Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.lang.prolog > #14792 > unrolled thread

WebPL is already outdated

Back to article view | Back to comp.lang.prolog

Contents

  WebPL is already outdated Mild Shock <janburse@fastmail.fm> - 2025-08-17 18:37 +0200
    Heap/Stack versus WAM resp. ZIP (Was: WebPL is already outdated) Mild Shock <janburse@fastmail.fm> - 2025-08-18 14:52 +0200
      Trealla knows Program Sharing (PS) Tricks ? (Was: Heap/Stack versus WAM resp. ZIP) Mild Shock <janburse@fastmail.fm> - 2025-08-18 15:06 +0200
        Smarter Partial Strings would use Program Sharing ['$append'/3] (Was: Trealla knows Program Sharing (PS) Tricks ?) Mild Shock <janburse@fastmail.fm> - 2025-08-18 15:42 +0200
          Static Shunting is even better [Dogelog Player] (Was: Smarter Partial Strings would use Program Sharing ['$append'/3]) Mild Shock <janburse@fastmail.fm> - 2025-08-18 15:49 +0200
    The Artificial Intelligence Flip: Acer Swift Go! (Was: WebPL is already outdated) Mild Shock <janburse@fastmail.fm> - 2025-08-31 23:56 +0200
      2025 will be last year we hear of Python (Re: The Artificial Intelligence Flip: Acer Swift Go!) Mild Shock <janburse@fastmail.fm> - 2025-09-01 00:45 +0200
        Apertus: With love, from Switzerland [02 Sept 2025] (Re: 2025 will be last year we hear of Python) Mild Shock <janburse@fastmail.fm> - 2025-09-05 00:36 +0100
        Don't try this (Was: Apertus: With love, from Switzerland [02 Sept 2025] ) Mild Shock <janburse@fastmail.fm> - 2025-09-05 01:03 +0100
        AI means ambracing Non-Linearity (Was: 2025 will be last year we hear of Python) Mild Shock <janburse@fastmail.fm> - 2025-09-19 10:01 +0200
          AI soaked PCs: Is there a Copilot+ Prolog? (Was: AI means ambracing Non-Linearity) Mild Shock <janburse@fastmail.fm> - 2025-09-19 10:10 +0200
            The morning coffee incident [Prolog Community] (Was: AI soaked PCs: Is there a Copilot+ Prolog?) Mild Shock <janburse@fastmail.fm> - 2025-09-19 14:38 +0200
              Root Cause Prediction for Your Brain (Was: The morning coffee incident [Prolog Community]) Mild Shock <janburse@fastmail.fm> - 2025-09-19 18:22 +0200
                Please delete my account and all my posts on SWI-Prolog discourse (Re: Root Cause Prediction for Your Brain) Mild Shock <janburse@fastmail.fm> - 2025-09-19 18:38 +0200
                  I will consult a Lawyer of mine (Was: Please delete my account and all my posts on SWI-Prolog discourse) Mild Shock <janburse@fastmail.fm> - 2025-09-19 18:42 +0200
    Scryer Prolog unify_with_occurs_check/2 doesn't scale (Was: WebPL is already outdated) Mild Shock <janburse@fastmail.fm> - 2025-09-19 16:08 +0200
      How bad is Rust, can JavaScript beat it? (Was: Scryer Prolog unify_with_occurs_check/2 doesn't scale) Mild Shock <janburse@fastmail.fm> - 2025-09-19 16:18 +0200
      unify_with_occurs_check/2 might have been fixed (Was: Scryer Prolog unify_with_occurs_check/2 doesn't scale) Mild Shock <janburse@fastmail.fm> - 2025-09-25 01:50 +0200
        Had to Rollback my Jaffar Unification (Was: unify_with_occurs_check/2 might have been fixed) Mild Shock <janburse@fastmail.fm> - 2025-09-25 01:59 +0200
          Trealla Prolog might apply "frozeness" to cyclic terms (Was: Had to Rollback my Jaffar Unification) Mild Shock <janburse@fastmail.fm> - 2025-09-25 02:06 +0200
            Non-intrusive through "frozen" subcategories (Was: Trealla Prolog might apply "frozeness" to cyclic terms) Mild Shock <janburse@fastmail.fm> - 2025-09-25 02:21 +0200
      Scryer Prolog occurs check cannot do hydra (Was: Scryer Prolog unify_with_occurs_check/2 doesn't scale) Mild Shock <janburse@fastmail.fm> - 2025-09-26 12:19 +0200
    WebPL and Scryer Prolog are bad examples (Was: WebPL is already outdated) Mild Shock <janburse@fastmail.fm> - 2025-10-13 09:49 +0200
      Who will win Shift-Reduce or Tabled DCG? [AI Boom] (Was: WebPL and Scryer Prolog are bad examples) Mild Shock <janburse@fastmail.fm> - 2025-10-13 15:09 +0200
    primes.pl mainly tests the Prolog ALU [mod/2 vs rem/2] (Was: WebPL is already outdated) Mild Shock <janburse@fastmail.fm> - 2025-10-15 02:38 +0200
      25-30% is insane, Neural Network Branch Prediction? (Was: primes.pl mainly tests the Prolog ALU) Mild Shock <janburse@fastmail.fm> - 2025-10-15 04:33 +0200
        NPUs (Neural Processing Units) are the new normal (Was: 25-30% is insane, Neural Network Branch Prediction?) Mild Shock <janburse@fastmail.fm> - 2025-10-15 16:04 +0200
          Ask Phind: AI inflection point right now [End 2025] (Was: NPUs (Neural Processing Units) are the new normal) Mild Shock <janburse@fastmail.fm> - 2025-10-15 16:10 +0200
            Eat Tteokbokki before SkyNet kills you [$100 ChatGPT] (Was: Ask Phind: AI inflection point right now [End 2025]) Mild Shock <janburse@fastmail.fm> - 2025-10-18 15:57 +0200
              Give Julio Di Egidio the bloody money [3 RMB¥ MiniMind] (Re: Eat Tteokbokki before SkyNet kills you [$100 ChatGPT]) Mild Shock <janburse@fastmail.fm> - 2025-10-18 16:19 +0200
                Vertex AI Training is more expensive? (Was: Give Julio Di Egidio the bloody money [3 RMB¥ MiniMind]) Mild Shock <janburse@fastmail.fm> - 2025-10-21 00:32 +0200
          The Love Affair: OpenAI and AMD (Was: NPUs (Neural Processing Units) are the new normal) Mild Shock <janburse@fastmail.fm> - 2025-10-18 18:59 +0200
          The NPU in your Browser [WebNN by W3C] (Was: NPUs (Neural Processing Units) are the new normal) Mild Shock <janburse@fastmail.fm> - 2025-10-26 08:39 +0100
            Fuzzy Alert: Boris the Loris on the Dancefloor (Was: The NPU in your Browser [WebNN by W3C]) Mild Shock <janburse@fastmail.fm> - 2025-10-26 11:33 +0100

Page 2 of 2 — ← Prev page 1 [2]

#14845 — Non-intrusive through "frozen" subcategories (Was: Trealla Prolog might apply "frozeness" to cyclic terms)

Hi,

If there would be a subcategories "acyclic"
and "cyclic" inside the "frozen" category.
One could indeed use safely a hybrid algorithm

that is non-intrusive for frozen terms, and
intrusive for non-frozen terms. Actually calling
it hybrid is a little overkill. It would just

stop at frozen terms. If it has the subcategories,
the built-in acyclic_term/1 could also stop, and
draw its result from the subcategory. This works

already in formerly Jekejeke Prolog now, but not
yet in Dogelog Player. That the rollback also
gave a 10x-20x factor slowdown for the unary

predicates is a little annonying. Must find a compromise.

