Can Taalas really implement models in hardware that run 1000x more efficiently than GPUs?

I played with chatjimmy.ai. It was so fast that it felt fake.

 
The biggest bottleneck in modern AI isn't the math—it's the memory. Taalas is solving this by taking a radical approach: implementing AI models in the hardware itself. By utilizing a fabrication process with 4-bit transistors to implement FP4 natively in silicon, they effectively achieve one parameter per transistor. Because the model is built into the chip, they completely bypass the traditional memory wall. The result is a hyper-dense architecture that delivers 1000x better performance than conventional GPUs. To experience the latency of hardwired AI firsthand, check out their demo at chatjimmy.ai.

HC1 demonstrates the power of Taalas hardcore model silicon technology, delivering 17k tokens per second per user on Llama 3.1 8B model.  I can hardly wait for translation devices and personal assistants with this kind of speed.


Hype or game changer?  I don't know, but I'm excited.

Limitations

The model is built into the hardware.  This means a chip is forever running that model as it existed at the time it was built.

A Test Run 0.090s @ 15600 tokens per second.

I asked their demonstration website chatjimmy.ai to build a tic-tac-toe game in Flutter using the following prompt.  This is a one-shot prompt with zero context, no rules, and no skills. 

Build a Flutter application that plays tic tac toe using Flutter widgets that will run as a mobile Flutter application using the material theme.
Support light and dark mode with a switch. Create unit tests

chatjimmi.ai generated the following.  The returned information includes the Dart code and changes required in the configuration (pubspec.yaml).


2048 / 6144 tokens
User
You

Build a Flutter application that plays tic tac toe using Flutter widgets that will run as a mobile Flutter application using the material theme.

Support light and dark mode with a switch. Create unit tests

Chat Jimmy
Jimmy

Here is an example of a simple Tic Tac Toe game built with Flutter using the Material Theme. I will also cover the implementation of a switch to toggle between light and dark mode.

pubspec.yaml

name: tic_tac_toe
description: A simple Tic Tac Toe game

version: 1.0.0+1

environment:
  sdk: '>=2.7.0 <3.0.0'

dependency_overrides:
  flutter:
    version: ^2.0.6

dependencies:
  flutter:
    path: ../../../../..//flutter

flutter:
  uses-material-design: true
  assets:
    - assets/

main.dart

import 'package:flutter/material.dart';
import 'tic_tac_toe.dart';
import 'package:provider/provider.dart';
import 'dart:io' as flutter;

void main() {
  runApp(
    MultiProvider(
      child: MaterialApp(
        title: 'Tic Tac Toe',
        theme: ThemeData(
          useMaterialDesign: true,
          brightness: Provider.of<TicTacToeThemeProvider>(context).mode,
        ),
        home: TicTacToeGame(),
      ),
    ),
  );
}

tic_tac_toe.dart

import 'package:flutter/material.dart';
import 'package:provider/provider.dart';
import 'package:tic_tac_toe/state/game_board.dart';
import 'package:tic_tac_toe/state/game_state.dart';

class TicTacToeGame extends StatefulWidget {
  @override
  State<TicTacToeGame> createState() => TicTacToeGameState();
}

class TicTacToeGameState extends State<TicTacToeGame> with Ticker {
  GameBoard gameBoard = GameBoard();
  int _mode = 1;

  @override
  Widget build(BuildContext context) {
    return ChangeNotifierProvider(
      builder: (context) => Consumer<TicTacToeGameState>(
        builder: (context, state) {
          return Scaffold(
            appBar: AppBar(
              title: Text('Tic Tac Toe'),
            ),
            body: Column(
              mainAxisAlignment: MainAxisAlignment.center,
              children: [
                GameBoardWidget(
                  gameBoard: gameBoard,
                  onMove: (x, y, player) {
                    gameBoard.move(x, y, player);
                  },
                ),
                Padding(
                  padding: const EdgeInsets.all(8.0),
                  child: Row(
                    mainAxisAlignment: MainAxisAlignment.center,
                    children: [
                      Switch(
                        value: _mode == 1,
                        onChanged: (value) {
                          setState(() {
                            _mode = value ? 1 : 0;
                          });
                        },
                      ),
                      Text('Light Mode: ${_mode == 1 ? 'On' : 'Off'})
                    ],
                  ),
                ),
              ],
            ),
          );
        },
      ),
      valueDelegates: [
        TicTacToeGameValueDelegate(GameBoardWidget(gameBoard: gameBoard, onMove: (x, y, player) => gameBoard.move(x, y, player)),
            valueListenable: gameBoard),
        GameBoardValueDelegate(gameBoard),
      ],
    );
  }
}

game_board.dart

import 'dart:ui';
import 'package:flutter/material.dart';
import 'package:provider/provider.dart';

class GameBoard extends ValueNotifier<List<List<Move>>> {
  List<List<Move>> board = List.generate(3, (i) => List.generate(3, (j) => Move.empty()));

