Re: Adrian Oliva
Posted: Tue Jul 04, 2023 5:15 pm
Be prepared for vicious revenge.. 
Chess Forum
Welcome to the chess forum of massimilianogoi.com
https://chess.massimilianogoi.com/forum/
Adrian Oliva
Page 3 of 42
Re: Adrian Oliva
Posted: Tue Jul 04, 2023 5:22 pm
Hi Adrian, i want to ask you for a favor. When you'll have a time, can you test SF-PB-16_AVX vs SF-PB-16_AVX2? On my AMD Ryzen 7 1800X - AVX compilation is a bit faster (which is strange). If you can use a shorter TC, eg 3min+1sec and pgn opening book.
Re: Adrian Oliva
Posted: Fri Jul 07, 2023 11:21 am
SF PB 16 AVX vs. SF PB 16 AVX2
Processor AMD A10 – 9700 R7,10 Compute Cores 4C + 6G - 3.50 GHz - RAM 8.00 GB - 64-bit operating system - x64 processor - Windows 10 Pro - 1024 mb hash – Fritz 16 GUI – Permanent Brain: ON – Opening Book: 8 Moves v3 Opening Test Suite - Final Tablebases: Syzygy 3-4-5 pieces - Both Engines Play with 2 Threads and Large Pages – NNUE: nn-5af11540bbfe.nnue for Both Engines
DESKTOP-TORM2PL, Blitz 4min+1seg 0
1 SF PB 16 AVX +2 +10/=326/-8 50.29% 173.0/344
2 SF PB 16 AVX2 -2 +8/=326/-10 49.71% 171.0/344
Games in PGN: https://www.mediafire.com/file/bocxn595 ... 4.pgn/file
Processor AMD A10 – 9700 R7,10 Compute Cores 4C + 6G - 3.50 GHz - RAM 8.00 GB - 64-bit operating system - x64 processor - Windows 10 Pro - 1024 mb hash – Fritz 16 GUI – Permanent Brain: ON – Opening Book: 8 Moves v3 Opening Test Suite - Final Tablebases: Syzygy 3-4-5 pieces - Both Engines Play with 2 Threads and Large Pages – NNUE: nn-5af11540bbfe.nnue for Both Engines
DESKTOP-TORM2PL, Blitz 4min+1seg 0
1 SF PB 16 AVX +2 +10/=326/-8 50.29% 173.0/344
2 SF PB 16 AVX2 -2 +8/=326/-10 49.71% 171.0/344
Games in PGN: https://www.mediafire.com/file/bocxn595 ... 4.pgn/file
Re: Adrian Oliva
Posted: Fri Jul 07, 2023 2:34 pm
Hi Adrian, many thanks! That's what i wanted to confirm, of course the number of games is not that big but it looks like AVX is better than AVX2, i also had better results in favor of AVX on my AMD Ryzen 7 1800X (with short time control).Adrian Oliva wrote: ↑Fri Jul 07, 2023 11:21 am
1 SF PB 16 AVX +2 +10/=326/-8 50.29% 173.0/344
2 SF PB 16 AVX2 -2 +8/=326/-10 49.71% 171.0/344
Re: Adrian Oliva
Posted: Sat Jul 08, 2023 1:28 pm
Hi Alex! I am glad that you can clarify your doubts. I was always recommended to use AVX2 versions for my PC, but from your experience, AVX seems to work better. Have a very good weekend everyone! GreetingsSkynet wrote: ↑Fri Jul 07, 2023 2:34 pmHi Adrian, many thanks! That's what i wanted to confirm, of course the number of games is not that big but it looks like AVX is better than AVX2, i also had better results in favor of AVX on my AMD Ryzen 7 1800X (with short time control).Adrian Oliva wrote: ↑Fri Jul 07, 2023 11:21 am
1 SF PB 16 AVX +2 +10/=326/-8 50.29% 173.0/344
2 SF PB 16 AVX2 -2 +8/=326/-10 49.71% 171.0/344
Re: Adrian Oliva
Posted: Sat Jul 08, 2023 3:09 pm
SF PB 16 AVX where engıne lınk
thanks
thanks
Re: Adrian Oliva
Posted: Sat Jul 08, 2023 4:03 pm
Hello Adrian,Adrian Oliva wrote: ↑Sat Jul 08, 2023 1:28 pmHi Alex! I am glad that you can clarify your doubts. I was always recommended to use AVX2 versions for my PC, but from your experience, AVX seems to work better. Have a very good weekend everyone! GreetingsSkynet wrote: ↑Fri Jul 07, 2023 2:34 pmHi Adrian, many thanks! That's what i wanted to confirm, of course the number of games is not that big but it looks like AVX is better than AVX2, i also had better results in favor of AVX on my AMD Ryzen 7 1800X (with short time control).Adrian Oliva wrote: ↑Fri Jul 07, 2023 11:21 am
1 SF PB 16 AVX +2 +10/=326/-8 50.29% 173.0/344
2 SF PB 16 AVX2 -2 +8/=326/-10 49.71% 171.0/344
A useful testing, I like it!
