Building the ultimate chess machine

by Albert Silver
11/18/2017 – In a recent discussion with top grandmasters, notably GM John Nunn, the topic came up on what the ideal configuration and setup would be for chess analysis, whether at home, or abroad. Naturally we are talking about reasonable budget restraints as well, since otherwise the sky is the limit. What is better? Several strong machines, or just one top computer? Read on to find out. | Photo: AMD

ChessBase 14 Download ChessBase 14 Download

Everyone uses ChessBase, from the World Champion to the amateur next door. Start your personal success story with ChessBase 14 and enjoy your chess even more!


Along with the ChessBase 14 program you can access the Live Database of 8 million games, and receive three months of free ChesssBase Account Premium membership and all of our online apps! Have a look today!

More...

The proposal

Pros travel around the world almost constantly, and the elite players more than any. You might imagine this would mean they are dependent on top-of-the-line laptops, which they certainly have as well, but they all know that a top desktop computer will always outgun even the most formidable laptop. The solution is simply to access their desktop computer remotely, via the internet.

There was a time this meant having some serious technical skills or access to some, but anyone with a subscription to ChessBase Account and a copy of ChessBase or Fritz can do this in a minute or two with minimal effort.

ChessBase Account Premium annual subscription

At the airport, in the hotel or at home on your couch: with the new ChessBase you always have access to the whole ChessBase world: the new ChessBase video library, tactics server, opening training App, the live database with eight million games, Let’s Check and web access to playchess.com

More...

The question is therefore what would that ideal desktop setup be? Presuming a spending limit of US$2000-2500 (for simplicity’s sake, US prices will be referenced) what is best? Two or three strong Intel i7 processors for example? Or perhaps a single much more powerful but more expensive machine, and if the latter what would it be?

Paradigm shift

Just a year ago, the answer to this question might have been completely different, but 2017 saw the entire consumer market stood on its head with the return of AMD into the high-end processor market. It isn’t that they had not tried before, but the previous attempts and promises, such as the much maligned Bulldozer series, fell flat on their face.

As such, the best multi-core processors were all by Intel, and the ones that went beyond the standard quad-core offerings they had been doling out over the last years thanks to their near monopoly, quickly went into the stratospheric price-range.

The IBM Blue Gene/P supercomputer "Intrepid" at Argonne National Laboratory runs 164,000 processor cores. It might be out of range of most readers. | Photo: Argonne National Laboratory

For example, in 2016, the absolute best consumer processor was the Intel i7-6950X Extreme Edition, that brought a huge ten-core processor, and ran for $1750, while the cheaper i7-6900X with eight cores retailed for $1100. However, that wasn’t the only problem. In those octa and deca-core models, each core also ran a good 25% slower than their respective colleagues in the quads for a classic three steps forward but two steps backwards situation (remember they also cost triple or quadruple).  As a result, it was hard to recommend the best CPU.

That was based on the existing standard entering 2017, and lasting until February, when AMD launched their new Ryzen series, which changed the CPU landscape from end to end. The Ryzen 1800X, which came with eight cores, yielded superior chess performance to that $1100 processor by Intel, but for just $500, and was just 10% worse than the $1700 processor by Intel. And that was just the beginning.  

In the last couple of months, AMD has also released their newest line of top processors, as if the Ryzen weren’t good enough, but catering to businesses, and consumers seeking elite performance. The special line has the pretty cool name ‘Threadripper’ (kudos to the marketing team who came up with that), comes with as many as sixteen cores, and is really aimed at professionals who do CPU intensive work such as video rendering, and more. Needless to say, it is also uniquely suited to chess engines, and delivers staggering performance.

The top processor is the AMD Ryzen Threadripper 1950X, which comes with 16 cores, all with hyperthreading like the Intel ilk, and capable of identical chess performance in Komodo to Intel’s best, the i9-7980XE, endowed with 18 cores. What impresses isn’t that AMD’s 16 cores do as well as Intel’s 18, it is the pricetag. While AMD’s offering retails for $999, reaching even a mere $799 on sale this coming Black Friday, the Intel goes for a cool $1999, and that is just the processor.