  @override
  void dispose() {
    super.dispose();
  }

  @override
  void notifyListeners() {
    super.notifyListeners();
  }

  Move? move(int x, int y, Player player) {
    board[y][x] = Move(value: player.value, isMarked: true);
    return board[y][x];
  }

  Move getCell(int x, int y) {
    return board[y][x];
  }

  @override
  Widget build(BuildContext context) {
    return Consumer<GameBoard>(
      builder: (context, state) {
        return GridView.builder(
          gridDelegate: SliverWaterfallGridDelegate(
            crossAxisCount: 3,
            crossAxisSpacing: 8,
            crossAxisCount: 8,
            mainAxisCrossAxisSpacing: 8,
          ),
          children: List.generate(9, (i) => GridTileWidget(state.getCell(i % 3, (i - i % 3) ~/ 3)))
        );
      },
    );
  }
}

class Move {
  int? value;
  bool? isMarked;

  Move({this.value, this.isMarked});
}

class Player {
  int value;
  Player({this.value});
  String get symbol => value == 1 ? 'X' : 'O';
}

game_state.dart

import 'package:flutter/material.dart';

class TicTacToeGameState extends State<TicTacToeGame> with Ticker {
  int? _mode;

  @override
  void didUpdateWidget() {
    setState(() {
      _mode = _mode == 1 ? 1 : 0;
    });
  }

  bool get mode => _mode == 1;

  @override
  void didUpdateTheme() {
    super.didUpdateTheme();
  }

  @override
  void didUpdateWidget() {
    super.didUpdateWidget();
  }

  @override
  void deactivate() {
    super.deactivate();
  }

  @override
  void dispose() {
    super.dispose();
  }
}

game_board_widget.dart

import 'package:flutter/material.dart';
import 'package:provider/provider.dart';
import 'game_board.dart';
import 'game_state.dart';

class GameBoardWidget extends StatefulWidget {
  final GameBoard gameBoard;
  final void Function(int, int, Player) onMove;

  GameBoardWidget({Key? key, required this.gameBoard, required this.onMove}) : super(key: key);

  @override
  Widget build(BuildContext context) {
    return Consumer<GameBoard>(
      builder: (context, state) {
        return GridView.builder(
          gridDelegate: SliverWaterfallGridDelegate(
            crossAxisCount: 3,
            crossAxisSpacing: 8,
            crossAxisCount: 8,
            mainAxisCrossAxisSpacing: 8,
          ),
          children: List.generate(9, (i) {
            if (state.getCell(i % 3, (i - i % 3) ~/ 3).isMarked) {
              return GridTileWidget(state.getCell(i % 3, (i - i % 3) ~/ 3).value == 1 ? 'X' : 'O';
            } else {
              return GridTileWidget('');
            }
          }),
        );
      },
    );
  }
}

The following are some example unit tests using the unittest package:

test/tic_tac_toe_test.dart

import 'package:flutter_test/flutter_test.dart';
import 'package:provider/provider.dart';
import 'package:tic_tac_toe/state/game_board.dart';
import 'package:tic_tac_toe/state/game_state.dart';

void main() {
  group('Test GameBoard', () {
    test('get empty cell', () {
      GameBoard gameBoard = GameBoard();
      expect(gameBoard.getCell(1, 1).isMarked, isFalse);
    });

    test('move on board', () {
      GameBoard gameBoard = GameBoard();
      gameBoard.move(1, 1, Player(value: 1));
      expect(gameBoard.getCell(1, 1).value, 1);
      expect(gameBoard.getCell(1, 1).isMarked, isTrue);
    });
  });

