SF8 has been released on abrok

[Frank Quisinsky]

Hi Gabor,

thats right.

Engines so strong as SF are not easy to test, also with a lot of opponents and many games.

But ASM SF is available and today I am thinking ...
Why not!

Easy to see that the version is faster ...

Only the name ... ASM Fish ... I don't like!

Right is more ...
Stockfish ... date ... ASM x64!

Think so!
But I am not sure, I will test the version the Stockfish team like to see. But if I get here no information I will test ASM because I am thinking the best one should be available in a rating list.

Best
Frank

[corres]

[quote="APassionForCriminalJustic"][quote="Lyudmil Tsvetkov"]Not released for me, until I see a post here by main developers with link to an official download.

I wonder when people will learn that faster compiles/asm versions bring almost nothing at LTC. They migth perform well at very fast TC, but, as tests have already shown, at LTC the benefit is non-existing. I would prefer a single functional change instead.[/quote]

Really? So why don't you go use some ancient-slow PC with Stockfish and see if you benefit from using it. Nah, actually, why do you not run the latest Stockfish development version on only one core. Forget about your eight-core or quad-core CPU. Speed does not matter right? What an absolutely ridiculous argument that fundamentally makes NO sense. Speed actually does matter. It doesn't matter if it is STC, LTC, or VLTC. Where one can improve the speed of an engine you would always choose that route. It's called progression. The question should be turned around to say, "When will people like yourself quit being in denial, and actually use their brain?!" If you can get millions of more positions searched per second even if the time control is 120 minutes per side the advantage will always lie with that engine that can search effectively more positions. Now I mean this to say two engines that are EXACTLY the same - but one is a lot faster than the other. Obviously you should not compare Komodo versus Stockfish in NPS differentials. In essence, I would prefer both the functional patches - and the speed improvements. Mohammed's asmFish is the best of both worlds. You get the functional patches plus an extra 20+ percent of speed over the Abrok compiles at absolutely no cost. What makes logical sense to you? The same goes with using the Ultimaiq compiles versus the Abrok compiles. I would take that extra eight percent of speed because it comes at NO COST.[/quote]

If the source code is the same totally the more speed the more Elo.
In the case of a given source the speed depends on the compiler used and the settings of that compiler.
Moreover there are lot of tricks to enhance program speed but these involve some modifications of code.
But the code of asmfish differs from "official" Stockfish in some measure.
Because there is no absolute precise translation from C++ to assembly code. So the higher speed does not call forth higher Elo in automatic mode. To prove this a lot of tests need with STC, LTC and VLTC.

[Lyudmil Tsvetkov]

Not released for me, until I see a post here by main developers with link to an official download.

I wonder when people will learn that faster compiles/asm versions bring almost nothing at LTC. They migth perform well at very fast TC, but, as tests have already shown, at LTC the benefit is non-existing. I would prefer a single functional change instead.

Really? So why don't you go use some ancient-slow PC with Stockfish and see if you benefit from using it. Nah, actually, why do you not run the latest Stockfish development version on only one core. Forget about your eight-core or quad-core CPU. Speed does not matter right? What an absolutely ridiculous argument that fundamentally makes NO sense. Speed actually does matter. It doesn't matter if it is STC, LTC, or VLTC. Where one can improve the speed of an engine you would always choose that route. It's called progression. The question should be turned around to say, "When will people like yourself quit being in denial, and actually use their brain?!" If you can get millions of more positions searched per second even if the time control is 120 minutes per side the advantage will always lie with that engine that can search effectively more positions. Now I mean this to say two engines that are EXACTLY the same - but one is a lot faster than the other. Obviously you should not compare Komodo versus Stockfish in NPS differentials. In essence, I would prefer both the functional patches - and the speed improvements. Mohammed's asmFish is the best of both worlds. You get the functional patches plus an extra 20+ percent of speed over the Abrok compiles at absolutely no cost. What makes logical sense to you? The same goes with using the Ultimaiq compiles versus the Abrok compiles. I would take that extra eight percent of speed because it comes at NO COST.

If the source code is the same totally the more speed the more Elo.
In the case of a given source the speed depends on the compiler used and the settings of that compiler.
Moreover there are lot of tricks to enhance program speed but these involve some modifications of code.
But the code of asmfish differs from "official" Stockfish in some measure.
Because there is no absolute precise translation from C++ to assembly code. So the higher speed does not call forth higher Elo in automatic mode. To prove this a lot of tests need with STC, LTC and VLTC.