Bye

Mild Shock schrieb:
> Hi,
> 
> I also tryed to measure Trealla Prolog. But the
> measurements are strange, always 0.001 secs or
> something. My suspicion is that Trealla Prolog
> 
> might apply "frozeness" to cyclic terms. A form
> of hash consing, which gives Trealla Prolog
> enough information to turn certain operations
> 
> practically into no-ops. I don't know yet how
> to proof my suspicion, and don't know how to
> deduce it from the source code.
> 
> That there are kind two types of "frozen" terms,
> acylic and cyclic, emerge a few days ago in
> formerly Jekejeke Prolog. I can represent it
> 
> inside the terms as null versus Variable[], in
> the variable spine. But I was not yet able to
> bring this feature to Dogelog Player. Because
> 
> copy_term/2 does not yet attempt a "frozenness"
> analysis. Frozen Prolog terms are only produced
> during transpilation, consult or assert,
> 
> but not yet during copy_term/2 in Dogelog Player.
> 
> Bye
> 
> Mild Shock schrieb:
>> Hi,
>>
>> The facinating result was the Jaffar Unification
>> beats Scryer Prolog even on the target JavaScript.
>> Not to speak of the Java target, which also beat it.
>>
>> But I rejected Jaffar Unification, because it
>> temporarily modifies my frozen terms, which might
>> impede some future program sharing across
>>
>> premptive threads. So I rolled back Pointer based
>> Jaffar Unification, and went back to Map Based
>> Union Find. Overall the Map and a slightly bigger
>>
>> stack incures a factor 3x slowdown. So for Java I get now:
>>
>> /* Dogelog Player 2.1.1 for Java */
>>
>> % ?- bench, bench, bench.
>> % [...]
>> % % Zeit 469 ms, GC 0 ms, Lips 42, Uhr 24.09.2025 20:00
>> % % Zeit 318 ms, GC 0 ms, Lips 62, Uhr 24.09.2025 20:00
>> % % Zeit 329 ms, GC 0 ms, Lips 60, Uhr 24.09.2025 20:00
>> % % Zeit 378 ms, GC 0 ms, Lips 52, Uhr 24.09.2025 20:00
>> % true.
>>
>> % ?- bench2, bench2, bench2.
>> % [...]
>> % % Zeit 847 ms, GC 0 ms, Lips 23, Uhr 25.09.2025 01:04
>> % % Zeit 506 ms, GC 0 ms, Lips 39, Uhr 25.09.2025 01:04
>> % % Zeit 186 ms, GC 0 ms, Lips 118, Uhr 25.09.2025 01:04
>> % % Zeit 418 ms, GC 0 ms, Lips 35, Uhr 25.09.2025 01:04
>> % true.
>>
>> In the binary predicates (bench) the factor 3x is pretty
>> much seen. But in the unary predicates (bench2) the
>> factor is much higher , something 10x - 20x. And JavaScript
>>
>> doesn't help. But this might be the price to pay for
>> a "non-intrusive" algorithm. Another name I have for my
>> current take is "non-tainting" algorithms.
>>
>> Should put a closer eye what could be done "non-intrusive",
>> or maybe device an algorithm that is a mixture of "non-
>> intrusive" and "intrucive".
>>
>> Bye
>>
>> Mild Shock schrieb:
>>> Hi,
>>>
>>> Scryer Prologs unify_with_occurs_check/2 might have
>>> been fixed. I can now test the following:
>>>
>>> /* Scryer Prolog 0.9.4-660 */
>>>
>>> % ?- bench, bench, bench.
>>> % [...]
>>> %    % CPU time: 0.148s, 57 inferences
>>> %    % CPU time: 0.126s, 57 inferences
>>> %    % CPU time: 0.214s, 58 inferences
>>> %    % CPU time: 0.213s, 58 inferences
>>> %    true.
>>>
>>> % ?- bench2, bench2, bench2.
>>> % [...]
>>> %    % CPU time: 0.036s, 58 inferences
>>> %    % CPU time: 0.042s, 58 inferences
>>> %    % CPU time: 0.018s, 59 inferences
>>> %    % CPU time: 0.096s, 56 inferences
>>> %    true.
>>>
>>> This was the test case, it includes
>>> unify_with_occurs_check/2:
>>>
>>> hydra(0, _) :- !.
>>> hydra(N, h(X, X)) :- N > 0, N0 is N-1, hydra(N0, X).
>>>
>>> hydra(0, A, A) :- !.
>>> hydra(N, h(X, X), A) :- N > 0, N0 is N-1, hydra(N0, X, A).
>>>
>>> bench :-
>>>     hydra(1048576, X), hydra(1048576, Y, Y),
>>>     time(X = Y),
>>>     time(unify_with_occurs_check(X, Y)),
>>>     time(X == Y),
>>>     time(compare(_, X, Y)), fail; true.
>>>
>>> bench2 :-
>>>     hydra(1048576, X), hydra(1048576, Y, Y),
>>>     time(copy_term(X-Y,_)),
>>>     time(term_variables(X-Y,_)),
>>>     time(\+ ground(X-Y)),
>>>     time(acyclic_term(X-Y)),
>>>     fail; true.
>>>
>>> Bye
>>>
>>> Mild Shock schrieb:
>>>> Hi,
>>>>
>>>> Since some idiots blocked me on Scryer Prolog issues,
>>>> I raise the issue here. Basically uniy_with_occurs_check/2
>>>> probably does use a different implementation of unification
>>>>
>>>> than find for (=)/2. Because it doesn't scale, I find:
>>>>
>>>> /* Scryer Prolog */
>>>> ?- test3(25).
>>>>     % CPU time: 0.001s, 57 inferences
>>>>     true.
>>>>
>>>> ?- test4(25).
>>>>     % CPU time: 2.133s, 57 inferences
>>>>     true.
>>>>
>>>> Expectation would be that unify_with_occurs_check/2
>>>> does just scale like it does in SWI-Prolog. In
>>>> SWI-Prolog I find:
>>>>
>>>> /* SWI-Prolog 9.3.30 */
>>>> ?- test3(25).
>>>> % -1 inferences, 0.000 CPU in 0.000 seconds (0% CPU, Infinite Lips)
>>>> true.
>>>>
>>>> ?- test4(25).
>>>> % -1 inferences, 0.000 CPU in 0.000 seconds (0% CPU, Infinite Lips)
>>>> true.
>>>>
>>>> The test case was simply a hydra variant. Actually the
>>>> last hydra modification posted by @kuniaki, which I
>>>> am currently ticking along now:
>>>>
>>>> hydra(0, _) :- !.
>>>> hydra(N, h(X, X)):- N>0, N0 is N-1, hydra(N0, X).
>>>>
>>>> hydra(0, A, A) :- !.
>>>> hydra(N, h(X, X), A):- N>0, N0 is N-1, hydra(N0, X, A).
>>>>
>>>> test3(N) :- hydra(N, X), hydra(N, Y, Y),
>>>>     time(X = Y).
>>>>
>>>> test4(N) :- hydra(N, X), hydra(N, Y, Y),
>>>>     time(unify_with_occurs_check(X, Y)).
>>>>
>>>> But of course there is a cut (!) in the first rules.
>>>>
>>>> Mild Shock schrieb:
>>>>> Hi,
>>>>>
>>>>> WebPL is already outdated I guess. It doesn't
>>>>> show the versions of the other Prolog systems
>>>>> it is using. While I had these results for
>>>>>
>>>>> the primes example in the WebPL playground:
>>>>>
>>>>> /* Trealla Prolog WASM */
>>>>> (23568.9ms)
>>>>>
>>>>> When I run the example here:
>>>>>
>>>>> https://php.energy/trealla.html
>>>>>
>>>>> I get better results:
>>>>>
>>>>> /* trealla-js 0.27.1 */
>>>>>
>>>>> ?- time(test).
>>>>> % Time elapsed 9.907s, 11263917 Inferences, 1.137 MLips
>>>>>
>>>>> Bye
>>>>
>>>
>>
> 

[toc] | [prev] | [next] | [standalone]

#14848 — Scryer Prolog occurs check cannot do hydra (Was: Scryer Prolog unify_with_occurs_check/2 doesn't scale)

Hi,

Was first thinking the unify_with_occurs_check/2 is
gone, when I tested this:

% ?- bench, bench, bench.
% [...]
%    % CPU time: 0.148s, 57 inferences
%    % CPU time: 0.126s, 57 inferences

But I did the test wrongly, basically the preceeding
(=)/2 did bind a variable, so that unify_with_occurs_check/2
didn't have to perform an occurs check.

If I undo the bind of a variable in (=)/2 before
going into the testing of unify_with_occurs_check/2,
I get the "bug" again:

% ?- bench, bench, bench.
%   % CPU time: 0.203s, 37 inferences
%   %%% hangs

But since ground/1 etc.. can do hydra, I suspect
the Scryer Prolog team will sooner or later figure
out how to do the occurs check so that it can

also do hyda. This is the test case now:

hydra(0, _) :- !.
hydra(N, h(X, X)) :- N > 0, N0 is N-1, hydra(N0, X).

hydra(0, A, A) :- !.
hydra(N, h(X, X), A) :- N > 0, N0 is N-1, hydra(N0, X, A).

bench :-
    hydra(1048576, X), hydra(1048576, Y, Y),
    time(\+ \+ X = Y),
    time(\+ \+ unify_with_occurs_check(X, Y)),
    time(\+ X == Y),
    time(compare(_, X, Y)), fail; true.

Bye

Mild Shock schrieb:
> Hi,
> 
> Since some idiots blocked me on Scryer Prolog issues,
> I raise the issue here. Basically uniy_with_occurs_check/2
> probably does use a different implementation of unification
> 
> than find for (=)/2. Because it doesn't scale, I find:
> 
> /* Scryer Prolog */
> ?- test3(25).
>     % CPU time: 0.001s, 57 inferences
>     true.
> 
> ?- test4(25).
>     % CPU time: 2.133s, 57 inferences
>     true.
> 
> Expectation would be that unify_with_occurs_check/2
> does just scale like it does in SWI-Prolog. In
> SWI-Prolog I find:
> 
> /* SWI-Prolog 9.3.30 */
> ?- test3(25).
> % -1 inferences, 0.000 CPU in 0.000 seconds (0% CPU, Infinite Lips)
> true.
> 
> ?- test4(25).
> % -1 inferences, 0.000 CPU in 0.000 seconds (0% CPU, Infinite Lips)
> true.
> 
> The test case was simply a hydra variant. Actually the
> last hydra modification posted by @kuniaki, which I
> am currently ticking along now:
> 
> hydra(0, _) :- !.
> hydra(N, h(X, X)):- N>0, N0 is N-1, hydra(N0, X).
> 
> hydra(0, A, A) :- !.
> hydra(N, h(X, X), A):- N>0, N0 is N-1, hydra(N0, X, A).
> 
> test3(N) :- hydra(N, X), hydra(N, Y, Y),
>     time(X = Y).
> 
> test4(N) :- hydra(N, X), hydra(N, Y, Y),
>     time(unify_with_occurs_check(X, Y)).
> 
> But of course there is a cut (!) in the first rules.
> 
> Mild Shock schrieb:
>> Hi,
>>
>> WebPL is already outdated I guess. It doesn't
>> show the versions of the other Prolog systems
>> it is using. While I had these results for
>>
>> the primes example in the WebPL playground:
>>
>> /* Trealla Prolog WASM */
>> (23568.9ms)
>>
>> When I run the example here:
>>
>> https://php.energy/trealla.html
>>
>> I get better results:
>>
>> /* trealla-js 0.27.1 */
>>
>> ?- time(test).
>> % Time elapsed 9.907s, 11263917 Inferences, 1.137 MLips
>>
>> Bye
> 

[toc] | [prev] | [next] | [standalone]

#14913 — WebPL and Scryer Prolog are bad examples (Was: WebPL is already outdated)

Hi,

Because GenX and later suffers from:

 > Somehow methods and tools to realize
 > efficient DCGs in Prolog are missing. Most
 > DCG attempts that one sees succumb
 >
 > to some declarative nonsense, creating
 > exponentially many spurious choice points,
 > you find rarely somebody mastering the Art.

Modern programmers fancy nothing else than
throwing a set of foreign library to their
Prolog system project. This is best seen in WebPL:

LALRPOP MIT/Apache-2.0 Generate the parser
https://github.com/w-henderson/WebPL/blob/main/dissertation.pdf

So there is no aim at creating a self hosting
Prolog system. There is a deep distrust in
DCGs. But why build a Prolog system that will

possibly ultimately have DCG, when you distrust
in DCGs? The second problem of GenX and later
is probably they don't know how to bootstrap

a Prolog system B via another Prolog system A.

Bye

P.S.: The result of using a Parser Tool are
often frustrating on the following levels:
- No operator table
- Directives are fixed
- Introducong DCGs need rebuid

Scryer Prolog has Operator Table, but mostlikely
used a Parser Tool some time in the project,
or programming templates borrow from Parser Tools.

Probably the worst recent example building a
Prolog system, which would have a reference for
the Parsing in Rust itself. So we have 2025

and there is not a single self hosting Prolog
yet, while all other programming languages such
as Java, golang, etc.. are self hosting.

