AMD Ryzen AI MAX+ 395 “Strix Halo” APU Delivers Over 3x Performance Compared to RTX 5080 in DeepSeek R1 AI Benchmarks

AMD has recently unveiled its Ryzen AI MAX+ 395 “Strix Halo, ”an advanced APU that has outperformed the RTX 5080 by more than three times in the DeepSeek R1 AI benchmark tests. This impressive achievement highlights AMD’s commitment to enhancing AI processing capabilities in consumer-grade laptops.

The Game-Changing Ryzen AI MAX+ 395 APU: Redefining AI Performance

Press Release: The Ryzen AI MAX+ 395 APU marks a substantial upgrade in artificial intelligence processing for high-end laptops. Integrating Zen 5 CPU cores, a powerful 50 TOPS XDNA 2 NPU, and an advanced GPU, this processor sets a new standard for AI performance in sleek, lightweight devices. The APU specifically excels in executing local AI models, especially in applications utilizing large language models (LLMs), demonstrating significant advantages over competing products.

AMD Ryzen AI MAX+ 395 performance comparison

This innovative processor is tailored for practical AI tasks, exemplified by its performance with llama.cpp-driven applications and LM Studio. These tools significantly simplify the use of local LLMs, enabling users to seamlessly run the latest language models without needing extensive technical expertise.

Notable Advantages of the AMD Ryzen AI MAX+ 395 Compared to Copilot+ Competitors:

Performance Metrics

Achieves up to 2.2 times the token throughput relative to the Intel Arc 140V.
Offers up to 4 times faster time to the initial token for smaller models such as Llama 3.2 3b Instruct.
Delivers up to 9.1 times quicker processing for models with 7-8B parameters.
Outperforms the Intel Core Ultra 258V by 12.2 times for 14B parameter models.

Vision Model Performance Enhancements

Achieves speeds as much as 7 times faster in IBM Granite Vision 3.2 3b.
Provides up to 4.6 times the efficiency in Google Gemma 3 4b.
Delivers performance enhancements of up to 6 times in Google Gemma 3 12b.

Memory Capacity

Supports up to 128GB of unified memory compared to just 32GB max from competing models.
Allows the conversion of up to 96GB into VRAM via Variable Graphics Memory.
Facilitates the operation of larger models, like Google Gemma 3 27B Vision, which are beyond the capabilities of other APUs.

Remarkably, Strix Halo APUs can utilize a substantial full 128 GB of VRAM, while the RTX 5080 is limited to 16 GB. Even the higher-tier RTX 5090 boasts only 32 GB VRAM but operates at a higher power consumption level—360W for the 5080 and 575W for the 5090.

While the performance of NVIDIA’s RTX 50 discrete GPUs remains superior for tasks fitting within their VRAM limits, the Strix Halo APU maintains a competitive edge as an integrated all-in-one solution, making it an appealing choice for consumers focused on balancing performance with energy efficiency.

Source & Images