The chart above is from Komodo benchmarks published by the French site, Hardware.fr. They also provide benchmarks on a wide variety of applications, games and more, with dozens of processors. Highly recommended. | Photo: Hardware.fr

Still, the question remains: what if you could put together three full computers, each with 50% of the performance, for the price of one Threadripper? In other words, getting 50% more analysis power at your fingertips. The recommendation would still be one Threadripper, and here is why:

Potential versus Reality

The extra 50% power available is only a factor if it is actually used. Since we are talking about three computers or more, that means that while that third machine could theoretically yield extra analysis, it still needs to be managed and used. It is absolutely not obvious that a chess pro, even with a very capable second, would be able to distribute enough analysis among three different machines for this to be useful. It is hard enough at home, so imagine trying to remotely connect to three machines, one after the other, and still keep track of which was doing what. If not exploited to the max, the user would be much better served with the single, extra powerful computer.

Another factor that comes into play is the hidden cost of having multiple desktop computers. It would imply a lot of extra space, and a good deal more power consumption, which would become quite expensive as time went by. Just picture three computers, all with monitors, keyboards, cables, and you can imagine that added burden and space needed.

Also, it needs to be mentioned that the single Threadripper can still do what the three individual machines cannot: put all 16 cores on a single task. In terms of productivity, the biggest likelihood for a player, whether alone, or with a second, is to call up one machine remotely, and have it analyze a position at a time. The option to use 16 cores over 8 is important as it means that similar analysis depth can be reached in half the time. Or likewise, an extra ply (a half-move) can be reached in the same amount of time.

The king of price/performance for maximum chess analysis: AMD Ryzen Threadripper 1950X | Photo: AMD

It bears remembering that the discussion here is chess engine performance pure and simple. If chess engine performance is not your priority, or is a lesser one, then specific benchmarks and results should be made and studied. For example, buying a 16-core machine for gaming, no matter the brand, would be a tremendous waste, since not only are there no games capable of taking advantage of so many cores, but the bottleneck would be the video card long before so many cores could even begin to be useful.

What if you are considering a slightly less elite chess machine for budgetary reasons? The good news is that you can still put together a fantastic 8-core Ryzen based machine for a very affordable sum, which could be consulted remotely from even the dinkiest laptop you had. The details of such a machine will be shared in a forthcoming article.

Fritz 16 - He just wants to play!

Fritz 16 is looking forward to playing with you, and you're certain to have a great deal of fun with him too. Tense games and even well-fought victories await you with "Easy play" and "Assisted analysis" modes.

More...

Building a Threadripper computer

Below is a general summary of recommendations to give an idea of what it would cost and what components you should be considering when putting together a 16-core Threadripper computer with all the trimmings. The exact brand and models are not set in stone, so use what follows as guidelines, though note that these were carefully researched for price and performance, as well as user reviews that might highlight longterm issues reviewers might miss.

The prices link to a product page in NewEgg, the largest online computer retailer in the US, though by all means use any you find convenient. Above all, the precise components will be easily identified. The exception is the Arctic 240 water cooler, which must be purchased from the company site in order to receive the special mount needed for the Threadripper. This need to purchase directly from the manufacturer is true of almost all cooling solutions for the Threadripper, due to its exceptional size.

Component Model Price
CPU / Processor AMD RYZEN Threadripper 1950X 16-Core $970
Motherboard ASRock X399 Professional Gaming sTR4 $399
Memory G.SKILL TridentZ Series 32GB (2 x 16GB) 288-Pin DDR4 3200 $329
Cooler Arctic Liquid Freezer 240 $85
Video card GIGABYTE GeForce GTX 1050 OC Low Profile 2GB $120
Power supply EVGA SuperNOVA 750 G3 $130
Case Antec Nine Hundred Black Steel ATX Mid Tower $100
SSD Windows Samsung 960 EVO M.2 500GB NVMe PCI-Express $234
SSD tablebases Samsung 960 EVO M.2 250GB NVMe PCI-Express $142

Total: US$ 2509

A few notes on the components chosen: the video card was chosen mostly thanks to its combination of good performance and above all ability to connect four separate displays at once. This can be immensely convenient and productive, and I have three displays connected myself. The case chosen is notable for its USB 3.0 ports on the front, excellent ventilation, and solid reputation. Finally, the choice of two SSDs for storage as opposed to a single centralized one is deliberate. One SSD should be reserved for the 6-piece Syzygy tablebases and not be mixed with the one used for the operating system and databases. It wil be consulted non-stop by engines in analysis, and it would be a mistake to place it on a drive that is also being used for other purposes, since it could lead to unnecessary slowdowns.