Mild Shock schrieb:
> Hi,
> 
> WebPL is already outdated I guess. It doesn't
> show the versions of the other Prolog systems
> it is using. While I had these results for
> 
> the primes example in the WebPL playground:
> 
> /* Trealla Prolog WASM */
> (23568.9ms)
> 
> When I run the example here:
> 
> https://php.energy/trealla.html
> 
> I get better results:
> 
> /* trealla-js 0.27.1 */
> 
> ?- time(test).
> % Time elapsed 9.907s, 11263917 Inferences, 1.137 MLips
> 
> Bye

[toc] | [prev] | [next] | [standalone]

#14914 — Who will win Shift-Reduce or Tabled DCG? [AI Boom] (Was: WebPL and Scryer Prolog are bad examples)

Hi,

Maybe it is with Prolog like with Dinosaurs,
when they got extinct by a meteor crash.
All that survived were some small rodents
as the story goes. Their advantage:
- Small Size
- Burrowing Behavior
- Omnivorous Diet
- Reproductive Speed

Now Scryer Prolog uses a Shift-Reduce under
the hood, the small rodent. But it might possibly
shove their end-users heavy Tabled DCG into
the face? So that this left recursion can be solved:

expr --> expr + factor

"constraint programming" got already killed
when ILOG was bought by IBM in 2008. ILOG's
optimization solver, CPLEX, has its roots in
the CHIP (Constraint Handling in Prolog)

language, 1985 at the European Computer-Industry
Research Centre (ECRC), initially using a Prolog
language interface. So its even not a Fench product.
By the time ILOG became a commercial powerhouse,

Prolog largely disappeared from their product
codebases. There was a Transition to C++ for
Performance and Industry Adoption. I have the
gut feeling that Tabled DCG is similarly dead,

especially in the light of large languages models (LLM).
But I cannot point the figure yet perfectly
at the issues. Currently exploring the sad problem
domain of this mostlikely dead horse.

A problem could be the overkill of "Logic Grammars",
that do not tolerate incorrect texts and that cannot
be applied so easy partially. Mostlikely one has
to scrutinize the assumptions behind Tabled DCG, and

review again the possibly options beyond the beaten paths.

Bye

Mild Shock schrieb:
> Hi,
> 
> Because GenX and later suffers from:
> 
>  > Somehow methods and tools to realize
>  > efficient DCGs in Prolog are missing. Most
>  > DCG attempts that one sees succumb
>  >
>  > to some declarative nonsense, creating
>  > exponentially many spurious choice points,
>  > you find rarely somebody mastering the Art.
> 
> Modern programmers fancy nothing else than
> throwing a set of foreign library to their
> Prolog system project. This is best seen in WebPL:
> 
> LALRPOP MIT/Apache-2.0 Generate the parser
> https://github.com/w-henderson/WebPL/blob/main/dissertation.pdf
> 
> So there is no aim at creating a self hosting
> Prolog system. There is a deep distrust in
> DCGs. But why build a Prolog system that will
> 
> possibly ultimately have DCG, when you distrust
> in DCGs? The second problem of GenX and later
> is probably they don't know how to bootstrap
> 
> a Prolog system B via another Prolog system A.
> 
> Bye
> 
> P.S.: The result of using a Parser Tool are
> often frustrating on the following levels:
> - No operator table
> - Directives are fixed
> - Introducong DCGs need rebuid
> 
> Scryer Prolog has Operator Table, but mostlikely
> used a Parser Tool some time in the project,
> or programming templates borrow from Parser Tools.
> 
> Probably the worst recent example building a
> Prolog system, which would have a reference for
> the Parsing in Rust itself. So we have 2025
> 
> and there is not a single self hosting Prolog
> yet, while all other programming languages such
> as Java, golang, etc.. are self hosting.
> 
> Mild Shock schrieb:
>> Hi,
>>
>> WebPL is already outdated I guess. It doesn't
>> show the versions of the other Prolog systems
>> it is using. While I had these results for
>>
>> the primes example in the WebPL playground:
>>
>> /* Trealla Prolog WASM */
>> (23568.9ms)
>>
>> When I run the example here:
>>
>> https://php.energy/trealla.html
>>
>> I get better results:
>>
>> /* trealla-js 0.27.1 */
>>
>> ?- time(test).
>> % Time elapsed 9.907s, 11263917 Inferences, 1.137 MLips
>>
>> Bye
> 

[toc] | [prev] | [next] | [standalone]

#14915 — primes.pl mainly tests the Prolog ALU [mod/2 vs rem/2] (Was: WebPL is already outdated)

Hi,

I spent some time thinking about my primes.pl
test. And came to the conclusion that it
mainly tests the Prolog ALU. Things like

integer successor or integer modulo. Then
I found that Java has Math.floorMod() which
I wasn't using yet. And peng results are better:

/* Dogelog Player 2.1.2 for Java, today */
?- time(test).
% Zeit 286 ms, GC 1 ms, Lips 26302430, Uhr 15.10.2025 02:31
true.

Maybe the Java backend picks a CPU instruction
for Math.floorMod() instead of executing the
longer code sequence that is needed to correct

rem/2 into mod/2. Who knows. I also reorganized
the code a little bit, and eliminated an extra
method call in all arithmetic functions, by

inlining the arithmetic function body in the
evaluable predicate definition code. Comparison
to old measurements and some measurements of

other Prolog systems:

/* Dogelog Player 2.1.2 for Java, weeks ago */
?- time(test).
% Zeit 378 ms, GC 1 ms, Lips 19900780, Uhr 28.08.2025 17:44
true.

/* SWI-Prolog 9.0.4 */
?- time(test).
% 7,506,639 inferences, 0.363 CPU in 0.362 seconds
(100% CPU, 20693560 Lips)
true.

/* Scryer Prolog 0.9.4-639 */
?- time(test).
% CPU time: 0.365s, 7_517_613 inferences
true.

/* Trealla Prolog 2.82.23-3 */
?- time(test).
% Time elapsed 0.868s, 11263917 Inferences, 12.983 MLips
true.

Bye

P.S.: The code uses the hated mathematical mod/2,
and not the cheaper rem/2 that CPUs usually have:

test :-
    len(L, 1000),
    primes(L, _).

primes([], 1).
primes([J|L], J) :-
    primes(L, I),
    K is I+1,
    search(L, K, J).

search(L, I, J) :-
    mem(X, L),
    I mod X =:= 0, !,
    K is I+1,
    search(L, K, J).
search(_, I, I).

mem(X, [X|_]).
mem(X, [_|Y]) :-
    mem(X, Y).

len([], 0) :- !.
len([_|L], N) :-
    N > 0,
    M is N-1,
    len(L, M).

Mild Shock schrieb:
> Hi,
> 
> WebPL is already outdated I guess. It doesn't
> show the versions of the other Prolog systems
> it is using. While I had these results for
> 
> the primes example in the WebPL playground:
> 
> /* Trealla Prolog WASM */
> (23568.9ms)
> 
> When I run the example here:
> 
> https://php.energy/trealla.html
> 
> I get better results:
> 
> /* trealla-js 0.27.1 */
> 
> ?- time(test).
> % Time elapsed 9.907s, 11263917 Inferences, 1.137 MLips
> 
> Bye

[toc] | [prev] | [next] | [standalone]

#14916 — 25-30% is insane, Neural Network Branch Prediction? (Was: primes.pl mainly tests the Prolog ALU)

Hi,

The change from 378 ms to 286 ms is around 25-30%
is insane. But I did both tests on a novel AI CPU.
To be precise on a AMD Ryzen AI 7 350.

But somehow I picked up rumors that AI CPUs now
might do Neural Network Branch Prediction. The
idea seems to exist in hardware at least since (2012):

Machine learning and artificial intelligence are
the current hype (again). In their new Ryzen
processors, AMD advertises the Neural Net
Prediction. It turns out this is was already
used in their older (2012) Piledriver architecture
used for example in the AMD A10-4600M. It is also
present in recent Samsung processors such as the
one powering the Galaxy S7. What is it really?
https://chasethedevil.github.io/post/the_neural_network_in_your_cpu/

It can be done with Convoluted Neural Networks (CNN):

BranchNet: A Convolutional Neural Network to
Predict Hard-To-Predict Branches
To this end, Tarsa et al. proposed using convolutional
neural networks (CNNs) that are trained at
compiletime to accurately predict branches that
TAGE cannot. Given enough profiling coverage, CNNs
learn input-independent branch correlations.
https://microarch.org/micro53/papers/738300a118.pdf

Interstingly the above shows cases a PGO based
Machine Learning for Branch Predictors. No clue
how they construct the CPU, that they can feed

it with offline constructed neural neutworks for
their own execution. Maybe an optimizer uses it?
But I guess a more modern  solutions would not only

use CNN, but also an Attention Mechanism.

Bye

Mild Shock schrieb:
> Hi,
> 
> I spent some time thinking about my primes.pl
> test. And came to the conclusion that it
> mainly tests the Prolog ALU. Things like
> 
> integer successor or integer modulo. Then
> I found that Java has Math.floorMod() which
> I wasn't using yet. And peng results are better:
> 
> /* Dogelog Player 2.1.2 for Java, today */
> ?- time(test).
> % Zeit 286 ms, GC 1 ms, Lips 26302430, Uhr 15.10.2025 02:31
> true.
> 
> Maybe the Java backend picks a CPU instruction
> for Math.floorMod() instead of executing the
> longer code sequence that is needed to correct
> 
> rem/2 into mod/2. Who knows. I also reorganized
> the code a little bit, and eliminated an extra
> method call in all arithmetic functions, by
> 
> inlining the arithmetic function body in the
> evaluable predicate definition code. Comparison
> to old measurements and some measurements of
> 
> other Prolog systems:
> 
> /* Dogelog Player 2.1.2 for Java, weeks ago */
> ?- time(test).
> % Zeit 378 ms, GC 1 ms, Lips 19900780, Uhr 28.08.2025 17:44
> true.
> 
> /* SWI-Prolog 9.0.4 */
> ?- time(test).
> % 7,506,639 inferences, 0.363 CPU in 0.362 seconds
> (100% CPU, 20693560 Lips)
> true.
> 
> /* Scryer Prolog 0.9.4-639 */
> ?- time(test).
> % CPU time: 0.365s, 7_517_613 inferences
> true.
> 
> /* Trealla Prolog 2.82.23-3 */
> ?- time(test).
> % Time elapsed 0.868s, 11263917 Inferences, 12.983 MLips
> true.
> 
> Bye
> 
> P.S.: The code uses the hated mathematical mod/2,
> and not the cheaper rem/2 that CPUs usually have:
> 
> test :-
>     len(L, 1000),
>     primes(L, _).
> 
> primes([], 1).
> primes([J|L], J) :-
>     primes(L, I),
>     K is I+1,
>     search(L, K, J).
> 
> search(L, I, J) :-
>     mem(X, L),
>     I mod X =:= 0, !,
>     K is I+1,
>     search(L, K, J).
> search(_, I, I).
> 
> mem(X, [X|_]).
> mem(X, [_|Y]) :-
>     mem(X, Y).
> 
> len([], 0) :- !.
> len([_|L], N) :-
>     N > 0,
>     M is N-1,
>     len(L, M).
> 
> Mild Shock schrieb:
>> Hi,
>>
>> WebPL is already outdated I guess. It doesn't
>> show the versions of the other Prolog systems
>> it is using. While I had these results for
>>
>> the primes example in the WebPL playground:
>>
>> /* Trealla Prolog WASM */
>> (23568.9ms)
>>
>> When I run the example here:
>>
>> https://php.energy/trealla.html
>>
>> I get better results:
>>
>> /* trealla-js 0.27.1 */
>>
>> ?- time(test).
>> % Time elapsed 9.907s, 11263917 Inferences, 1.137 MLips
>>
>> Bye
> 

[toc] | [prev] | [next] | [standalone]

#14917 — NPUs (Neural Processing Units) are the new normal (Was: 25-30% is insane, Neural Network Branch Prediction?)

