, so his current computer would be fine, as long as it supports that GPU.
I guess his power supply may not be strong enough for two powerful GPUs.
I guess his components are old, and therefore somewhat prone to failure.
Is putting top flight new gear into a very old box asking for trouble?
A nice, new computer does not cost very much.

SF is better tactically.
Why not have a box that can run both?

We humans are bad at prediction. Something else may come along that is better than both.
Let me change that. Something else WILL come along that is better than both.
That's how it works. Always has worked that way, and I guess it will keep going like that.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

User avatar
Ovyron
Posts: 2829
Joined: Tue Jul 03, 2007 2:30 am

Re: buying a new computer

Post by Ovyron » Wed Jul 10, 2019 5:45 am

Dann Corbit wrote:
Wed Jul 10, 2019 3:30 am
Why not have a box that can run both?
Because you'd rather spend your whole budget into getting the best CPU that you can for A/B engines, or the entire budget to get the best, and most GPUs that you can for NNs, instead of something that is mediocre at both.

I recently defeated the best hybrid effort that exists, with hardware similar to what Uri would buy, with my 10 year old CPU that is 500kn/s weaker than what Uri already has, so Uri would be very disappointed by the mediocre way's performance (I know he wouldn't play with the thing and wants mainly analysis, but what quality is there to expect?), going this way would be a gamble, hoping that someone figures out one day how to mix NN with A/B to create something that isn't worse than NN or A/B alone (as things stand right now.) At the very least Uri would be better to wait to see a successful hybrid appear before going all in for it.

I also connect daily to play people on hardware that is 10 times faster than mine, and I don't really know what they're doing with the extra 18000kn/s or more, because I draw them 90% of the time, so I think going for fastest CPU is dubious.

But what about the guy that is using Leela with 3 piled up GPUs? That uses a 3 move opening book and lets Leela play from the start? He beats me on a regular basis, despite having some big time disadvantage after I come out of book, and my Stockfish loving its out of book position with high scores that turn out to be false, playing games that go into positions where Stockfish says 0.00 and Leela says -1.50, and Leela is always right, beating me while his entire CPU remains idle doing nothing.

So it seems Leela is the way to go, because for analysis you can check for tactical shoots with the hardware Uri already has, so you don't really need a better CPU.

At least empirical evidence supports this, the rest is speculation.
Great spirits have always encountered violent opposition from mediocre minds.

Zenmastur
Posts: 515
Joined: Sat May 31, 2014 6:28 am

Re: buying a new computer

Post by Zenmastur » Wed Jul 10, 2019 1:05 pm

Ovyron wrote:
Wed Jul 10, 2019 5:45 am

Because you'd rather spend your whole budget into getting the best CPU that you can for A/B engines, or the entire budget to get the best, and most GPUs that you can for NNs, instead of something that is mediocre at both.

I recently defeated the best hybrid effort that exists, with hardware similar to what Uri would buy, with my 10 year old CPU that is 500kn/s weaker than what Uri already has, so Uri would be very disappointed by the mediocre way's performance (I know he wouldn't play with the thing and wants mainly analysis, but what quality is there to expect?), going this way would be a gamble, hoping that someone figures out one day how to mix NN with A/B to create something that isn't worse than NN or A/B alone (as things stand right now.) At the very least Uri would be better to wait to see a successful hybrid appear before going all in for it.
A question. What is the best hybrid system right now in your opinion. My idea for a hybrid system would be something like training a small NN to do the LMR for a AB engine ( this seems to be where SF seems to have a weakness). Same for which TT entries to evict during replacement operations. Integrating a self-learning book that selects lines of play based an engine performance in those lines, not necessarily using NN for this one since there is a way to do it that uses less intensive calculations methods.