I hope you find the above of interest and look forward to reading about your next super chess machine.



Born in the US, he grew up in Paris, France, where he completed his Baccalaureat, and after college moved to Rio de Janeiro, Brazil. He had a peak rating of 2240 FIDE, and was a key designer of Chess Assistant 6. In 2010 he joined the ChessBase family as an editor and writer at ChessBase News. He is also a passionate photographer with work appearing in numerous publications.
Discussion and Feedback Join the public discussion or submit your feedback to the editors


Discuss

Rules for reader comments

 
 

Not registered yet? Register

Nordlandia Nordlandia 11/28/2017 02:56
Fairly well. Btw concerning engine matches from cloud service, it's probably better idea to use wired connection with ethernet cables than wireless. Wired connection is more secure and reliable opposed to wireless wifi. Just discovered this today.

This probably solves the issue with connection issues mid-game, as i experienced earlier.


Cloud engine matches allow for using ponder without inflicted speed penalty.
Bertman Bertman 11/27/2017 06:52
@Nordlandia - I am aware, and Ronald de Man has stated that not only does it add nothing, other than a cosmetic change, but is a huge waste of disk space. Personally, I keep the 6-pc Syzygy on a 128GB pendrive.
Budd Budd 11/26/2017 11:07
Chessbmb and Middlewate are right. Threadrippers chess performance will be halved by using only 2 dimms. Not recommended if you are looking high plys.
SambalOelek SambalOelek 11/23/2017 10:02
chess is dominated by engines, not by brain

in ten years time ROBOTS will compete for the world championship of chess...

Sad or Revolutionary?
Gildor Gildor 11/23/2017 01:52
@middlewave Thank you very much!
Nordlandia Nordlandia 11/22/2017 08:06
@Bertman - to ensure you're aware, the maker of syzygy is generating brand new depth to mate format. Estimated size is 500 GB +/-

Once wdl find certain TB win in search, dtm kicks in with exact distance to mate. It does not add any elo, just cosmetic change.

Syzygy + DTM utilize about ~750 GB, which perfectly fits on 1TB SSD, with 250 GB available for usage elsewhere.
middlewave middlewave 11/22/2017 02:33
@Gildor
It all depends on adequate cooling. I suppose water cooling should work best, and even if you are inexperienced in this field you should not despair: there are pretty good water-cooling All-In-Ones in the market, very easy to install. But a good air cooler should suffice as well; just make sure your case is large enough. I would strongly suggest to look at Noctua air coolers, they have stellar reputation for performance and reliability, and they are pretty quiet too.
Btw water-cooling is not 100% quiet: the coolers DO use fans as well! But they don't need to operate at high speeds, as with an air cooler, so it will be pretty quiet overall.
Gildor Gildor 11/21/2017 07:44
Thanks for the great article! I am looking for a computer that is best for very long analyses of positions. It should be able to continuously analyze 24/7 without heating up. It also needs to be quiet. Would the proposed set-up be capable of meeting these two requirements?
daniel7472 daniel7472 11/21/2017 07:35
@chessmbs, middlewave , thank you for your reply.
I asked about the hyperthreading since I am about to but a PC and I am still not decided between i5 8400 and ryzen 5 1600. Ryzen will give a better score in Stockfish and Komodo but I have also read about the inefficiency of hyperthreading hence for the question.
chessmbs chessmbs 11/21/2017 07:10
@SambalOelek
Name a few...well, no...Just 1 usable and working example with links please...
SambalOelek SambalOelek 11/20/2017 11:47
guys why are you all building a pc??

juts put it in the cloud...use internet servers to do the calculations

you are so 90's...