Hi,

It seems I am having problems pacing with
all the new fancy toys. Wasn't able to really
benchmark my NPU from a Desktop AI machine,

picked the wrong driver. Need to try again.
What worked was benchmarking Mobile AI machines.
I just grabbed Geekbench AI and some devices:

USA Fab, M4:

     sANN    hANN    qANN
iPad CPU    4848    7947    6353
iPad GPU    9752    11383    10051
iPad NPU    4873    36544    *51634*

China Fab, Snapdragon:

     sANN    hANN    qANN
Redmi CPU    1044    950    1723
Redmi GPU    480    905    737
Redmi NNAPI    205    205    469
Redmi QNN    226    226    *10221*

Speed-Up via NPU is factor 10x. See the column
qANN which means quantizised artificial neural
networks, when NPU or QNN is picked.

The mobile AI NPUs are optimized using
mimimal amounts of energy, and minimal amounts
of space squeezing (distilling) everything

into INT8 and INT4.

Bye

Mild Shock schrieb:
> Hi,
> 
> The change from 378 ms to 286 ms is around 25-30%
> is insane. But I did both tests on a novel AI CPU.
> To be precise on a AMD Ryzen AI 7 350.
> 
> But somehow I picked up rumors that AI CPUs now
> might do Neural Network Branch Prediction. The
> idea seems to exist in hardware at least since (2012):
> 
> Machine learning and artificial intelligence are
> the current hype (again). In their new Ryzen
> processors, AMD advertises the Neural Net
> Prediction. It turns out this is was already
> used in their older (2012) Piledriver architecture
> used for example in the AMD A10-4600M. It is also
> present in recent Samsung processors such as the
> one powering the Galaxy S7. What is it really?
> https://chasethedevil.github.io/post/the_neural_network_in_your_cpu/
> 
> It can be done with Convoluted Neural Networks (CNN):
> 
> BranchNet: A Convolutional Neural Network to
> Predict Hard-To-Predict Branches
> To this end, Tarsa et al. proposed using convolutional
> neural networks (CNNs) that are trained at
> compiletime to accurately predict branches that
> TAGE cannot. Given enough profiling coverage, CNNs
> learn input-independent branch correlations.
> https://microarch.org/micro53/papers/738300a118.pdf
> 
> Interstingly the above shows cases a PGO based
> Machine Learning for Branch Predictors. No clue
> how they construct the CPU, that they can feed
> 
> it with offline constructed neural neutworks for
> their own execution. Maybe an optimizer uses it?
> But I guess a more modern  solutions would not only
> 
> use CNN, but also an Attention Mechanism.
> 
> Bye
> 
> Mild Shock schrieb:
>> Hi,
>>
>> I spent some time thinking about my primes.pl
>> test. And came to the conclusion that it
>> mainly tests the Prolog ALU. Things like
>>
>> integer successor or integer modulo. Then
>> I found that Java has Math.floorMod() which
>> I wasn't using yet. And peng results are better:
>>
>> /* Dogelog Player 2.1.2 for Java, today */
>> ?- time(test).
>> % Zeit 286 ms, GC 1 ms, Lips 26302430, Uhr 15.10.2025 02:31
>> true.
>>
>> Maybe the Java backend picks a CPU instruction
>> for Math.floorMod() instead of executing the
>> longer code sequence that is needed to correct
>>
>> rem/2 into mod/2. Who knows. I also reorganized
>> the code a little bit, and eliminated an extra
>> method call in all arithmetic functions, by
>>
>> inlining the arithmetic function body in the
>> evaluable predicate definition code. Comparison
>> to old measurements and some measurements of
>>
>> other Prolog systems:
>>
>> /* Dogelog Player 2.1.2 for Java, weeks ago */
>> ?- time(test).
>> % Zeit 378 ms, GC 1 ms, Lips 19900780, Uhr 28.08.2025 17:44
>> true.
>>
>> /* SWI-Prolog 9.0.4 */
>> ?- time(test).
>> % 7,506,639 inferences, 0.363 CPU in 0.362 seconds
>> (100% CPU, 20693560 Lips)
>> true.
>>
>> /* Scryer Prolog 0.9.4-639 */
>> ?- time(test).
>> % CPU time: 0.365s, 7_517_613 inferences
>> true.
>>
>> /* Trealla Prolog 2.82.23-3 */
>> ?- time(test).
>> % Time elapsed 0.868s, 11263917 Inferences, 12.983 MLips
>> true.
>>
>> Bye
>>
>> P.S.: The code uses the hated mathematical mod/2,
>> and not the cheaper rem/2 that CPUs usually have:
>>
>> test :-
>>     len(L, 1000),
>>     primes(L, _).
>>
>> primes([], 1).
>> primes([J|L], J) :-
>>     primes(L, I),
>>     K is I+1,
>>     search(L, K, J).
>>
>> search(L, I, J) :-
>>     mem(X, L),
>>     I mod X =:= 0, !,
>>     K is I+1,
>>     search(L, K, J).
>> search(_, I, I).
>>
>> mem(X, [X|_]).
>> mem(X, [_|Y]) :-
>>     mem(X, Y).
>>
>> len([], 0) :- !.
>> len([_|L], N) :-
>>     N > 0,
>>     M is N-1,
>>     len(L, M).
>>
>> Mild Shock schrieb:
>>> Hi,
>>>
>>> WebPL is already outdated I guess. It doesn't
>>> show the versions of the other Prolog systems
>>> it is using. While I had these results for
>>>
>>> the primes example in the WebPL playground:
>>>
>>> /* Trealla Prolog WASM */
>>> (23568.9ms)
>>>
>>> When I run the example here:
>>>
>>> https://php.energy/trealla.html
>>>
>>> I get better results:
>>>
>>> /* trealla-js 0.27.1 */
>>>
>>> ?- time(test).
>>> % Time elapsed 9.907s, 11263917 Inferences, 1.137 MLips
>>>
>>> Bye
>>
> 

[toc] | [prev] | [next] | [standalone]

#14918 — Ask Phind: AI inflection point right now [End 2025] (Was: NPUs (Neural Processing Units) are the new normal)

Hi,

But not only Mobie AI and Desktop AI are making
a broader imprint now. We might also experience
Workstation AI, with a 3'000.- USD price tag:

 > You Can't Buy This... Yet! The NVIDIA GB10 from Dell
 > The New Superchip that Terrifies the Cloud!
 > https://www.youtube.com/watch?v=x1qViw4xyVo

So whats going on? I was asking Phind, which is
driven by a 70B model tailored towards developers:

Q: Is there an AI inflection point right now ,
    with NPUs in mobile, desktop and workstation

A: Evidence of the Inflection Point

- Mobile Leadership
   NPUs originated in smartphones
   Now becoming ubiquitous across all device types
   Enabling sophisticated AI features at consumer price points

- Desktop Revolution
   Major manufacturers implementing NPUs across product lines
   Apple's Neural Engine integrated into M-series chips
   Qualcomm, Intel, and AMD incorporating AI accelerators

- Workstation Transformation
   Professional-grade NPUs in mobile workstations
   Demonstrated superior performance for AI-specific tasks
   Enabling local processing of previously cloud-dependent workloads

