The evolution of Local Artificial Intelligence (AI) reaches a new level with the launch of MLPerf Client 1.0. This robust update represents a significant advance over the previous version (0.6), consolidating it as an indispensable tool for measuring the AI performance directly into your hardware. Its main highlight is the introduction of a user-friendly graphic interface (GUI)designed to democratize access to the benchmark AI, attracting everyone from casual users to professional developers and testers.

MLPerf Client 1.0 considerably expands support for several AI models and optimizes hardware acceleration on a wider range of devices. The tool now allows you to test the performance of popular large language models (LLMs)such as Llama 2 7B Chat e Llama 3.1 8B Instruct of the Target, and the Phi 3.5 Mini Instruct from Microsoft. This enhanced capability makes benchmark to evaluate the behavior of different artificial intelligences in various interaction scenarios.

In addition, the tool supports the experimental model Phi 4 Reasoning 14BThis is a clear example of the focus on performance testing with state-of-the-art language models which have a larger set of parameters and advanced features. MLPerf Client 1.0 has also improved the variety of types of promptsThis allows code analysis - a crucial feature for developers - and the measurement of content summarization with large context windows (4,000 or 8,000 tokens), an innovative experimental feature.

To download MLPerf Client 1.0 and start testing AI performance on your device, just go to the official GitHub page. This software is an initiative of the MLCommons consortium and its MLPerf Client working group, which collaborates with the main hardware and software suppliers to maintain a benchmark industry-standard customer service.

Required Hardware and Comprehensive Support

For hardware testers and enthusiasts, MLPerf Client 1.0 provides a more scalable and comprehensive set of workloads on multiple devices, allowing for an accurate assessment of the hardware's performance. local AI performance in various configurations. It's important to note that some workloads in this version require a GPU with 16 GB VRAM for efficient execution. This requirement underlines the tool's focus on prioritizing high-end hardware, expanding the scope beyond integrated graphics and NPUs.

The hardware and software stacks for Customer AI are dynamic, and there are countless ways to accelerate AI workloads locally. Fortunately, MLPerf Client 1.0 expands the coverage of these acceleration paths to more hardware than ever before, with a notable focus on devices Qualcomm e Apple.

Below, we detail the hardware and technologies supported, demonstrating the vast versatility of benchmark to AI acceleration:

• AMD's hybrid NPU and GPU via ONNX Runtime GenAI and the Ryzen AI SDK;
• GPUs AMD, Intel e NVIDIA via ONNX Runtime GenAI-DirectML;
• NPU and Intel GPU via OpenVINO;
• Hybrid NPU and CPU from Qualcomm Technologies via Qualcomm Genie and the QAIRT SDK;
• GPUs for Apple Mac via MLX;
• NPU and Intel GPU via Microsoft Windows ML and the OpenVINO execution provider;
• NVIDIA GPUs via Llama.cpp-CUDA;
• GPUs for Apple Mac via Llama.cpp-Metal.

A New Graphical Interface for Ease of Use

One of the most striking enhancements of the MLPerf Client 1.0 is the implementation of a graphical user interface (GUI). This new feature allows users to easily explore the full range of benchmarks available for their hardware, selecting and running the tests intuitively. This transition is an important milestone, as previous versions of MLPerf Client operated exclusively via the command line, making it less accessible.

In addition to easy selection, the new version offers real-time monitoring of a system's various hardware resources. This is essential so that users can quickly check whether the chosen execution path is using the GPU or NPU expected for the AI acceleration. This ease of use will certainly broaden the appeal of the benchmarkThis makes it accessible to both casual users who want to test their devices and professional hardware testers who need to collect and compare results on various hardware and software configurations.

The AI Scenario: Cloud vs. Local and the Relevance of the MLPerf Client

The exponential growth of Artificial Intelligence is undeniable, driving a real boom in the sector. Companies such as NVIDIA reaps rewards with record revenueswhile others, such as Intel, recognized challenges and face restructuring, including layoffs.

Although many technologies interact with the most advanced versions of AI models in the cloud - with leading services such as ChatGPT, Claude and Gemini remaining cloud-based - running AI models locally still has significant advantages. For reasons including privacy, research and control data, locally executed AI models remain of great interest to users and companies. In this context, the ability to reliably and neutrally measure the AI performance in client systems, using GPUs and NPUs, is absolutely crucial.

A Customer AI continues to be a vital development space, as suppliers of hardware e software work to define the types of workloads best suited to local execution and the ideal computing resources to support them. Tools such as MLPerf Client 1.0 are therefore essential for driving innovation and providing transparency in the promising local AI ecosystem.

Follow more news on artificial intelligence and technology:

• New NVIDIA-backed technology promises to reduce AI costs by up to 50% with DDR5 memory over Ethernet
• Intel XeSS 2 gains official support for AMD and NVIDIA GPUs
• Microsoft launches Copilot Mode for Edge on a trial basis