Skynet DB (private tests).

[Skynet]

Greetings to all!
Below is a link with games (pgn file) with all my private tests about 4200 clean games (and some games without results = Line). All tests was made with TC: 6min+2sec and 3 cores per engine.
https://krakenfiles.com/view/ed94d4f543/file.html

[Sedat Canbaz]

Hello Aleks,

Many thanks...thousands of games..
And not many are sharing such useful data (especially games on same PC)

However,
After checking your database ... I noticed several games lost on time
Even at 6m+2s

So I just wonder...
Can you give us more details about your played conditions:
- Hardware ?
- Windows or Linux or other operating system ?
- Ponder ON/OFF ?
- Hyper-Threading ON/OFF ?
- Hash-Table Size ?
- Move Overhead is default ?
- GUI ?

Btw, (in SCCT)
I never noticed any single game lost on time (in latest tours via SF based engines)
Even at 3 minutes...sure this is my tournament PC, which seems to be quite stable...
But not sure about others used conditions, systems..for example yours...

And now again the same story came in my mind (for anyone who missed)
In the past.. (in the early of 2000 years)
I had 2 assistants in Antalya (they were not so much experienced)
In short, I installed the same engines on their PCs,
I mean by myself prepared the same tour configurations (as in mine)
And after 1-2 weeks...I checked their played games...
One of the assistant's database was terrible...many games lost on time etc
I asked him why all this happened... because all same...exc. testers..?
He replayed: I was playing other online non -chess games (during active tours)

Sure...I do not mean that you are as beginner etc..
But it seems there is something, which is not so stable in your test...

Best,
Sedat

[Skynet]

Can you give us more details about your played conditions:
- Hardware ?
- Windows or Linux or other operating system ?
- Ponder ON/OFF ?
- Hyper-Threading ON/OFF ?
- Hash-Table Size ?
- Move Overhead is default ?
- GUI ?

Hardware: AMD Athlon X4 870K - 3.8 GHz, Quad Core, 16 GB RAM, Operating system: Windows 7, 64 bit
Ponder=Off / Hyper Threading=ON / Hash on CuteChess=1024 - on Fritz 16=1536 / Move Overhead=Always default
GUIs CuteChess 1.2, and Fritz 16 (which was buggy and some games ended without results=Line, and now it is uninstalled because i don't want to make any book tests anymore).

[Sedat Canbaz]

Can you give us more details about your played conditions:
- Hardware ?
- Windows or Linux or other operating system ?
- Ponder ON/OFF ?
- Hyper-Threading ON/OFF ?
- Hash-Table Size ?
- Move Overhead is default ?
- GUI ?

Hardware: AMD Athlon X4 870K - 3.8 GHz, Quad Core, 16 GB RAM, Operating system: Windows 7, 64 bit
Ponder=Off / Hyper Threading=ON / Hash on CuteChess=1024 - on Fritz 16=1536 / Move Overhead=Always default
GUIs CuteChess 1.2, and Fritz 16 (which was buggy and some games ended without results=Line, and now it is uninstalled because i don't want to make any book tests anymore).

Thanks for the info...

So...your engines are played with 3 threads (not 3 cores)
I mean when HT is disabled =Cores....when HT is enabled = Threads

But if I was in your shoes (for maximum and stable performance):
- Disable HT from BIOS
- Enable Pondering...
- Disable large pages (İf already are activated...)
- Increase Overhead to min: 100
- You can use 512 MB HASH...on my test is the best...(for Blitz + MP)

Btw, be careful with Cutechess GUI...I had bad experience with it...
Because..(in case of parallel matches) Cutechess force the engines to play from same exe
Use older Fritz GUIs, such Fritz 13 Fritz 14...e.g book learning works properly...
And run matches in Gauntlet mode or Round-Robin...but not recommended in eng mode
And do not forget, for each book player to install other copy engine...
That's all...which in my mind now...