https://www.phind.com/search/cmgs1s6jv00023h67g5z2aaa0

Bye

Mild Shock schrieb:
> Hi,
> 
> It seems I am having problems pacing with
> all the new fancy toys. Wasn't able to really
> benchmark my NPU from a Desktop AI machine,
> 
> picked the wrong driver. Need to try again.
> What worked was benchmarking Mobile AI machines.
> I just grabbed Geekbench AI and some devices:
> 
> USA Fab, M4:
> 
>      sANN    hANN    qANN
> iPad CPU    4848    7947    6353
> iPad GPU    9752    11383    10051
> iPad NPU    4873    36544    *51634*
> 
> China Fab, Snapdragon:
> 
>      sANN    hANN    qANN
> Redmi CPU    1044    950    1723
> Redmi GPU    480    905    737
> Redmi NNAPI    205    205    469
> Redmi QNN    226    226    *10221*
> 
> Speed-Up via NPU is factor 10x. See the column
> qANN which means quantizised artificial neural
> networks, when NPU or QNN is picked.
> 
> The mobile AI NPUs are optimized using
> mimimal amounts of energy, and minimal amounts
> of space squeezing (distilling) everything
> 
> into INT8 and INT4.
> 
> Bye
> 
> Mild Shock schrieb:
>> Hi,
>>
>> The change from 378 ms to 286 ms is around 25-30%
>> is insane. But I did both tests on a novel AI CPU.
>> To be precise on a AMD Ryzen AI 7 350.
>>
>> But somehow I picked up rumors that AI CPUs now
>> might do Neural Network Branch Prediction. The
>> idea seems to exist in hardware at least since (2012):
>>
>> Machine learning and artificial intelligence are
>> the current hype (again). In their new Ryzen
>> processors, AMD advertises the Neural Net
>> Prediction. It turns out this is was already
>> used in their older (2012) Piledriver architecture
>> used for example in the AMD A10-4600M. It is also
>> present in recent Samsung processors such as the
>> one powering the Galaxy S7. What is it really?
>> https://chasethedevil.github.io/post/the_neural_network_in_your_cpu/
>>
>> It can be done with Convoluted Neural Networks (CNN):
>>
>> BranchNet: A Convolutional Neural Network to
>> Predict Hard-To-Predict Branches
>> To this end, Tarsa et al. proposed using convolutional
>> neural networks (CNNs) that are trained at
>> compiletime to accurately predict branches that
>> TAGE cannot. Given enough profiling coverage, CNNs
>> learn input-independent branch correlations.
>> https://microarch.org/micro53/papers/738300a118.pdf
>>
>> Interstingly the above shows cases a PGO based
>> Machine Learning for Branch Predictors. No clue
>> how they construct the CPU, that they can feed
>>
>> it with offline constructed neural neutworks for
>> their own execution. Maybe an optimizer uses it?
>> But I guess a more modern  solutions would not only
>>
>> use CNN, but also an Attention Mechanism.
>>
>> Bye
>>
>> Mild Shock schrieb:
>>> Hi,
>>>
>>> I spent some time thinking about my primes.pl
>>> test. And came to the conclusion that it
>>> mainly tests the Prolog ALU. Things like
>>>
>>> integer successor or integer modulo. Then
>>> I found that Java has Math.floorMod() which
>>> I wasn't using yet. And peng results are better:
>>>
>>> /* Dogelog Player 2.1.2 for Java, today */
>>> ?- time(test).
>>> % Zeit 286 ms, GC 1 ms, Lips 26302430, Uhr 15.10.2025 02:31
>>> true.
>>>
>>> Maybe the Java backend picks a CPU instruction
>>> for Math.floorMod() instead of executing the
>>> longer code sequence that is needed to correct
>>>
>>> rem/2 into mod/2. Who knows. I also reorganized
>>> the code a little bit, and eliminated an extra
>>> method call in all arithmetic functions, by
>>>
>>> inlining the arithmetic function body in the
>>> evaluable predicate definition code. Comparison
>>> to old measurements and some measurements of
>>>
>>> other Prolog systems:
>>>
>>> /* Dogelog Player 2.1.2 for Java, weeks ago */
>>> ?- time(test).
>>> % Zeit 378 ms, GC 1 ms, Lips 19900780, Uhr 28.08.2025 17:44
>>> true.
>>>
>>> /* SWI-Prolog 9.0.4 */
>>> ?- time(test).
>>> % 7,506,639 inferences, 0.363 CPU in 0.362 seconds
>>> (100% CPU, 20693560 Lips)
>>> true.
>>>
>>> /* Scryer Prolog 0.9.4-639 */
>>> ?- time(test).
>>> % CPU time: 0.365s, 7_517_613 inferences
>>> true.
>>>
>>> /* Trealla Prolog 2.82.23-3 */
>>> ?- time(test).
>>> % Time elapsed 0.868s, 11263917 Inferences, 12.983 MLips
>>> true.
>>>
>>> Bye
>>>
>>> P.S.: The code uses the hated mathematical mod/2,
>>> and not the cheaper rem/2 that CPUs usually have:
>>>
>>> test :-
>>>     len(L, 1000),
>>>     primes(L, _).
>>>
>>> primes([], 1).
>>> primes([J|L], J) :-
>>>     primes(L, I),
>>>     K is I+1,
>>>     search(L, K, J).
>>>
>>> search(L, I, J) :-
>>>     mem(X, L),
>>>     I mod X =:= 0, !,
>>>     K is I+1,
>>>     search(L, K, J).
>>> search(_, I, I).
>>>
>>> mem(X, [X|_]).
>>> mem(X, [_|Y]) :-
>>>     mem(X, Y).
>>>
>>> len([], 0) :- !.
>>> len([_|L], N) :-
>>>     N > 0,
>>>     M is N-1,
>>>     len(L, M).
>>>
>>> Mild Shock schrieb:
>>>> Hi,
>>>>
>>>> WebPL is already outdated I guess. It doesn't
>>>> show the versions of the other Prolog systems
>>>> it is using. While I had these results for
>>>>
>>>> the primes example in the WebPL playground:
>>>>
>>>> /* Trealla Prolog WASM */
>>>> (23568.9ms)
>>>>
>>>> When I run the example here:
>>>>
>>>> https://php.energy/trealla.html
>>>>
>>>> I get better results:
>>>>
>>>> /* trealla-js 0.27.1 */
>>>>
>>>> ?- time(test).
>>>> % Time elapsed 9.907s, 11263917 Inferences, 1.137 MLips
>>>>
>>>> Bye
>>>
>>
> 

[toc] | [prev] | [next] | [standalone]

#14922 — Eat Tteokbokki before SkyNet kills you [$100 ChatGPT] (Was: Ask Phind: AI inflection point right now [End 2025])

Hi,

Thinks are definitively accelerating. I really would
like to use an AI that knows about all the News of today.
This bloody cut date is so annoying.

Further indicative that AI is accelerating:

In August 2025, Sam Altman dropped a bombshell:

*months, not years: Rushing GPT-6*
In August 2025, Sam Altman dropped a bombshell:
GPT-6 is already in development and coming sooner
than you think. Not in two years, but
potentially in months.
https://www.youtube.com/watch?v=44mJb5sKji0

Karpathy, coined vibe coding, released in October 2025:

*nanochat: The best ChatGPT that $100 can buy*
This repo is a full-stack implementation of an
LLM like ChatGPT in a single, clean, minimal,
hackable, dependency-lite codebase. nanochat is
designed to run on a single 8XH100 node via
scripts like speedrun.sh, that run the
entire pipeline start to end.
https://github.com/karpathy/nanochat

Bye

Mild Shock schrieb:
> Hi,
> 
> But not only Mobie AI and Desktop AI are making
> a broader imprint now. We might also experience
> Workstation AI, with a 3'000.- USD price tag:
> 
>  > You Can't Buy This... Yet! The NVIDIA GB10 from Dell
>  > The New Superchip that Terrifies the Cloud!
>  > https://www.youtube.com/watch?v=x1qViw4xyVo
> 
> So whats going on? I was asking Phind, which is
> driven by a 70B model tailored towards developers:
> 
> Q: Is there an AI inflection point right now ,
>     with NPUs in mobile, desktop and workstation
> 
> A: Evidence of the Inflection Point
> 
> - Mobile Leadership
>    NPUs originated in smartphones
>    Now becoming ubiquitous across all device types
>    Enabling sophisticated AI features at consumer price points
> 
> - Desktop Revolution
>    Major manufacturers implementing NPUs across product lines
>    Apple's Neural Engine integrated into M-series chips
>    Qualcomm, Intel, and AMD incorporating AI accelerators
> 
> - Workstation Transformation
>    Professional-grade NPUs in mobile workstations
>    Demonstrated superior performance for AI-specific tasks
>    Enabling local processing of previously cloud-dependent workloads
> 
> https://www.phind.com/search/cmgs1s6jv00023h67g5z2aaa0
> 
> Bye
> 
> Mild Shock schrieb:
>> Hi,
>>
>> It seems I am having problems pacing with
>> all the new fancy toys. Wasn't able to really
>> benchmark my NPU from a Desktop AI machine,
>>
>> picked the wrong driver. Need to try again.
>> What worked was benchmarking Mobile AI machines.
>> I just grabbed Geekbench AI and some devices:
>>
>> USA Fab, M4:
>>
>>      sANN    hANN    qANN
>> iPad CPU    4848    7947    6353
>> iPad GPU    9752    11383    10051
>> iPad NPU    4873    36544    *51634*
>>
>> China Fab, Snapdragon:
>>
>>      sANN    hANN    qANN
>> Redmi CPU    1044    950    1723
>> Redmi GPU    480    905    737
>> Redmi NNAPI    205    205    469
>> Redmi QNN    226    226    *10221*
>>
>> Speed-Up via NPU is factor 10x. See the column
>> qANN which means quantizised artificial neural
>> networks, when NPU or QNN is picked.
>>
>> The mobile AI NPUs are optimized using
>> mimimal amounts of energy, and minimal amounts
>> of space squeezing (distilling) everything
>>
>> into INT8 and INT4.
>>
>> Bye
>>
>> Mild Shock schrieb:
>>> Hi,
>>>
>>> The change from 378 ms to 286 ms is around 25-30%
>>> is insane. But I did both tests on a novel AI CPU.
>>> To be precise on a AMD Ryzen AI 7 350.
>>>
>>> But somehow I picked up rumors that AI CPUs now
>>> might do Neural Network Branch Prediction. The
>>> idea seems to exist in hardware at least since (2012):
>>>
>>> Machine learning and artificial intelligence are
>>> the current hype (again). In their new Ryzen
>>> processors, AMD advertises the Neural Net
>>> Prediction. It turns out this is was already
>>> used in their older (2012) Piledriver architecture
>>> used for example in the AMD A10-4600M. It is also
>>> present in recent Samsung processors such as the
>>> one powering the Galaxy S7. What is it really?
>>> https://chasethedevil.github.io/post/the_neural_network_in_your_cpu/
>>>
>>> It can be done with Convoluted Neural Networks (CNN):
>>>
>>> BranchNet: A Convolutional Neural Network to
>>> Predict Hard-To-Predict Branches
>>> To this end, Tarsa et al. proposed using convolutional
>>> neural networks (CNNs) that are trained at
>>> compiletime to accurately predict branches that
>>> TAGE cannot. Given enough profiling coverage, CNNs
>>> learn input-independent branch correlations.
>>> https://microarch.org/micro53/papers/738300a118.pdf
>>>
>>> Interstingly the above shows cases a PGO based
>>> Machine Learning for Branch Predictors. No clue
>>> how they construct the CPU, that they can feed
>>>
>>> it with offline constructed neural neutworks for
>>> their own execution. Maybe an optimizer uses it?
>>> But I guess a more modern  solutions would not only
>>>
>>> use CNN, but also an Attention Mechanism.
>>>
>>> Bye
>>>
>>> Mild Shock schrieb:
>>>> Hi,
>>>>
>>>> I spent some time thinking about my primes.pl
>>>> test. And came to the conclusion that it
>>>> mainly tests the Prolog ALU. Things like
>>>>
>>>> integer successor or integer modulo. Then
>>>> I found that Java has Math.floorMod() which
>>>> I wasn't using yet. And peng results are better:
>>>>
>>>> /* Dogelog Player 2.1.2 for Java, today */
>>>> ?- time(test).
>>>> % Zeit 286 ms, GC 1 ms, Lips 26302430, Uhr 15.10.2025 02:31
>>>> true.
>>>>
>>>> Maybe the Java backend picks a CPU instruction
>>>> for Math.floorMod() instead of executing the
>>>> longer code sequence that is needed to correct
>>>>
>>>> rem/2 into mod/2. Who knows. I also reorganized
>>>> the code a little bit, and eliminated an extra
>>>> method call in all arithmetic functions, by
>>>>
>>>> inlining the arithmetic function body in the
>>>> evaluable predicate definition code. Comparison
>>>> to old measurements and some measurements of
>>>>
>>>> other Prolog systems:
>>>>
>>>> /* Dogelog Player 2.1.2 for Java, weeks ago */
>>>> ?- time(test).
>>>> % Zeit 378 ms, GC 1 ms, Lips 19900780, Uhr 28.08.2025 17:44
>>>> true.
>>>>
>>>> /* SWI-Prolog 9.0.4 */
>>>> ?- time(test).
>>>> % 7,506,639 inferences, 0.363 CPU in 0.362 seconds
>>>> (100% CPU, 20693560 Lips)
>>>> true.
>>>>
>>>> /* Scryer Prolog 0.9.4-639 */
>>>> ?- time(test).
>>>> % CPU time: 0.365s, 7_517_613 inferences
>>>> true.
>>>>
>>>> /* Trealla Prolog 2.82.23-3 */
>>>> ?- time(test).
>>>> % Time elapsed 0.868s, 11263917 Inferences, 12.983 MLips
>>>> true.
>>>>
>>>> Bye
>>>>
>>>> P.S.: The code uses the hated mathematical mod/2,
>>>> and not the cheaper rem/2 that CPUs usually have:
>>>>
>>>> test :-
>>>>     len(L, 1000),
>>>>     primes(L, _).
>>>>
>>>> primes([], 1).
>>>> primes([J|L], J) :-
>>>>     primes(L, I),
>>>>     K is I+1,
>>>>     search(L, K, J).
>>>>
>>>> search(L, I, J) :-
>>>>     mem(X, L),
>>>>     I mod X =:= 0, !,
>>>>     K is I+1,
>>>>     search(L, K, J).
>>>> search(_, I, I).
>>>>
>>>> mem(X, [X|_]).
>>>> mem(X, [_|Y]) :-
>>>>     mem(X, Y).
>>>>
>>>> len([], 0) :- !.
>>>> len([_|L], N) :-
>>>>     N > 0,
>>>>     M is N-1,
>>>>     len(L, M).
>>>>
>>>> Mild Shock schrieb:
>>>>> Hi,
>>>>>
>>>>> WebPL is already outdated I guess. It doesn't
>>>>> show the versions of the other Prolog systems
>>>>> it is using. While I had these results for
>>>>>
>>>>> the primes example in the WebPL playground:
>>>>>
>>>>> /* Trealla Prolog WASM */
>>>>> (23568.9ms)
>>>>>
>>>>> When I run the example here:
>>>>>
>>>>> https://php.energy/trealla.html
>>>>>
>>>>> I get better results:
>>>>>
>>>>> /* trealla-js 0.27.1 */
>>>>>
>>>>> ?- time(test).
>>>>> % Time elapsed 9.907s, 11263917 Inferences, 1.137 MLips
>>>>>
>>>>> Bye
>>>>
>>>
>>
> 