mmm well i am so 60's with my books :)
chessmbs chessmbs 11/20/2017 06:52
@middlewave
Well :-) that would be rather strange....offsetting the mainboard capability into the future just to be able to upgrade later ... to be able to use it at it full potential...actually the way they ment to be from the beginning....With a build of this level (remember: flagship CPU with flagship motherboard) one who invest that amount of money it does it to make the best use of it from the very first moment.
chessmbs chessmbs 11/20/2017 06:39
@middlewave
@daniel7472

"Hyper-threading
If your computer supports hyper-threading it is usually recommended not using more threads than physical cores. The additional hyper-threads will yield about 25% to 30% extra node speed, but the inefficiency of the parallel alpha-beta search with the higher number of threads will partially offset this speed gain. This means that the extra hyper-threads will produce only a small increase in Elo – probably at most 10 Elo.

To illustrate this, in a 12 vs 24 threads test match on a 12-core computer (Intel Xeon processor) the outcome after 1500 games was +7 ±10 Elo in favor of the 24-thread engine. Even with 1500 games played the measured Elo difference was still inside the error margin!

If your CPU can be overclocked it may actually be wiser not to use hyper-threading. By not using the hyper-threading you will reduce the thermal load of the CPU which will allow you to reach a higher overclocking frequency."

excerpt from the Houdini 6 manual Cores and Threads Management section.
middlewave middlewave 11/20/2017 05:43
@chessmbs
However, there is one other aspect about Albert's RAM configuration which should be taken into account: populating the RAM slots with fewer DIMMs gives you more flexibility and freedom for future upgrades, eg. if you later wish to go to 64Gb or beyond! I know most X399 motherboards have 8 RAM slots, but still, if you later decide to increase RAM you will have to replace the existing DIMMs to achieve quad-channel...
So, yes, a quad-channel setup is optimal for Threadripper, but on the other hand Albert's choice is more future-proof!
middlewave middlewave 11/20/2017 03:46
@daniel7472
It depends on the engine you are using, and also on the CPU architecture. Until recently it was judged that HyperThreading not only didn't provide any benefits, but would actually slow the engine down! But recent advances in CPU tech and also engine programming have provided examples where HT will actually cause no harm, and even provide a small boost.It all depends on the specific use case.
I am not an expert on this matter, this is just a general overview. If you want a (very general) piece of advice: switch HT off (works better in most situations). But this is very vague :-)
middlewave middlewave 11/20/2017 03:35
@chessmbs
Although I am not a hardware expert, and certainly not on AMD's chips, I think the same about RAM.
Threadripper design puts a lot of emphasis on RAM speed, because RAM operates the "Infinity Fabric" tech that interconnects the two processor dies (1950X is NOT a monolithic die with 16 cores, but two 8-core Ryzen dies that interconnect).
Thus, CPU performance should be affected by the speed of the entire RAM setup: fast RAM chips greatly improve performance, and a quad-channel setup (as suggested by chessmbs) will (at least in theory) be faster than a dual-channel one.
When using Intel chips, RAM speed will only play a minor role, because the technology interconnecting the multiple cores is different and not RAM-dependent (the interconnecting mesh runs at a stable frequency regardless of RAM speed). But with Threadripper things are entirely different!
chessmbs chessmbs 11/20/2017 09:41
Hi. without entering in details regarding the components of the rig, there is one MAJOR shortcoming with this build, and derives from the lack of understanding what the architecture of the TR4 socket / x399 platform + the Threadripper CPU brings in to the game...and this shortcoming is even more important chess wise, as it would halve the performance of any engine that is run on it (tested with the exactly same Komodo benchmark)!
And this is related to the memory used, which says: 32GB (2 x 16GB) Big mistake! This rig will run into a serious bottleneck, caused by using only dual channel whereas one of the major pluses for the TR is the Quad Channel architectural approach toward memory management. Therefor it is a matter of utmost importance to use 4 memory modules with this mobos! So, for a 32GB total memory, one should use a 4 x 8GB quad channel kit.
Just my 2 cents...
daniel7472 daniel7472 11/20/2017 07:37
Hi,
is there any benefit from hyperthreading?
Bertman Bertman 11/20/2017 07:09
@Crispy - It is in the plans and will appear in the next couple of days.
crispyambulance crispyambulance 11/19/2017 10:57
Albert

Can you suggest a setup at a cheaper price point for the less wealthy readers? eg 1500$ and 1000$

