buying a new computer

[jp]

You sound focused on the results of games. We're not debating whether a human can beat an engine.

I was just waiting for an NN engine to play in its game a true good opening novelty that human GMs would then adopt. I don't think it has yet. I don't think NN engines have contributed to opening theory (yet) any more than traditional engines.

Engines play much stronger chess than humans so why does it matter if humans fail to adopt engine novelties. I think it's completely irrelevant. When I play ICCF games I don't look at GM games to help me with opening theory. I used to, but found so many errors that I started ignoring human played games. When I use any games as references (which isn't often) they are high level ICCF games and even then I double check every move before I consider following a line of play because many of these have errors in them.

Human-approved latest opening theory is the best opening theory we have. If humans don't adopt engine novelties, their judgement is that those novelties are bad. Do you really not trust the World Champion and super-GMs to be able to judge good and bad novelties?

But I can step back and ask anyone to show me any opening moves by NN engines that are "opening novelties", without worrying about whether they are good and bad. Have the NN engines played any original opening moves, good or bad?

Again, that would be interesting. We could then see whether we thought they were good or bad.

[jp]

Engines play much stronger than any humans do and I don't see any engines voluntarily adopting human novelties.

Of course engines aren't voluntarily adopting human novelties.

Engines don't do anything voluntarily because they don't think. They don't have intelligence.
Humans either give them opening books or don't. The engines don't have any say in it.

If engines involuntarily "play" human openings by being forced by humans to use a human-chosen opening book, they will be harder to beat. An engine with a good book is much harder to beat than a bookless engine. This has been demonstrated countless times, including by people on this forum.

They play stronger than humans by brute-force calculation. That doesn't mean that they on their own can teach humans how to play the opening.

[jp]

Human opening theory is almost irrelevant to modern engine play even though I think engines suck at opening play as compared to their middle game play. AND human opening theory could be made completely irrelevant if learning books were widely used in engines.

This sounds like stuff debated in this thread:
viewtopic.php?f=2&t=71033

Do you agree with what Ovyron says in that thread, esp. in the last few pages? (Maybe you could comment in that thread.)

[zullil]

Human opening theory is almost irrelevant to modern engine play even though I think engines suck at opening play as compared to their middle game play. AND human opening theory could be made completely irrelevant if learning books were widely used in engines.

This sounds like stuff debated in this thread:
viewtopic.php?f=2&t=71033

Do you agree with what Ovyron says in that thread, esp. in the last few pages? (Maybe you could comment in that thread.)

Ovyron says many things, like "did you know white's Giuoco Piano has been busted? By black? That's amazing! I can beat in my 10 years old computer someone with the latest thing if they play the Italian as white"

Getting confirmation is impossible.

[jp]

Well, it sounds like many CC players think OTB GMs are blundering fools even in their favorite prepared and memorized openings.
e.g.

When I play ICCF games I don't look at GM games to help me with opening theory. I used to, but found so many errors that I started ignoring human played games. When I use any games as references (which isn't often) they are high level ICCF games and even then I double check every move before I consider following a line of play because many of these have errors in them.

[jdart]

Human opening theory is almost irrelevant to modern engine play even though I think engines suck at opening play as compared to their middle game play. AND human opening theory could be made completely irrelevant if learning books were widely used in engines.

What we call human opening theory these days is heavily computer-assisted, including I am sure some strong players who are using Lc0. Humans with computer aid continue to find important novelties. This includes correspondence players who have days to find a move and can use a very large CPU time allotment on their computers. Many of these moves will not be found by computer search at ordinary tournament time controls, so having a "book" with known good moves is still an important asset for computer players IMO.

--Jon

[Zenmastur]

Engines play much stronger than any humans do and I don't see any engines voluntarily adopting human novelties.

Of course engines aren't voluntarily adopting human novelties.

I guess you're finally start to see the light. Maybe....

Engines don't do anything voluntarily because they don't think. They don't have intelligence.
Humans either give them opening books or don't. The engines don't have any say in it.

Well you make a good point. Although, I think the reason the don't do anything voluntarily isn't because they don't think, but because they haven't been told that they could (or been shown how). But that could easily be changed.

If engines involuntarily "play" human openings by being forced by humans to use a human-chosen opening book, they will be harder to beat. An engine with a good book is much harder to beat than a bookless engine. This has been demonstrated countless times, including by people on this forum.