[toc] | [prev] | [next] | [standalone]

#14923 — Give Julio Di Egidio the bloody money [3 RMB¥ MiniMind] (Re: Eat Tteokbokki before SkyNet kills you [$100 ChatGPT])

Hi,

Give Julio Di Egidio the bloody money. He is
craving for 300 USD so that he can buy the
ISO Prolog core standard. Just imagine he would want

to build a MiniMind. Just lets put some more
prespective on the current costs:

This open-source project aims to train a super-small
language model MiniMind with only 3 RMB cost and
2 hours, starting completely from scratch. The
MiniMind series is extremely lightweight, with the
smallest version being 1/7000 the size of GPT-3,
making it possible to train quickly on even the
most ordinary personal GPUs.
https://github.com/jingyaogong/minimind/blob/master/README_en.md

ChatGPT tells me that most of the numbers
are correct when you rent a GPU by the hour.
But what about a 100% ownership of a GPU for

a year. I find this might cost 12'000 USD.
One has to separate platforms for execution from
those platforms for training:

GEX44: for AI inference
Nvidia RTX™ 4000, 184 EUR / month

GEX130: for AI training
NVIDIA RTX™ 6000, 813 EUR / month
https://www.hetzner.com/dedicated-rootserver/matrix-gpu/

Bye

Mild Shock schrieb:
> Hi,
> 
> Thinks are definitively accelerating. I really would
> like to use an AI that knows about all the News of today.
> This bloody cut date is so annoying.
> 
> Further indicative that AI is accelerating:
> 
> In August 2025, Sam Altman dropped a bombshell:
> 
> *months, not years: Rushing GPT-6*
> In August 2025, Sam Altman dropped a bombshell:
> GPT-6 is already in development and coming sooner
> than you think. Not in two years, but
> potentially in months.
> https://www.youtube.com/watch?v=44mJb5sKji0
> 
> Karpathy, coined vibe coding, released in October 2025:
> 
> *nanochat: The best ChatGPT that $100 can buy*
> This repo is a full-stack implementation of an
> LLM like ChatGPT in a single, clean, minimal,
> hackable, dependency-lite codebase. nanochat is
> designed to run on a single 8XH100 node via
> scripts like speedrun.sh, that run the
> entire pipeline start to end.
> https://github.com/karpathy/nanochat
> 
> Bye
> 
> Mild Shock schrieb:
>> Hi,
>>
>> But not only Mobie AI and Desktop AI are making
>> a broader imprint now. We might also experience
>> Workstation AI, with a 3'000.- USD price tag:
>>
>>  > You Can't Buy This... Yet! The NVIDIA GB10 from Dell
>>  > The New Superchip that Terrifies the Cloud!
>>  > https://www.youtube.com/watch?v=x1qViw4xyVo
>>
>> So whats going on? I was asking Phind, which is
>> driven by a 70B model tailored towards developers:
>>
>> Q: Is there an AI inflection point right now ,
>>     with NPUs in mobile, desktop and workstation
>>
>> A: Evidence of the Inflection Point
>>
>> - Mobile Leadership
>>    NPUs originated in smartphones
>>    Now becoming ubiquitous across all device types
>>    Enabling sophisticated AI features at consumer price points
>>
>> - Desktop Revolution
>>    Major manufacturers implementing NPUs across product lines
>>    Apple's Neural Engine integrated into M-series chips
>>    Qualcomm, Intel, and AMD incorporating AI accelerators
>>
>> - Workstation Transformation
>>    Professional-grade NPUs in mobile workstations
>>    Demonstrated superior performance for AI-specific tasks
>>    Enabling local processing of previously cloud-dependent workloads
>>
>> https://www.phind.com/search/cmgs1s6jv00023h67g5z2aaa0
>>
>> Bye
>>
>> Mild Shock schrieb:
>>> Hi,
>>>
>>> It seems I am having problems pacing with
>>> all the new fancy toys. Wasn't able to really
>>> benchmark my NPU from a Desktop AI machine,
>>>
>>> picked the wrong driver. Need to try again.
>>> What worked was benchmarking Mobile AI machines.
>>> I just grabbed Geekbench AI and some devices:
>>>
>>> USA Fab, M4:
>>>
>>>      sANN    hANN    qANN
>>> iPad CPU    4848    7947    6353
>>> iPad GPU    9752    11383    10051
>>> iPad NPU    4873    36544    *51634*
>>>
>>> China Fab, Snapdragon:
>>>
>>>      sANN    hANN    qANN
>>> Redmi CPU    1044    950    1723
>>> Redmi GPU    480    905    737
>>> Redmi NNAPI    205    205    469
>>> Redmi QNN    226    226    *10221*
>>>
>>> Speed-Up via NPU is factor 10x. See the column
>>> qANN which means quantizised artificial neural
>>> networks, when NPU or QNN is picked.
>>>
>>> The mobile AI NPUs are optimized using
>>> mimimal amounts of energy, and minimal amounts
>>> of space squeezing (distilling) everything
>>>
>>> into INT8 and INT4.
>>>
>>> Bye
>>>
>>> Mild Shock schrieb:
>>>> Hi,
>>>>
>>>> The change from 378 ms to 286 ms is around 25-30%
>>>> is insane. But I did both tests on a novel AI CPU.
>>>> To be precise on a AMD Ryzen AI 7 350.
>>>>
>>>> But somehow I picked up rumors that AI CPUs now
>>>> might do Neural Network Branch Prediction. The
>>>> idea seems to exist in hardware at least since (2012):
>>>>
>>>> Machine learning and artificial intelligence are
>>>> the current hype (again). In their new Ryzen
>>>> processors, AMD advertises the Neural Net
>>>> Prediction. It turns out this is was already
>>>> used in their older (2012) Piledriver architecture
>>>> used for example in the AMD A10-4600M. It is also
>>>> present in recent Samsung processors such as the
>>>> one powering the Galaxy S7. What is it really?
>>>> https://chasethedevil.github.io/post/the_neural_network_in_your_cpu/
>>>>
>>>> It can be done with Convoluted Neural Networks (CNN):
>>>>
>>>> BranchNet: A Convolutional Neural Network to
>>>> Predict Hard-To-Predict Branches
>>>> To this end, Tarsa et al. proposed using convolutional
>>>> neural networks (CNNs) that are trained at
>>>> compiletime to accurately predict branches that
>>>> TAGE cannot. Given enough profiling coverage, CNNs
>>>> learn input-independent branch correlations.
>>>> https://microarch.org/micro53/papers/738300a118.pdf
>>>>
>>>> Interstingly the above shows cases a PGO based
>>>> Machine Learning for Branch Predictors. No clue
>>>> how they construct the CPU, that they can feed
>>>>
>>>> it with offline constructed neural neutworks for
>>>> their own execution. Maybe an optimizer uses it?
>>>> But I guess a more modern  solutions would not only
>>>>
>>>> use CNN, but also an Attention Mechanism.
>>>>
>>>> Bye
>>>>
>>>> Mild Shock schrieb:
>>>>> Hi,
>>>>>
>>>>> I spent some time thinking about my primes.pl
>>>>> test. And came to the conclusion that it
>>>>> mainly tests the Prolog ALU. Things like
>>>>>
>>>>> integer successor or integer modulo. Then
>>>>> I found that Java has Math.floorMod() which
>>>>> I wasn't using yet. And peng results are better:
>>>>>
>>>>> /* Dogelog Player 2.1.2 for Java, today */
>>>>> ?- time(test).
>>>>> % Zeit 286 ms, GC 1 ms, Lips 26302430, Uhr 15.10.2025 02:31
>>>>> true.
>>>>>
>>>>> Maybe the Java backend picks a CPU instruction
>>>>> for Math.floorMod() instead of executing the
>>>>> longer code sequence that is needed to correct
>>>>>
>>>>> rem/2 into mod/2. Who knows. I also reorganized
>>>>> the code a little bit, and eliminated an extra
>>>>> method call in all arithmetic functions, by
>>>>>
>>>>> inlining the arithmetic function body in the
>>>>> evaluable predicate definition code. Comparison
>>>>> to old measurements and some measurements of
>>>>>
>>>>> other Prolog systems:
>>>>>
>>>>> /* Dogelog Player 2.1.2 for Java, weeks ago */
>>>>> ?- time(test).
>>>>> % Zeit 378 ms, GC 1 ms, Lips 19900780, Uhr 28.08.2025 17:44
>>>>> true.
>>>>>
>>>>> /* SWI-Prolog 9.0.4 */
>>>>> ?- time(test).
>>>>> % 7,506,639 inferences, 0.363 CPU in 0.362 seconds
>>>>> (100% CPU, 20693560 Lips)
>>>>> true.
>>>>>
>>>>> /* Scryer Prolog 0.9.4-639 */
>>>>> ?- time(test).
>>>>> % CPU time: 0.365s, 7_517_613 inferences
>>>>> true.
>>>>>
>>>>> /* Trealla Prolog 2.82.23-3 */
>>>>> ?- time(test).
>>>>> % Time elapsed 0.868s, 11263917 Inferences, 12.983 MLips
>>>>> true.
>>>>>
>>>>> Bye
>>>>>
>>>>> P.S.: The code uses the hated mathematical mod/2,
>>>>> and not the cheaper rem/2 that CPUs usually have:
>>>>>
>>>>> test :-
>>>>>     len(L, 1000),
>>>>>     primes(L, _).
>>>>>
>>>>> primes([], 1).
>>>>> primes([J|L], J) :-
>>>>>     primes(L, I),
>>>>>     K is I+1,
>>>>>     search(L, K, J).
>>>>>
>>>>> search(L, I, J) :-
>>>>>     mem(X, L),
>>>>>     I mod X =:= 0, !,
>>>>>     K is I+1,
>>>>>     search(L, K, J).
>>>>> search(_, I, I).
>>>>>
>>>>> mem(X, [X|_]).
>>>>> mem(X, [_|Y]) :-
>>>>>     mem(X, Y).
>>>>>
>>>>> len([], 0) :- !.
>>>>> len([_|L], N) :-
>>>>>     N > 0,
>>>>>     M is N-1,
>>>>>     len(L, M).
>>>>>
>>>>> Mild Shock schrieb:
>>>>>> Hi,
>>>>>>
>>>>>> WebPL is already outdated I guess. It doesn't
>>>>>> show the versions of the other Prolog systems
>>>>>> it is using. While I had these results for
>>>>>>
>>>>>> the primes example in the WebPL playground:
>>>>>>
>>>>>> /* Trealla Prolog WASM */
>>>>>> (23568.9ms)
>>>>>>
>>>>>> When I run the example here:
>>>>>>
>>>>>> https://php.energy/trealla.html
>>>>>>
>>>>>> I get better results:
>>>>>>
>>>>>> /* trealla-js 0.27.1 */
>>>>>>
>>>>>> ?- time(test).
>>>>>> % Time elapsed 9.907s, 11263917 Inferences, 1.137 MLips
>>>>>>
>>>>>> Bye
>>>>>
>>>>
>>>
>>
> 

