PAGE

MACE Compatibility Guide

Welcome to the MACE Compatibility Guide for version 1.3.0, the most advanced release of MACE to date. With this version, MACE introduces a powerful new Prompt Generation Engine, enabling direct text-to-audio generation of high-quality samples inside your DAW. This guide outlines hardware recommendations, OS requirements, performance expectations, and tips for managing system resources effectively. This engine can generate high-quality and stereo samples up to 11 seconds long. This means drum loops, melodies, one-shots, drum shots, ambience and more!

✨ What’s New in MACE 1.3.0

At the core of this update is the Prompt Engine — a cutting-edge generative system capable of transforming text prompts into rich, musical samples using machine learning models. This is a technical milestone in plugin design and brings with it new demands on your system.

Note: While optimized extensively, prompt generation speed varies significantly based on your hardware — especially whether or not you have a compatible GPU or a high-performance CPU.

⚙️ Performance Overview

MACE is designed for modern hardware. The speed and performance of prompt generation depends heavily on your system configuration.

✅ Recommended Hardware

  • CPU: Intel i7 or better (Intel i9 recommended if no GPU) | M3 or M4 Series CPU (Mac)
  • GPU: NVIDIA RTX 3060+ or equivalent (DirectX 12-capable)
  • RAM: 32–64 GB
  • Disk: 10–20 GB free for generation cache and exports
  • System Type: Desktop PC or M-Series Mac preferred. (Some Windows Laptops may experience performance hits)

⚠️ Minimum Requirements

  • Windows: Windows 10+ (Version 19041 or later)
  • macOS: Sequoia 15.0+ (required for CoreML acceleration)
  • RAM: Minimum 16 GB
  • CPU: i5 or Apple M1 (Intel Macs supported, but very slow)

🖥️ macOS vs Windows Differences

  • macOS: M-series Macs benefit from CoreML acceleration, dramatically reducing generation times. Intel Macs are supported, but may experience slow performance and compatibility issues.
  • Windows: MACE uses DirectML GPU acceleration, which works across most modern NVIDIA, AMD, and Intel GPUs.

🧠 Prompt Engine Behavior

When you open MACE for the first time in a session, it will load the Prompt Engine. This includes multiple machine learning models, which can briefly freeze your DAW UI — this is expected.

Once one instance of MACE has loaded the Prompt Engine, all other instances will share it.

  • Initial Load Time: 2–15 seconds depending on hardware
  • RAM Usage:
    • With shared engine: ~7–9 GB total
    • Without shared engine (i.e. if MACEModelManager.dll/.dylib fails to load): Each instance may use 5–7 GB individually

If you’re seeing RAM usage that scales per-instance or goes well above 9 GB total, MACE is likely not using the shared memory manager, which indicates a faulty install.

🧱 Fixed Models [Legacy Models]

In addition to the new Prompt Generation Engine, MACE 1.3.0 continues to include the original models, now labeled as Fixed Models in the Model Browser.

These models are not prompt-based and were trained on short, unlabeled samples. As a result:

  • 🔊 Sound Character: Slightly grainier and more limited in variation, but still uniquely textured — many users appreciate their distinct sonic signature.
  • ⚡ Performance: Extremely fast generation times, even on low-end CPUs.
  • 🎯 Use Case: Best suited for short glitchy one-shots (e.g., snares, kicks, FX, stabs).

💡 Tip: Fixed Models will still work even if your machine can’t run the Prompt Engine effectively. They’re ideal for machines with older hardware or when you want instant generation without waiting.

To access them, simply open the Model Browser dropdown and look under “Fixed Models [Legacy]”. They’re kept in 1.3.0 specifically for their utility, speed, and character.

🎮 GPU Compatibility (Windows)

MACE uses DirectML for GPU inference on Windows. While not CUDA-based, it still requires modern graphics cards that support DirectX 12 (Feature Level 12_0).

Supported GPUs

  • NVIDIA: GTX 900 series and newer
  • AMD: RX 400 series and newer
  • Intel: UHD 600 series, Iris Xe, and newer

Minimum Driver Versions

  • NVIDIA: 451.48+
  • AMD: 20.5.1+
  • Intel: 27.20.100.8190+

While MACE should use your GPU just fine out of the box, it is recommend to use the Studio Drivers for stability, and keeping your drivers up to date using NVIDIA’s GeForce Experience.

NOTE: There is a Force CPU setting in the Settings Panel within MACE. If you find an issue with your GPU and MACE is not gracefully falling back to CPU, you can bypass the GPU check.

MACE’s Setting Panel will list if GPU Acceleration or CoreML is enabled.

💾 Generation Cache & Disk Management

MACE uses a dedicated Generation Cache folder to store generated .wav files. These can quickly grow in size depending on how much content you generate.

Default Locations

  • Windows: Documents\Tensorpunk\MACE\GenerationCache
  • macOS: ~/Documents/Tensorpunk/MACE/GenerationCache

Important: Avoid using OneDrive or cloud-synced folders for cache storage.

On Windows, your default path can be changed in the installer. Both versions support changing the path within MACE itself.

Built-in Cache Tools

Inside MACE’s Settings Panel, you’ll find:

  • Wipe Generation Cache
    Deletes all unprotected .wav files from the cache.
  • Protect This Session From Cache Wipe
    Ensures samples used in a saved DAW project are not deleted. When you save a project with a MACE instance in the session, the Protect This Session From Cache Wipe automatically gets enabled for ease of mind.

💡 Tip: Dragged samples into your DAW are saved in your Render Output Directory and are safe from cache wipes.

The Settings Panel contains Generation Cache path, the Protection setting Cache Size, and Wiping Cache.
Before Wiping Cache, a warning dialog will throw. For safety, its recommended to backup your generation cache first. Note that this wipe only sends .wav files in the cache to your system’s Recycle Bin.
For further cleanup you will need to empty it — or on an accidental deletion, you have the option to restore it.


💡 New Feature: Consolidate Active Cells to Folder
In the Presets menu dropdown, there is a new action “Consolidate Active Cells to Folder“

This feature will export all 16 currently loaded cell’s samples to a user-chosen directory and updates references, protecting the session from accidental cache loss. This is also useful for transferring sessions across machines.

🔍 Troubleshooting High RAM or Crashes

If MACE seems to use RAM independently per instance or crashes frequently:

  1. Verify that MACEModelManager.dll (Windows) or MACEModelManager.dylib (macOS) is present: Windows: C:\ProgramData\Tensorpunk\Mace\dlls\tpv201_r
    Mac: /Users/Shared/Mace/lib
  2. Ensure it’s not blocked by antivirus or permissions.
  3. Reinstall MACE and run the installer as Admin (Windows) or from a verified .pkg (Mac).
  4. Still having trouble? Contact support: [email protected]

🧪 Final Notes

MACE 1.3.0 is a bleeding-edge tool pushing the boundaries of local AI audio generation inside a plugin. While optimized for stability and performance, it still demands modern hardware. We’re actively improving compatibility and performance across systems — and your feedback helps drive that forward.