Why do not you stop all this nonsense?

If someone has been reading the tournaments and matches thread, Pierre Dominici ran a match of 84 games between asm and SF, and the score stood at 42.5:41.5 for asm, 30 min per game.

So, the 20 elo or so measured by Stefan at 1 min. drop to about 1 elo or so.
You are not interested in testing a version that is only 1 elo stronger, are you?

Besides, I get headache from all those versions that are identically the same: SF,asm-fish, brain-fish,...I already forgot which one is the real SF and which are the derivatives.

[syzygy]

But the code of asmfish differs from "official" Stockfish in some measure.

Not if you take pedantFish instead of asmFish. PedantFish produces the exact same node counts as Stockfish (or it will, once it has been updated to SF8).

Because there is no absolute precise translation from C++ to assembly code.

What does a C++ compiler do? Translate from C++ to assembly (and then run the result through an assembler to obtain object code).

There is no single translation from C++ to assembly and indeed every compiler and every compiler version and every different set of compiler options will result in different object code, but the behaviour of the program should be exactly the same. In the absence of compiler bugs, the only difference is speed. And the same holds for pedantFish.

And of course any speed increase of a functionally identical program will translate in better chess, whether at STC or at LTC or at VLTC. 8.5% extra speed makes the faster program get as much out of 60 minutes of thinking time as the slower program gets out of a little over 65 minutes. Anybody who certainly understands this but nevertheless denies it, is just looking for attention (and as always, the best thing is to just ignore it).

[Werewolf]

Ronald,

Which do you prefer, asmfish or Pedantfish?

Can you explain why.

[tmokonen]

Official compiles, courtesy of Kiran Panditrao of the FishCooking forum, have now been uploaded to https://stockfishchess.org/

No 32 bit, Linux or Mac compiles as of yet.

[IWB]

I really hope the rating sites test the asm version when it becomes available.

They never test the asm version, because they are ways to old fashioned .

In their heads the asm version is a not-illegal unofficial clone.

This is my personal opinion about cegt and ipon.

I was always unsure about what's going on in my head. Good to know that once and for all, thx.

Ingo

[Leto]

Official Stockfish 8 POPCNT vs Brainfish 103016 NUMA POPCNT, starting position, 3 minute infinite analysis, AMD FX8300 8 core:


SF8 POPCNT: 3 minutes, d31/43 1.d4, 5471kn/s
Brainfish 301016 NUMA POPCNT: 3 minutes, d31/39, 1.d4 5849kn/s

[corres]

[quote="syzygy"][quote="corres"]But the code of asmfish differs from "official" Stockfish in some measure.[/quote]
Not if you take pedantFish instead of asmFish. PedantFish produces the exact same node counts as Stockfish (or it will, once it has been updated to SF8).

[quote]Because there is no absolute precise translation from C++ to assembly code.[/quote]
What does a C++ compiler do? Translate from C++ to assembly (and then run the result through an assembler to obtain object code).

There is no single translation from C++ to assembly and indeed every compiler and every compiler version and every different set of compiler options will result in different object code, but the behaviour of the program should be exactly the same. In the absence of compiler bugs, the only difference is speed. And the same holds for pedantFish.

And of course any speed increase of a functionally identical program will translate in better chess, whether at STC or at LTC or at VLTC. 8.5% extra speed makes the faster program get as much out of 60 minutes of thinking time as the slower program gets out of a little over 65 minutes. Anybody who certainly understands this but nevertheless denies it, is just looking for attention (and as always, the best thing is to just ignore it).[/quote]

Thanks for the detailed answer.
I do not know the difference between asmfish and pedantfish.
I think that the parameters of a compiler affect not only to the speed but the behavior of the engine. Because the different parameters affect the different part of the code with different mode.

[syzygy]

I do not know the difference between asmfish and pedantfish.

So now you know: pedantFish does exactly what Stockfish does.

AsmFish does a few things differently to get slightly better speed.

I think that the parameters of a compiler affect not only to the speed but the behavior of the engine. Because the different parameters affect the different part of the code with different mode.

They do not affect the behaviour of the program. The compiler produces a binary that does exactly what the source code says it should do. Otherwise the compiler is buggy.