[toc] | [prev] | [next] | [standalone]

#14931 — Vertex AI Training is more expensive? (Was: Give Julio Di Egidio the bloody money [3 RMB¥ MiniMind])

Hi,

Vertex AI Training is more expensive:

"Vertex AI provides a managed training
service that enables you to operationalize
large scale model training. You can use
Vertex AI to run training applications
based on any machine learning (ML) framework
on Google Cloud infrastructure. Vertex AI
also has integrated support that simplifies
the preparation process for model training
and serving for the PyTorch, TensorFlow,
scikit-learn, and XGBoost frameworks.
https://cloud.google.com/products/calculator

For 10 Jobs à 10 hours, it wants 1000 USD
from me. The accelerator is TPU V3. Nevertheless
might be the cheaper option to use the RTX™ 6000

for 720 hours. ChatGPT gives me this calculation:

Scenario I: Single TPU, RTX 6000 wins
4.43×10^19 FLOPs versus ~2.36×10^20 FLOPs

Scenario II: 8 chips TPU, RTX 6000 looses
~3.54×10^20 FLOPs versus ~2.36×10^20 FLOPs

Bye

Mild Shock schrieb:
> Hi,
> 
> Give Julio Di Egidio the bloody money. He is
> craving for 300 USD so that he can buy the
> ISO Prolog core standard. Just imagine he would want
> 
> to build a MiniMind. Just lets put some more
> prespective on the current costs:
> 
> This open-source project aims to train a super-small
> language model MiniMind with only 3 RMB cost and
> 2 hours, starting completely from scratch. The
> MiniMind series is extremely lightweight, with the
> smallest version being 1/7000 the size of GPT-3,
> making it possible to train quickly on even the
> most ordinary personal GPUs.
> https://github.com/jingyaogong/minimind/blob/master/README_en.md
> 
> ChatGPT tells me that most of the numbers
> are correct when you rent a GPU by the hour.
> But what about a 100% ownership of a GPU for
> 
> a year. I find this might cost 12'000 USD.
> One has to separate platforms for execution from
> those platforms for training:
> 
> GEX44: for AI inference
> Nvidia RTX™ 4000, 184 EUR / month
> 
> GEX130: for AI training
> NVIDIA RTX™ 6000, 813 EUR / month
> https://www.hetzner.com/dedicated-rootserver/matrix-gpu/
> 
> Bye
> 
> Mild Shock schrieb:
>> Hi,
>>
>> Thinks are definitively accelerating. I really would
>> like to use an AI that knows about all the News of today.
>> This bloody cut date is so annoying.
>>
>> Further indicative that AI is accelerating:
>>
>> In August 2025, Sam Altman dropped a bombshell:
>>
>> *months, not years: Rushing GPT-6*
>> In August 2025, Sam Altman dropped a bombshell:
>> GPT-6 is already in development and coming sooner
>> than you think. Not in two years, but
>> potentially in months.
>> https://www.youtube.com/watch?v=44mJb5sKji0
>>
>> Karpathy, coined vibe coding, released in October 2025:
>>
>> *nanochat: The best ChatGPT that $100 can buy*
>> This repo is a full-stack implementation of an
>> LLM like ChatGPT in a single, clean, minimal,
>> hackable, dependency-lite codebase. nanochat is
>> designed to run on a single 8XH100 node via
>> scripts like speedrun.sh, that run the
>> entire pipeline start to end.
>> https://github.com/karpathy/nanochat
>>
>> Bye
>>
>> Mild Shock schrieb:
>>> Hi,
>>>
>>> But not only Mobie AI and Desktop AI are making
>>> a broader imprint now. We might also experience
>>> Workstation AI, with a 3'000.- USD price tag:
>>>
>>>  > You Can't Buy This... Yet! The NVIDIA GB10 from Dell
>>>  > The New Superchip that Terrifies the Cloud!
>>>  > https://www.youtube.com/watch?v=x1qViw4xyVo
>>>
>>> So whats going on? I was asking Phind, which is
>>> driven by a 70B model tailored towards developers:
>>>
>>> Q: Is there an AI inflection point right now ,
>>>     with NPUs in mobile, desktop and workstation
>>>
>>> A: Evidence of the Inflection Point
>>>
>>> - Mobile Leadership
>>>    NPUs originated in smartphones
>>>    Now becoming ubiquitous across all device types
>>>    Enabling sophisticated AI features at consumer price points
>>>
>>> - Desktop Revolution
>>>    Major manufacturers implementing NPUs across product lines
>>>    Apple's Neural Engine integrated into M-series chips
>>>    Qualcomm, Intel, and AMD incorporating AI accelerators
>>>
>>> - Workstation Transformation
>>>    Professional-grade NPUs in mobile workstations
>>>    Demonstrated superior performance for AI-specific tasks
>>>    Enabling local processing of previously cloud-dependent workloads
>>>
>>> https://www.phind.com/search/cmgs1s6jv00023h67g5z2aaa0
>>>
>>> Bye
>>>
>>> Mild Shock schrieb:
>>>> Hi,
>>>>
>>>> It seems I am having problems pacing with
>>>> all the new fancy toys. Wasn't able to really
>>>> benchmark my NPU from a Desktop AI machine,
>>>>
>>>> picked the wrong driver. Need to try again.
>>>> What worked was benchmarking Mobile AI machines.
>>>> I just grabbed Geekbench AI and some devices:
>>>>
>>>> USA Fab, M4:
>>>>
>>>>      sANN    hANN    qANN
>>>> iPad CPU    4848    7947    6353
>>>> iPad GPU    9752    11383    10051
>>>> iPad NPU    4873    36544    *51634*
>>>>
>>>> China Fab, Snapdragon:
>>>>
>>>>      sANN    hANN    qANN
>>>> Redmi CPU    1044    950    1723
>>>> Redmi GPU    480    905    737
>>>> Redmi NNAPI    205    205    469
>>>> Redmi QNN    226    226    *10221*
>>>>
>>>> Speed-Up via NPU is factor 10x. See the column
>>>> qANN which means quantizised artificial neural
>>>> networks, when NPU or QNN is picked.
>>>>
>>>> The mobile AI NPUs are optimized using
>>>> mimimal amounts of energy, and minimal amounts
>>>> of space squeezing (distilling) everything
>>>>
>>>> into INT8 and INT4.
>>>>
>>>> Bye
>>>>
>>>> Mild Shock schrieb:
>>>>> Hi,
>>>>>
>>>>> The change from 378 ms to 286 ms is around 25-30%
>>>>> is insane. But I did both tests on a novel AI CPU.
>>>>> To be precise on a AMD Ryzen AI 7 350.
>>>>>
>>>>> But somehow I picked up rumors that AI CPUs now
>>>>> might do Neural Network Branch Prediction. The
>>>>> idea seems to exist in hardware at least since (2012):
>>>>>
>>>>> Machine learning and artificial intelligence are
>>>>> the current hype (again). In their new Ryzen
>>>>> processors, AMD advertises the Neural Net
>>>>> Prediction. It turns out this is was already
>>>>> used in their older (2012) Piledriver architecture
>>>>> used for example in the AMD A10-4600M. It is also
>>>>> present in recent Samsung processors such as the
>>>>> one powering the Galaxy S7. What is it really?
>>>>> https://chasethedevil.github.io/post/the_neural_network_in_your_cpu/
>>>>>
>>>>> It can be done with Convoluted Neural Networks (CNN):
>>>>>
>>>>> BranchNet: A Convolutional Neural Network to
>>>>> Predict Hard-To-Predict Branches
>>>>> To this end, Tarsa et al. proposed using convolutional
>>>>> neural networks (CNNs) that are trained at
>>>>> compiletime to accurately predict branches that
>>>>> TAGE cannot. Given enough profiling coverage, CNNs
>>>>> learn input-independent branch correlations.
>>>>> https://microarch.org/micro53/papers/738300a118.pdf
>>>>>
>>>>> Interstingly the above shows cases a PGO based
>>>>> Machine Learning for Branch Predictors. No clue
>>>>> how they construct the CPU, that they can feed
>>>>>
>>>>> it with offline constructed neural neutworks for
>>>>> their own execution. Maybe an optimizer uses it?
>>>>> But I guess a more modern  solutions would not only
>>>>>
>>>>> use CNN, but also an Attention Mechanism.
>>>>>
>>>>> Bye
>>>>>
>>>>> Mild Shock schrieb:
>>>>>> Hi,
>>>>>>
>>>>>> I spent some time thinking about my primes.pl
>>>>>> test. And came to the conclusion that it
>>>>>> mainly tests the Prolog ALU. Things like
>>>>>>
>>>>>> integer successor or integer modulo. Then
>>>>>> I found that Java has Math.floorMod() which
>>>>>> I wasn't using yet. And peng results are better:
>>>>>>
>>>>>> /* Dogelog Player 2.1.2 for Java, today */
>>>>>> ?- time(test).
>>>>>> % Zeit 286 ms, GC 1 ms, Lips 26302430, Uhr 15.10.2025 02:31
>>>>>> true.
>>>>>>
>>>>>> Maybe the Java backend picks a CPU instruction
>>>>>> for Math.floorMod() instead of executing the
>>>>>> longer code sequence that is needed to correct
>>>>>>
>>>>>> rem/2 into mod/2. Who knows. I also reorganized
>>>>>> the code a little bit, and eliminated an extra
>>>>>> method call in all arithmetic functions, by
>>>>>>
>>>>>> inlining the arithmetic function body in the
>>>>>> evaluable predicate definition code. Comparison
>>>>>> to old measurements and some measurements of
>>>>>>
>>>>>> other Prolog systems:
>>>>>>
>>>>>> /* Dogelog Player 2.1.2 for Java, weeks ago */
>>>>>> ?- time(test).
>>>>>> % Zeit 378 ms, GC 1 ms, Lips 19900780, Uhr 28.08.2025 17:44
>>>>>> true.
>>>>>>
>>>>>> /* SWI-Prolog 9.0.4 */
>>>>>> ?- time(test).
>>>>>> % 7,506,639 inferences, 0.363 CPU in 0.362 seconds
>>>>>> (100% CPU, 20693560 Lips)
>>>>>> true.
>>>>>>
>>>>>> /* Scryer Prolog 0.9.4-639 */
>>>>>> ?- time(test).
>>>>>> % CPU time: 0.365s, 7_517_613 inferences
>>>>>> true.
>>>>>>
>>>>>> /* Trealla Prolog 2.82.23-3 */
>>>>>> ?- time(test).
>>>>>> % Time elapsed 0.868s, 11263917 Inferences, 12.983 MLips
>>>>>> true.
>>>>>>
>>>>>> Bye
>>>>>>
>>>>>> P.S.: The code uses the hated mathematical mod/2,
>>>>>> and not the cheaper rem/2 that CPUs usually have:
>>>>>>
>>>>>> test :-
>>>>>>     len(L, 1000),
>>>>>>     primes(L, _).
>>>>>>
>>>>>> primes([], 1).
>>>>>> primes([J|L], J) :-
>>>>>>     primes(L, I),
>>>>>>     K is I+1,
>>>>>>     search(L, K, J).
>>>>>>
>>>>>> search(L, I, J) :-
>>>>>>     mem(X, L),
>>>>>>     I mod X =:= 0, !,
>>>>>>     K is I+1,
>>>>>>     search(L, K, J).
>>>>>> search(_, I, I).
>>>>>>
>>>>>> mem(X, [X|_]).
>>>>>> mem(X, [_|Y]) :-
>>>>>>     mem(X, Y).
>>>>>>
>>>>>> len([], 0) :- !.
>>>>>> len([_|L], N) :-
>>>>>>     N > 0,
>>>>>>     M is N-1,
>>>>>>     len(L, M).
>>>>>>
>>>>>> Mild Shock schrieb:
>>>>>>> Hi,
>>>>>>>
>>>>>>> WebPL is already outdated I guess. It doesn't
>>>>>>> show the versions of the other Prolog systems
>>>>>>> it is using. While I had these results for
>>>>>>>
>>>>>>> the primes example in the WebPL playground:
>>>>>>>
>>>>>>> /* Trealla Prolog WASM */
>>>>>>> (23568.9ms)
>>>>>>>
>>>>>>> When I run the example here:
>>>>>>>
>>>>>>> https://php.energy/trealla.html
>>>>>>>
>>>>>>> I get better results:
>>>>>>>
>>>>>>> /* trealla-js 0.27.1 */
>>>>>>>
>>>>>>> ?- time(test).
>>>>>>> % Time elapsed 9.907s, 11263917 Inferences, 1.137 MLips
>>>>>>>
>>>>>>> Bye
>>>>>>
>>>>>
>>>>
>>>
>>
> 