I also connect daily to play people on hardware that is 10 times faster than mine, and I don't really know what they're doing with the extra 18000kn/s or more, because I draw them 90% of the time, so I think going for fastest CPU is dubious.
I have long been of the opinion that iterative deepening, while it works very well, is a sub-optimal search strategy once a certain fraction of the available move time is used. The exponential growth in search time with search depth seems to bear this out. This also inhibits plan finding in AB engines because in too many cases no useful plan can be found with the limited search depths. Often times the plans that are found are sub-optimal again as a result of limited search depth. A mixed search strategy based on shorter compound ID searches seems like a reasonable approach.
But what about the guy that is using Leela with 3 piled up GPUs? That uses a 3 move opening book and lets Leela play from the start? He beats me on a regular basis, despite having some big time disadvantage after I come out of book, and my Stockfish loving its out of book position with high scores that turn out to be false, playing games that go into positions where Stockfish says 0.00 and Leela says -1.50, and Leela is always right, beating me while his entire CPU remains idle doing nothing.
I would note that just like AB progams NN programs could greatly benefit from a self learned opening book. This would free up more time once out of book and could change the effectiveness a particular size NN.

As far as your opponents are concerned, they are using a lot more powerful hardware and a lot more power. I think the break point is when you analyze the games after and you can't find any mistakes using an AB engine. When this starts happening on a regular basis then I would be worried. Until then I think it's just a huge difference in computing power.
So it seems Leela is the way to go, because for analysis you can check for tactical shoots with the hardware Uri already has, so you don't really need a better CPU.
You could, but who would want to? It's way too tedious with slow hardware. In CC chess you are likely to be playing 50 games at a time (or more), with time controls of 10 moves in 50 days. This mean your average mover rate must be 10 moves per day. If you work, sleep and have a “normal” life this leaves very little time for the analysis. This is why you run your computer 24/7 365. You clearly get much better analysis if you are running 25M nps with a 64GB TT than if running 2.5M nps with an 8GB TT.
In addition when sitting at the machines your interactive analysis goes much faster, so you get more done.

Regards,

Zenmastur
Only 2 defining forces have ever offered to die for you.....Jesus Christ and the American Soldier. One died for your soul, the other for your freedom.

corres
Posts: 1779
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: buying a new computer

Post by corres » Wed Jul 10, 2019 2:08 pm

In general if one uses two systems like an AB engine and an NN engine for making analysis that one should be a good chess player because only that one should choose the right move if there is a difference in the value of the analyzed position between the two engines.
So for a weaker chess player it is better to use only one - AB or NN - system than two.
Maybe a cross analysis can help in this situation too but it is a very time consuming method.

MikeB
Posts: 3559
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: buying a new computer

Post by MikeB » Wed Jul 10, 2019 5:54 pm

corres wrote:
Wed Jul 10, 2019 2:08 pm
In general if one uses two systems like an AB engine and an NN engine for making analysis that one should be a good chess player because only that one should choose the right move if there is a difference in the value of the analyzed position between the two engines.
So for a weaker chess player it is better to use only one - AB or NN - system than two.
Maybe a cross analysis can help in this situation too but it is a very time consuming method.
Respectfully disagree, as a weaker player using both an AB engine and NN engine, through trial and error, will learn over time which engine to trust in which type of positions. In the US, we call this "OJT", or "On the Job Training". Of course, it will take some work on their part.

User avatar
Ovyron
Posts: 2829
Joined: Tue Jul 03, 2007 2:30 am

Re: buying a new computer

Post by Ovyron » Thu Jul 11, 2019 1:48 am

Zenmastur wrote:
Wed Jul 10, 2019 1:05 pm
A question. What is the best hybrid system right now in your opinion.
It's called Eman XXX. I have no idea about how it works, as apparently it's some private thing owned by a few individuals. But Leela and Stockfish (Eman is strongest derivative) are run in parallel and the system does something to pick a move.

My guess is that there's some positions where Leela is worse than Stockfish and others where Stockfish is worse than Leela, and the system is picking moves from the worse entity, which proves fatal, or something.

So the hybrid performs worse than Stockfish or Leela alone.