No argument from me on the first point. But, I will point out that it's entirely possible, even likely, that an engine that creates it's own book ( or modifies and extends a human book ) will be much harder to beat than if using a human made book or no book at all.

They play stronger than humans by brute-force calculation. That doesn't mean that they on their own can teach humans how to play the opening.

Brute-force calculations is only one aspect of computers abilities related to chess. Another aspect which hasn't seen much effective use by the computer chess community is rapid and accurate storage and retrieval of huge quantities of information. This aspect has been given almost know attention by chess programmers. Very rudimentary opening books, game collections and endgame table bases is about the extent of it. That is changing though. When you think about it a NN is simply a complicated and efficient way to store information about the game. But it's computationally intensive and not very elegant when performing certain tasks. An MAB opening book, that is kept and extended by the engine, seem like a good if not perfect fit to change the way engines play the opening and will allow them to find their own novelties. One problem is the ridiculous way certain aspects of game play are handled by chess GUI's. In fact this is a large hinderance to better engine play.

I haven't read the thread you mentioned yet. When I get Time I'll look at it.

Regards,

Zenmastur

[Gregory Owett]

Hi Everyone,
I am in the case of Daniel, I want to build a machine in a box Cooler Master HAF 932 advanced existing because the amd FX 8150 processor, that I had, is KO.

On a MSI X470 Gaming Pro, I want to put an AMD Rysen 9 3090X, and an MSI GeForce RTX 2060 Ventus OC graphics card.
As RAM: Corsair Vengeance RGB DDR4 2 x 8 GB 3466 MHz CASE 16. With a cooler Corsair Hydro Series - H100i Pro RGB.
All this, for about 1,500 euros.

What do you think of this configuration? Can it work properly?
Thanks for some advice!

The X470 Gaming pro is OK, BUT it doesn't have Bios flashback capability so you should have your vendor flash the bios with one that is compatible with the Ryzen 3000 series of chips before they ship it to you. It would suck if you got it and it wouldn't boot your CPU due to an older bios being installed.

The ram you picked is a NON-standard speed. While it should work fine, generally non-standard speed ram is more expensive than more standard speeds like DDR4 3200 or 3600 and they will often have unusual or slower timings. So, you should double check the prices of both DDR4 3200 and 3600 ram sticks and compare their primary timings before purchasing these sticks. I found ones similar (not sure they are the exact same SKU) to yours for both $189.99 and $321.30 on Newegg both with 16-18-18-36 timings. That's a very wide price spread and both seem overpriced. DDR4 3200 with CL 14-14-14-35 timings will be as fast or faster and may be cheaper!

The HAF 932 is a huge case! It's also an older design that's built like a tank! Not sure why you need such a huge case but I'm sure it will work and it's designed with water cooling in mind.

The Corsair Hydro Series H100i PRO RGB is just OK. It's nothing special (depending on what you're paying for it.) Personally I avoid water coolers unless I'm doing below ambient cooling and even then I don't care for them much. I personally would go for a Noctua NH-D15S. I believe this will fit in your case and it's cheaper. The difference in cooling MAY be noticeable but unimportant in the long run as very little performance is gained by overclocking the 3900X.

Which leaves just the power supply unaccounted for.

Having said that, I don't see anything wrong with the components you listed. I'm sure they'll work just fine.

Regards,

Zenmastur

Hi Zenmastur,
Your indications are very useful! I thank you! I will apply them.
I especially hope, that my hard drive will work with the win7 system installed on it, without losing the data !
Thanks again !
The best
Gregory

[Gregory Owett]

It's better to buy DDR4 32 or 64 GB RAM.

Yes, indeed, 32 will be better!
Thanks !

[dragontamer5788]

Hi Everyone,
I am in the case of Daniel, I want to build a machine in a box Cooler Master HAF 932 advanced existing because the amd FX 8150 processor, that I had, is KO.

On a MSI X470 Gaming Pro, I want to put an AMD Rysen 9 3090X, and an MSI GeForce RTX 2060 Ventus OC graphics card.
As RAM: Corsair Vengeance RGB DDR4 2 x 8 GB 3466 MHz CASE 16. With a cooler Corsair Hydro Series - H100i Pro RGB.
All this, for about 1,500 euros.

What do you think of this configuration? Can it work properly?
Thanks for some advice!

