Fixstars Launches Enhanced AIBooster with Advanced AI Diagnostics and Customizable Observability

Accelerating the Development Cycle with Deeper Performance Insights and Streamlined Model Optimization

IRVINE, Calif., Jan. 8, 2026 /PRNewswire/ -- Fixstars Corporation (TSE Prime: 3687, US Headquarters: Irvine, CA), a leading company in performance engineering technology, today announced the release of the latest version of its AI processing acceleration platform, "Fixstars AIBooster" (hereinafter "AIBooster").

The latest version of AIBooster introduces enhanced flexibility and deeper diagnostic capabilities: advanced diagnostic reports for AcuiRT — AIBooster's integrated framework for inference-optimized AI model conversion — and customizable performance observability with user-defined metrics and tags. These updates empower engineers to minimize observation overhead while accelerating the development cycle through rapid identification and refactoring of model conversion bottlenecks. Fixstars remains committed to providing the tools necessary for high-performance, cost-effective AI operations, ensuring that customers can achieve seamless and efficient AI deployment across any environment.

New Features of the Latest AIBooster

1. Advanced Diagnostic Reports for AcuiRT Model Conversion

AcuiRT now features significantly strengthened diagnostic capabilities. Engineers can now resolve deployment issues via CLI or detailed reports featuring:

• Conversion Result Visualization: Provides a detailed breakdown of conversion success and failure, including layer-by-layer success rates, specific error messages for failed layers, and inference accuracy comparisons for detecting potential degradation.
• Performance Profiling: Analyzes post-conversion characteristics, including overall inference latency and a layer-by-layer breakdown of processing time to pinpoint bottlenecks.

2. Customizable Performance Observability Tailored to User Environments

The performance observability feature in AIBooster provides a macro view of performance trends across clusters while monitoring hardware efficiency at the node and device levels. In this release, users can now easily customize the following parameters to achieve a more flexible observation experience:

• Metric Collection Intervals: Users can adjust the granularity of data collection to minimize overhead. For long-term trend analysis, increasing the interval reduces resource consumption, making it easier to deploy in diverse environments.
• User-Defined Tags for Traces: These tags allow users to classify workloads based on their own criteria, such as model type, execution settings, or specific datasets. This enables deeper performance analysis from unique perspectives.

Use Case: Faster and More Reliable Model Conversion for NVIDIA GPUs

A 2D object detection model (Detection Transformer: DETR) was converted from a trained PyTorch model to an NVIDIA GPU–optimized format (TensorRT), using AcuiRT. In the initial conversion attempt, only 16% of the model's layers were successfully converted, and because the model was only partially converted, inference performance was worse than before conversion. By leveraging a new diagnostic report feature to identify the failed layers and underlying errors, the model was refactored accordingly. As a result, 100% layer conversion was achieved with just four hours of refactoring, and the converted model demonstrated approximately a 1.25× improvement in inference speed.

About Fixstars Corporation
Fixstars is a technology company dedicated to accelerating AI inference and training through advanced software optimization solutions. It supports innovation in healthcare, manufacturing, finance, mobility, and other industries. For more information, visit: https://www.fixstars.com/

Media Contact
Public Relations, Fixstars Corporation
Email: [email protected]
Tel: (408) 400-3679

SOURCE Fixstars