Meanwhile, can you inform me please:
Large Pages are enabled on your machine ?
And Hyper-Threading (SMT) is enabled too ?
Why I ask these questions,
Depending on various hardwares, configurations etc
The used conditions may down/damage our testings..
E.g in the past..I noticed that (in one of my older machines)
When Hyper-Threading was as ON + Ponder ON suffered...
I mean some programs produced unexpected, strange results!
For this reason, if running long-matches+Ponder ON on same machine
I suggest to be disabled HT...plus LP to be disabled as well
Anyhow, after checking your test games,
We see that the results are almost identical,
I mean that we can not say AVX2 is weaker..
Btw, 344 games data is good, but not enough
To say exactly that AVX is much stronger...
Meanwhile, for anyone missed, interested...
I have many compile/binary testings as AVX2, BMI2 etc :
https://sites.google.com/site/computers ... g-strength
One thing more, during my beta SF-PB benchmark testings:
https://sites.google.com/site/computers ... benchmarks
I found that AVX2 is best/solved faster on my latest machine (2x Epyc 7B12)
On other hand, actually the shocking news was that,
Stockfish 10 x64 or even 32-bit( for older processors), performed even better on some hardwares
https://sites.google.com/site/computers ... benchmarks
As conclusion,
Any engine compile's speed depends on author, hardware, testings etc.!
Just my 2 cents over this issue )
Greetings
Re: Adrian Oliva
Posted: Sat Jul 08, 2023 4:13 pm
Re: Adrian Oliva
Posted: Sat Jul 08, 2023 5:30 pm
thanks
Re: Adrian Oliva
Posted: Sat Jul 08, 2023 7:11 pm
Sedat Canbaz wrote: ↑Sat Jul 08, 2023 4:03 pmHello Adrian,Adrian Oliva wrote: ↑Sat Jul 08, 2023 1:28 pmHi Alex! I am glad that you can clarify your doubts. I was always recommended to use AVX2 versions for my PC, but from your experience, AVX seems to work better. Have a very good weekend everyone! GreetingsSkynet wrote: ↑Fri Jul 07, 2023 2:34 pmHi Adrian, many thanks! That's what i wanted to confirm, of course the number of games is not that big but it looks like AVX is better than AVX2, i also had better results in favor of AVX on my AMD Ryzen 7 1800X (with short time control).Adrian Oliva wrote: ↑Fri Jul 07, 2023 11:21 am
1 SF PB 16 AVX +2 +10/=326/-8 50.29% 173.0/344
2 SF PB 16 AVX2 -2 +8/=326/-10 49.71% 171.0/344
A useful testing, I like it!
Meanwhile, can you inform me please:
Large Pages are enabled on your machine ?
And Hyper-Threading (SMT) is enabled too ?
Why I ask these questions,
Depending on various hardwares, configurations etc
The used conditions may down/damage our testings..
E.g in the past..I noticed that (in one of my older machines)
When Hyper-Threading was as ON + Ponder ON suffered...
I mean some programs produced unexpected, strange results!
For this reason, if running long-matches+Ponder ON on same machine
I suggest to be disabled HT...plus LP to be disabled as well
Anyhow, after checking your test games,
We see that the results are almost identical,
I mean that we can not say AVX2 is weaker..
Btw, 344 games data is good, but not enough
To say exactly that AVX is much stronger...
Meanwhile, for anyone missed, interested...
I have many compile/binary testings as AVX2, BMI2 etc :
https://sites.google.com/site/computers ... g-strength
One thing more, during my beta SF-PB benchmark testings:
https://sites.google.com/site/computers ... benchmarks
I found that AVX2 is best/solved faster on my latest machine (2x Epyc 7B12)
On other hand, actually the shocking news was that,
Stockfish 10 x64 or even 32-bit( for older processors), performed even better on some hardwares
https://sites.google.com/site/computers ... benchmarks
As conclusion,
Any engine compile's speed depends on author, hardware, testings etc.!
Just my 2 cents over this issue )
Hello Sedat! As always, thanks for your invaluable comments. I do the tests with Permanent Brain ON and LP ON.
Greetings
Greetings