Greetings,
Sedat

[Researcher]

Can you give us more details about your played conditions:
- Hardware ?
- Windows or Linux or other operating system ?
- Ponder ON/OFF ?
- Hyper-Threading ON/OFF ?
- Hash-Table Size ?
- Move Overhead is default ?
- GUI ?

Hardware: AMD Athlon X4 870K - 3.8 GHz, Quad Core, 16 GB RAM, Operating system: Windows 7, 64 bit
Ponder=Off / Hyper Threading=ON / Hash on CuteChess=1024 - on Fritz 16=1536 / Move Overhead=Always default
GUIs CuteChess 1.2, and Fritz 16 (which was buggy and some games ended without results=Line, and now it is uninstalled because i don't want to make any book tests anymore).

Thanks for the info...

So...your engines are played with 3 threads (not 3 cores)
I mean when HT is disabled =Cores....when HT is enabled = Threads

But if I was in your shoes (for maximum and stable performance):
- Disable HT from BIOS
- Enable Pondering...
- Disable large pages (İf already are activated...)
- Increase Overhead to min: 100
- You can use 512 MB HASH...on my test is the best...(for Blitz + MP)

Btw, be careful with Cutechess GUI...I had bad experience with it...
Because..(in case of parallel matches) Cutechess force the engines to play from same exe
Use older Fritz GUIs, such Fritz 13 Fritz 14...e.g book learning works properly...
And run matches in Gauntlet mode or Round-Robin...but not recommended in eng mode
And do not forget, for each book player to install other copy engine...
That's all...which in my mind now...

Greetings,
Sedat

I wish to few observations here.
1. When HT is enabled or disabled, threads can be set by the user.
When HT is enabled, max. no. of threads that user can set = twice the no. of real cores.
When HT is disabled, max. no. of threads that user can set = no. of real cores.
2. Hyperthreading is good. The only issue is it generates more heat that when it is kept off. So if hyperthreading is used, we only need a cool environment for better performance.
3. Ponder ON (This will generate more heat. Same, using cool environment helps.) I requested one of the engine authors to give useful advice and I came to know that keeping ponder on interferes with the opponent engine's performance when it is not there time to move.
(I now believe, for tours results to be accurate, Ponder OFF is best and when playing online, Ponder ON is best.)
4. Large Pages are good if playing online games. For tours, large pages ON/OFF affects both engines nearly equally. Because after some games the system cannot make use of the Large Pages and falls flat on default OFF.
For enabling it again in best way, one has to restart the PC.
LP ON/OFF will not affect the accuracy of the result of the tours.
But I agree with Sedat on OFF condition because, if we keep it ON, then we cannot say that for the entire tour it was on. It gets OFF after some time and needs RAM defragmentation possible by restart of the system.
5. Overhead is right, even 200 ms is all right. I agree on this.
6. The hash table size selection is more dependent on the time controls.
One one thread, Peter has given a detailed account on this. Many thanks to him.

Best regards!!

[Skynet]

Well, in my case when the engine is using more than 1 core/thread, HT are disabled automatically (found out it today). As for the Large Pages, they are not supported by my PC.

[Skynet]

My personal database, books tests (mixed engines), tc: 3+1 and up, 3cores and up, clear pgn - duplicates removed.
https://pixeldrain.com/u/82KkVY3s

[SkyNet]

Hello, here is 112 pgns from my personal book tests made this year, just in case if someone want to use them for creating book. TC: 2+1 and 3+1 / 3 cores per engine.
https://krakenfiles.com/view/hOvYXNxYNL/file.html

[catalano7777]

Hello, here is 112 pgns from my personal book tests made this year, just in case if someone want to use them for creating book. TC: 2+1 and 3+1 / 3 cores per engine.
https://krakenfiles.com/view/hOvYXNxYNL/file.html

Thank you so much, Skynet!!

[SkyNet]

Here is another DB from my personal book tests, made on my AMD Ryzen 7 with different TCs and number of cores, it's including 19 pgns (with at least 600+ games each).
https://krakenfiles.com/view/F3VTlqCPMa/file.html