Water cooling is stupid IMO. You need to be drawing 200W+ before water cooling is even worthwhile. Anything under 200W should be air-cooled no problem, especially by high-end air coolers.

The $100ish (USD) air coolers can cool up to 250W power consumption (although water-cooling is way better at this point). At least, there are plenty of people putting 250W TDP machines on air (aka: how many hardcore server setups are on air, in tiny 2U cases... that you know of?). Air is perfectly capable at ~250W or less TDP. BTW: The AMD Ryzen 3900x is expected to use 170W worst-case

Now if you had plans to reach 200W+ or 300W power consumption (say... are you going to overclock severely ?), then maybe you wanna go water cooling. But if you're sticking with basic system, just keep it simple with air cooling.

A good, high-end air cooler is the Noctua NH-D15. It will never leak, its pump will never fail, and algae will never grow in its water. Because air-coolers are just a simple metal heatpipe connected to a simple fan. They're practically bulletproof and dead simple to use. Water coolers have lots of issues you can run into (and if it decides to spring a leak and leak all over your system... well... you won't be happy).

Water Cooling is necessary in the highest end builds because of a steep overclock, where total-system power consumption hits over 500W average (GPUs + CPUs are all overclocked and need serious cooling)... and if you stuff a case with 2x or even 4x GPUs the air flow begins to suffer. So yeah, water-cooling has a place but you have a bog-standard computer design there. There's no reason to use water in your situation.

And if you ever decide to use water-cooling, it should be an All-copper water system to avoid galvanic corrosion, with a large, external (and reliable) water pumps, and an easily accessed reservoir for easy-maintenance. Your "cheap" $100 AIO water cooler is going to be difficult to maintain, has an undersized pump that is prone to failure, and has a copper+aluminum mix that has galvanic corrosion issues.

Don't get me wrong: the Asetek AIO water cooler (of which the Corsair Hydro is just a rebrand of) is a modern engineering marvel. But its tradeoffs are rather poor compared to quality air-cooling designs. And its a maintenance nightmare because all of its parts have been miniaturized. I'm glad that a company is out there trying to make water-cooling simplier and easier for everyone, but I don't think their design is recommend worthy quite yet.

------------

You just gotta spend some time thinking about air-flow when using an air-cooler. You want cold-air blowing into the system through the front, rising due to the PSU fan at the bottom, and then air distributed between the GPU, RAM, VRMs, and CPU cooler. Finally, the computer should exhaust all hot-air through the back and top of the case. A single, powerful, stream of cold air to cool the whole system will be more reliable AND cheaper than any water cooling system. You just gotta put some thought into it.

Make sure all your fans inside the case work with this air-flow, and you'll have a cooling system that lasts for 10 years, easily. Use Water cooling, and you will be lucky to make it to year 3 before maintaining something. (Ex: fluid refresh the biocide / anti-corrosion juices inside of the water)

What do you think of this configuration? Can it work properly?

Anyone building their own computer should run Memtest86 for 24-hours (maybe 48-hours if you're paranoid) before installing Windows. If you're building your own computer, you are going to be using your own QA work.

Computer QA is guess-and-check for the most part. Stick parts together, and they'll probably work. Modern parts have software that auto-configure pretty much everything. But you should test for hardware failures.

If you assume a 1-in-a-billion chance of failure, you have 256 failures in 32GB of RAM (remember: 32GB == 256 Gigabits). Silent-bitflips in RAM is one of the most difficult things to debug, so an "isolated" test like Memtest86 (which launches its own OS) is best for the job.

I don't have a methodology for testing hard drives / SSDs unfortunately. But RAM should absolutely be tested, and tested for 24-hours straight.

----------

The most important thing is you are buying "out of spec" RAM. JEDEC standards are only 1.2V at 2666 MHz. But the RAM you're buying will be run at 3466MHz at 1.35V. However, AMD Ryzen loves fast RAM, so you should do everything in your power to overclock the RAM. To run at 3466 MHz, you will have to configure the BIOS to use "XMP settings".

Its super easy to do, but I recommend testing the RAM if you do this. It never hurts to be too careful.

-----

Don't forget storage (I recommend NVMe SSD. HP Ex950 is quite good and cheap in the USA, but I hear its harder to source in Europe. I don't know what Europeans should buy instead). And don't forget the Windows Tax