mclane wrote: ↑Sat Dec 07, 2019 11:04 am
You think any decent programmer could make a chess engine that does less then 1 node per second and play decent chess ??
No, because Mephisto doesn't play decent chess I was saying they could make a program much stronger than Mephisto while searching as many or fewer nodes than Mephisto.
But there's no incentive, who's going to spend their time doing this? Post a $1000 bounty (whoever does it first keeps it) or something and maybe then someone will make it.
But right now the neighbors' grandma's dog already has hardware fast enough that misusing it by searching fewer nodes would be stupid.
It's like asking the martial arts world champion to fight blindfolded, with arms and legs tied to a heavy rock and ask them to fight some decrepit old guy that can barely move who was champion 40 years ago, and then complain that today's fighters don't know how to fight well...
The software progress is maybe only there because the hardware allows more resources.
Or do you think Stockfish could gain the „progress“ as you call it with 32 kb rom and 4-8 kb ram and 8 Bit ?
So you are not honest, the software progress is only possible because hardware has allowed the programmers to use hash and parallel cores and bigger memory bandwidth.
So this is not really software progress at all. Without the hardware progress this would not be possible .
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....
mclane wrote: ↑Sun Dec 08, 2019 8:34 am
How do you let modern programs run on 68030 hardware ?
I have made a cross-compiler for 68k architecture. I took Cfish 8 and compiled it for 68030 linux target using gcc.
I wanted to explore a thought experiment, that if programmers circa 1990 had known about software advances 25 years into the future, how much stronger would their programs be on their 1990 hardware against period top programs. The only problem is I do not own any 1990 top programs (or top dedicated machines) to test. I sold all my Fidelity period machines but even they were not in the top league like Mephisto.
I like these „experiments“ too.
Only I still have old machines, even my old home computers. Or all this old chess „shit“.
So let’s find out about it.
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....
I would prefer facebook messenger over Skype or whats APP chat.
Machines could be e.g.
Millennium the King that has 300 mhz arm cpu and can be reduced in mhz speed to calibrate strength.
Or fidelity on 68030 cpu.
Or Ed Schroeder chess machine 16 or 30 mhz arm
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....
mclane wrote: ↑Sun Dec 08, 2019 1:27 pm
So this is not really software progress at all. Without the hardware progress this would not be possible .
We should compare the strength of chess engines running them by hand with an abacus. Anything more is just exploiting hardware progress and should be forbidden !!! Right ?!?
Reality check : good software strives to properly use existing hardware, instead of wasting available resources.
You're like a guy who'd claim that modern 3D games, that can't run on 1985 hardware, are trash and really a low-res NES game is the pinnacle of computer graphics.
Mclane loves to artificially reduce Stockfish nodes to absurdly low amounts that can be reached in 0.001s on a weak entry-level CPU. Why not instead give Mephisto the benefit of modern hardware allowing it to run at (tens of ?) thousands of NPS ? Why not show how Mephisto scales up ?
The answer, of course, is that Mephisto is tremendously bad at using more time to play better moves.
mclane wrote: ↑Sun Dec 08, 2019 8:34 am
How do you let modern programs run on 68030 hardware ?
That's the point. New hardware capabilities lead to new software capabilities. IE Cray Blitz runs terribly on a PC. It runs like the blazes on a Cray, even if the hardware clock speeds are the same. We did a lot of cute stuff with vectors that was simply way too slow on a scalar architecture. Faster hardware allows you to get away with doing more evaluation computations without killing search depth. At 1000 NPS a program will do who knows what that is bad at that depth (IE with reductions and such, you might see a search with an effective depth of < 1 ply.
Yes, hardware has been a major boost. But it also allowed a lot of software boosts that would be impractical without the better hardware. If I were designing (re-designing) Crafty to run on a 68030 or something similar, I'd probably throw out ⅔ of what I currently do, if not more. Not to mention all the parallel search stuff. The 64 bit stuff would be somewhat problematic in terms of speed on a 2 x 32 bit word implementation. The list goes on. Probably a more accurate test would be to run Mephisto or something similar on todays hardware and compare with today's software. But that would still be inaccurate, because compromises due to hardware speed were made in Mephisto, which would not be made with today's hardware.
I was saying they could make a program much stronger than Mephisto while searching as many or fewer nodes than Mephisto.