[toc] | [prev] | [next] | [standalone]

#14924 — The Love Affair: OpenAI and AMD (Was: NPUs (Neural Processing Units) are the new normal)

Hi,

This is probably what my AI Laptop can do. The
demo shows AMD Ryzen AI 7 340 micro. I have a
AMD Ryzen AI 7 350 laptop.

 > 100% Powered by AMD Ryzen™ AI NPU
 > https://www.youtube.com/watch?v=0t8ijUPg4A0

Only I am too stupid / too lazy to dig up the right
drivers and install FastFlowLM. But one sees in
the video how the NPU gets 30% - 60% occupied,

and it does the transcription of a YouTube video,
into text (via Whisper-large-v3-turbo from OpenAI).
The demo then switches to summarize mode

(via GPT-OSS-20B from OpenAI). And boom the,
NPU goes to 100%!

Bye

P.S.: I didn't know about the OpenAI and AMD partnership,
also buying an AMD AI Laptop was not motivated by
this development. Not sure whether its really a big thing:

AMD and OpenAI announce partnership
https://openai.com/index/openai-amd-strategic-partnership/

It might be a good thing for end users like me, if
Edge uses cases like the above become more common.
But they still depend on models trained not on the Edge,

but in a Data Center. The AMD Instinct product line directly
competes with Nvidia's Tesla and Intel's Xeon Phi and Data
Center GPU lines of machine learning and GPGPU cards.

Mild Shock schrieb:
> Hi,
> 
> It seems I am having problems pacing with
> all the new fancy toys. Wasn't able to really
> benchmark my NPU from a Desktop AI machine,
> 
> picked the wrong driver. Need to try again.
> What worked was benchmarking Mobile AI machines.
> I just grabbed Geekbench AI and some devices:
> 
> USA Fab, M4:
> 
>      sANN    hANN    qANN
> iPad CPU    4848    7947    6353
> iPad GPU    9752    11383    10051
> iPad NPU    4873    36544    *51634*
> 
> China Fab, Snapdragon:
> 
>      sANN    hANN    qANN
> Redmi CPU    1044    950    1723
> Redmi GPU    480    905    737
> Redmi NNAPI    205    205    469
> Redmi QNN    226    226    *10221*
> 
> Speed-Up via NPU is factor 10x. See the column
> qANN which means quantizised artificial neural
> networks, when NPU or QNN is picked.
> 
> The mobile AI NPUs are optimized using
> mimimal amounts of energy, and minimal amounts
> of space squeezing (distilling) everything
> 
> into INT8 and INT4.
> 
> Bye

[toc] | [prev] | [next] | [standalone]

#14938 — The NPU in your Browser [WebNN by W3C] (Was: NPUs (Neural Processing Units) are the new normal)

Hi,

Candidate Recommendation Draft - 30 September 2025
https://www.w3.org/TR/webnn

WebNN samples by Ningxin Hu, Intel, Shanghai
https://github.com/webmachinelearning/webnn-samples

Bye

Mild Shock schrieb:
> Hi,
> 
> It seems I am having problems pacing with
> all the new fancy toys. Wasn't able to really
> benchmark my NPU from a Desktop AI machine,
> 
> picked the wrong driver. Need to try again.
> What worked was benchmarking Mobile AI machines.
> I just grabbed Geekbench AI and some devices:
> 
> USA Fab, M4:
> 
>      sANN    hANN    qANN
> iPad CPU    4848    7947    6353
> iPad GPU    9752    11383    10051
> iPad NPU    4873    36544    *51634*
> 
> China Fab, Snapdragon:
> 
>      sANN    hANN    qANN
> Redmi CPU    1044    950    1723
> Redmi GPU    480    905    737
> Redmi NNAPI    205    205    469
> Redmi QNN    226    226    *10221*
> 
> Speed-Up via NPU is factor 10x. See the column
> qANN which means quantizised artificial neural
> networks, when NPU or QNN is picked.
> 
> The mobile AI NPUs are optimized using
> mimimal amounts of energy, and minimal amounts
> of space squeezing (distilling) everything
> 
> into INT8 and INT4.
> 
> Bye

[toc] | [prev] | [next] | [standalone]

#14939 — Fuzzy Alert: Boris the Loris on the Dancefloor (Was: The NPU in your Browser [WebNN by W3C])

Hi,

Boris the Loris and Julio Di Egidio the Nazi Retard,
are going for an afterwork beer. They are still
highly confused by Fuzzy Testing:

Star Trek - The 70's Disco Generation
https://www.youtube.com/watch?v=505zvAvnreg

The favorite hangout is Spock's Logic Dancefloor,
which is known for its sharp unfuzzy wit. They
have  a chat with Data about Disco Math,

the only Math which has no Fuzzy Logic in it.

Bye

Mild Shock schrieb:
> Hi,
> 
> Candidate Recommendation Draft - 30 September 2025
> https://www.w3.org/TR/webnn
> 
> WebNN samples by Ningxin Hu, Intel, Shanghai
> https://github.com/webmachinelearning/webnn-samples
> 
> Bye
> 
> Mild Shock schrieb:
>> Hi,
>>
>> It seems I am having problems pacing with
>> all the new fancy toys. Wasn't able to really
>> benchmark my NPU from a Desktop AI machine,
>>
>> picked the wrong driver. Need to try again.
>> What worked was benchmarking Mobile AI machines.
>> I just grabbed Geekbench AI and some devices:
>>
>> USA Fab, M4:
>>
>>      sANN    hANN    qANN
>> iPad CPU    4848    7947    6353
>> iPad GPU    9752    11383    10051
>> iPad NPU    4873    36544    *51634*
>>
>> China Fab, Snapdragon:
>>
>>      sANN    hANN    qANN
>> Redmi CPU    1044    950    1723
>> Redmi GPU    480    905    737
>> Redmi NNAPI    205    205    469
>> Redmi QNN    226    226    *10221*
>>
>> Speed-Up via NPU is factor 10x. See the column
>> qANN which means quantizised artificial neural
>> networks, when NPU or QNN is picked.
>>
>> The mobile AI NPUs are optimized using
>> mimimal amounts of energy, and minimal amounts
>> of space squeezing (distilling) everything
>>
>> into INT8 and INT4.
>>
>> Bye

[toc] | [prev] | [standalone]

Page 2 of 2 — ← Prev page 1 [2]

Back to top | Article view | comp.lang.prolog

csiph-web