Thanks
therook1357 therook1357 11/18/2017 11:11
what happened to books,there cheaper!!
Bertman Bertman 11/18/2017 09:58
@Aighearach - One other thing: I do not think you understood the point of the two separate SSDs. RAID is not going to solve the problem at all of tablebase reads not being affected by software or OS reads at the same time.
Bertman Bertman 11/18/2017 09:21
@Aighearch - If you read the notes, you will note that the video card is to be able to support multiple displays. As a gaming card it is pretty weak, and its power draw is negligible. That said, you are correct that a more modest PSU is quite doable. As to the case, I don't see how a $20 case is going to help prevent the CPU from melting down, or provide frontal USB3 or even fit the water-cooler.
Bertman Bertman 11/18/2017 09:06
@ShackAhlen - According to Hardware.fr (http://www.hardware.fr/articles/967-11/ia-echecs-stockfish-komodo.html) from the opening position the average is about 28 million NPS for Komodo, and close to 31 million NPS for Stockfish.
Aighearach Aighearach 11/18/2017 09:02
I think it is funny you include a $120 video card. LOL It looks like the machine was chosen for playing video games between rounds, not a machine for a chess professional! You could also spend $20 on a case instead of $100 since it didn't even come with a power supply, and once you've dropped the gaming video card you no longer need a $130 power supply! You can get a high quality super-quiet one for $85 if you're not powering that video monster.

I'd also not believe that 2 SSDs with one constantly loaded would be faster than two of the same size as a RAID. I've been building workstations most of my life, and I've just never seen that be true; normally when you split the drives that way it is because the cheap one is a lot cheaper and they're not compatible to run as RAID. Here where they are the same technology, only a benchmark running your actual software would be believable; the mere argument falls flat. Maybe spend the savings from the video card on a second 500G HD if 500 is really needed, and then use RAID 0. But you could also get 2 of the 250s and then a regular cheap HD for the OS, and it would still cost less, and the performance should be a lot higher!
SambalOelek SambalOelek 11/18/2017 08:54
@chessSpawn49 you need a coffee or a rest?
SchackAhlen SchackAhlen 11/18/2017 08:52
What would the typical nps be ( at the opening phase) be for the mentioned setup?
ChessSpawn49 ChessSpawn49 11/18/2017 08:24
Yawn.........
SambalOelek SambalOelek 11/18/2017 08:11
Are u guys using windows or LINUX? probably Chessbase doesn't support LInux...

Might as well make all open source...

Anyway, more calculations, more gimmick...not needed.

Just do the old ways, without pc. Fischer did it. So did Capablanca.

Do we all need to be come chess Cyborgs?
John Timmel John Timmel 11/18/2017 07:55
Can anyone offer pricing examples of implementing a parallel chess engine and database (Chessbase)
on any of the elastic/flexible cloud service such as from Google, Microsoft Azure, Amazon, IBM, etc etc.?
These can be a pay per use, and/or dedicated to varying degrees, and can dynamically grow or shrink depending on load. Fritz and Chessbase Docker images would be quite interesting.
Bertman Bertman 11/18/2017 07:53
@Nordlandia - You're mistaken on a few accounts: Syszygy is not two times smaller than Nalimov it is over 6 times smaller. Also, it does not eat RAM as you suggest. the full set of 6-piece Syzygy takes up 160GB, and a curated set takes less, In fact, personally I keep the full set on a 128GB USB3 pendrive that I use on whatever computer I am analyzing with. The suggestion of a 256GB SSD unit is already larger than necessary to hold the full set. The choice of a PCIe drive is also to ensure maximum read speeds, since it is at least 6-8 times faster than a standard SSD drive, depending on the reference of course.
Nordlandia Nordlandia 11/16/2017 03:41
I'd rather go for Samsung 960 PRO 1TB M.2 for tablebases. New syzygy depth to mate is currently being generated at ~ 50% size of nalimov. This renders nalimov format obsolete.

Besides 128 Gb is enough to cache 6-men WDL in RAM. 32 GB tend to cause paging, which downgrade performance considerably. One solution for 32 GB users is to disable page file (swap file), remember syzygy eat RAM.
1