I know it's the best hybrid because these people want to win very badly and are using the best that they can, if a better hybrid existed they'd be using that instead. So it turns out we're at a point where the computer plays stronger with the CPU sitting idle than with the strongest A/B engine chunking out moves and scores that just weaken Leela, and Leela is improving faster than hybrid efforts, so unless Leela hits a brick wall at some point (flat lines on her improvements), gambling for hybrid looks very risky.
Zenmastur wrote:
Wed Jul 10, 2019 1:05 pm
My idea for a hybrid system would be something like training a small NN to do the LMR for a AB engine ( this seems to be where SF seems to have a weakness). Same for which TT entries to evict during replacement operations. Integrating a self-learning book that selects lines of play based an engine performance in those lines, not necessarily using NN for this one since there is a way to do it that uses less intensive calculations methods.
Right, looks like hybrid is still on its infancy and that a good approach would need to do something different at the learning phase of the NN.
Zenmastur wrote:
Wed Jul 10, 2019 1:05 pm
I have long been of the opinion that iterative deepening, while it works very well, is a sub-optimal search strategy once a certain fraction of the available move time is used. The exponential growth in search time with search depth seems to bear this out. This also inhibits plan finding in AB engines because in too many cases no useful plan can be found with the limited search depths. Often times the plans that are found are sub-optimal again as a result of limited search depth. A mixed search strategy based on shorter compound ID searches seems like a reasonable approach.
Yeah, I can't help but think A/B engines are wasting so many resources and that a better multicore strategy must be devised. When I take a look at people with 40 cores playing 10 ply deeper than me I just think I should be losing the game badly, so drawing with ease just means A/B engines are using the extra resources improperly.
I would note that just like AB progams NN programs could greatly benefit from a self learned opening book. This would free up more time once out of book and could change the effectiveness a particular size NN.
Currently self-learning books have been tried for Eman (with its experience file) and SugarNN (where people start without book and let the thing improve and learn by itself) and for the former Khalid (Eman's author) abandoned it in favor of normal books (the engine uses experience file only after leaving book) and SugarNN is playing weaker than Eman, but I guess it'll improve as time goes by...
I think the break point is when you analyze the games after and you can't find any mistakes using an AB engine. When this starts happening on a regular basis then I would be worried.
It's already happening on a regular basis with A/B engine Vs. A/B engine, regardless of hardware (unlike 5 years ago, where a hardware like mine was suicide, and this was against top hardware of that time, which is 5 year old hardware by now). It's still happening rarely with A/B Vs. Leela. Looks like Leela's success is arriving at positions where A/B engines are much more likely to blunder, regardless of hardware (A/B engine being "outevaluated" so it doesn't matter what depth they reach), but where Leela is also likely to blunder (again, regardless of hardware, so if the position requires high tactical capability, the strongest piled GPUs will lose against A/B in hardware slower than mine.)
You could, but who would want to? It's way too tedious with slow hardware. In CC chess you are likely to be playing 50 games at a time (or more), with time controls of 10 moves in 50 days. This mean your average mover rate must be 10 moves per day. If you work, sleep and have a “normal” life this leaves very little time for the analysis. This is why you run your computer 24/7 365. You clearly get much better analysis if you are running 25M nps with a 64GB TT than if running 2.5M nps with an 8GB TT.
In addition when sitting at the machines your interactive analysis goes much faster, so you get more done.
Diminishing returns and self-refutation (where after enough analysis you find the best move and must play it, because more analysis leads to refuting it - even if the opponent wouldn't - and second best move "catching up" so you no longer know what move is best) are a huge problem in correspondence games, so that instead of playing your games at a little bit higher level (50/10 time control is long enough that you have enough time to reach the breaking point) you're better off just playing more games.

Upgrading hardware to be able to play 80 games instead of 50 doesn't make sense to me, because there's no benefit in playing more games. The branching factor of chess just doesn't make upgrading cost-effective after you reach a certain point, and I believe Uri is already beyond this point.
Great spirits have always encountered violent opposition from mediocre minds.

Post Reply