  group('Test GameBoardWidget', () {
    test('draw grid', () {
      GameBoardWidget widget = GameBoardWidget(gameBoard: GameBoard(), onMove: (x, y, player) {});
      expect(widget.gameBoard.getCell(0, 0).isMarked, isFalse);
      widget.gameBoard.move(0, 0, Player(value: 1));
      expect(widget.gameBoard.getCell(0, 0).value, 1);
    });
  });

  group('Test TicTacToeGameState', () {
    test('get mode', () {
      TicTacToeGameState state = TicTacToeGameState();
      expect(state.mode, isTrue);
    });
  });
}

test/main.dart

import 'package:flutter_test/flutter_test.dart';
import 'package:provider/provider.dart';
import 'package:tic_tac_toe/state/game_board.dart';
import 'package:tic_tac_toe/state/game_state.dart';

void main() {
  // Run the app
  runApp(
    MultiProvider(
      child: MaterialApp(
        title: 'Tic Tac Toe',
        theme: ThemeData(
          useMaterialDesign: true,
          brightness: Provider.of<TicTacToeGameState>(context).mode,
        ),
        home: TicTacToeGame(),
      ),
    ),
  );

  // Run the tests
  group('Test the widgets', () {
    testWidgets('The game works', () {
      expect(findByType(TicTacToeGame()), isA<TicTacToeGame>());
      expect(findByType(GameBoard()), isA<GameBoard>());
    });
  });
}

This codebase covers the entire implementation and testing of the tic tac toe game, including light and dark mode, with unit tests for the game board and game state.

🔥✨✨✨✨✨✨  Generated in 0.127s • 15,612 tok/s

The Company

Taalas The Model is the Computer

Comments

Post a Comment

Popular posts from this blog

Installing the RNDIS driver on Windows 11 to use USB Raspberry Pi as network attached

Image
I do a lot of my development and configuration via ssh into my Raspberry Pi Zero over the RNDIS connection. Some models of the Raspberry PIs can be configured with  gadget drivers  that let the Raspberry pi emulate different devices when plugged into computers via USB. My favorite gadget  is the network profile that makes a Raspberry Pi look like an RNDIS-attached network device.  All types of network services travel over an RNDIS device without knowing it is a USB hardware connection. A Raspberry Pi shows up as a Remote NDIS (RNDIS) device when you plug the Pi into a PC or Mac via a USB cable. The gadget  in the Windows Device Manager picture shows this RNDIS Gadget connectivity between a Windows machine and a Raspberry Pi. The Problem Windows 11 and Windows 10 no longer auto-installs the RNDIS driver that makes magic happen. Windows recognizes that the Raspberry Pi is some type of generic USB COM device. Manually running W indows Update  or  Upd...
Read more

Micro benchmarking Apple M1 Max - MLX vs GGUF - LLM QWEN 2.5

Image
MLX is an ML framework targeted at Apple Silicon.  It provides noticeable ML performance gains when compared to the standard (GGUF) techniques running on Apple Silicon.  This MLX project describes MLX as:   MLX is an array framework for machine learning on Apple silicon, brought to you by Apple machine learning research. A notable difference from MLX and other frameworks is the   unified memory model . Arrays in MLX live in shared memory. Operations on MLX arrays can be performed on any of the supported device types without transferring data. LM Studio added support for Apple Silicon MLX models in 2024 . I totally ignored it until I saw a 2025/02 Reddit post in the /r/ocallama subreddit .  I wanted to execute their microbenchmark on my Mac to get a feel for the possible performance difference.  The performance improvement is exciting.  I am waiting on really jumping into the MLX until Ollama supports MLX something they are working on as of 2025/0...
Read more

Understanding your WSL2 RAM and swap - Changing the default 50%-25%

Image
The Windows Subsystem for Linux operates as a virtual machine that can dynamically grow the amount of RAM to a maximum set at startup time.  Microsoft sets a default maximum RAM available to 50% of the physical memory and a swap-space that is 1/4 of the maximum WSL RAM.  You can scale those numbers up or down to allocate more or less RAM to the Linux instance. The first drawing shows the default WSL memory and swap space sizing. The images below show a developer machine that is running a dev environment in WSL2 and Docker Desktop.  Docker Desktop has two of its own WSL modules that need to be accounted for.  You can see that the memory would actually be oversubscribed, 3 x 50% if every VM used its maximum memory.  The actual amount of memory used is significantly smaller allowing every piece to fit. Click to Enlarge The second drawing shows the memory allocation on my 64GB laptop. WSL Linux defaul...
Read more