New benchmarks show AMD EPYC CPUs outperform Nvidia Grace in AI and data center workloads, while Ryzen AI MAX+ 395 redefines laptop AI.
AMD showcased competitive benchmarks for its EPYC CPUs against Nvidia’s Grace processors. They also unveiled the Ryzen AI MAX+ 395 processor for consumer laptops and introduced the space-qualified Versal AI Edge adaptive System-on-Chip (SoC).
AMD EPYC CPUs Take the Lead Over Nvidia Grace in AI Performance Tests
AMD presented benchmark comparisons demonstrating clear performance advantages of its EPYC processors over Nvidia’s Grace CPUs, particularly in data center and AI workloads. EPYC processors, leveraging mature x86 architecture, offer robust performance, scalability, and compatibility advantages compared to newer Arm-based solutions like Nvidia Grace.
AMD highlighted several key benchmarks showcasing EPYC processors’ advantages over Nvidia’s Grace, including:
- Unparalleled Density: EPYC processors offer 33% higher density and 192 core (per socket) SKUs than Grace Superchip’s 144 arm Neoverse cores.
- Power Efficiency: In dual-socket SPECpower® tests, EPYC CPUs delivered 2.75x better power efficiency, making them significantly more energy-efficient for large-scale deployments.
- Database Performance: EPYC processors demonstrated over twice the performance in MySQL TPROC-C workloads. This is increasingly significant as we deploy more autonomous RAG systems that require substantial vector databases.
- Video Encoding: When encoding video using the FFmpeg VP9 codec, EPYC achieved 2.9x higher throughput, enabling faster and more efficient media processing.
- GPU-Accelerated Workloads: EPYC CPUs provided up to 20% better throughput than other leading x86 solutions, making them particularly effective for GPU-intensive AI tasks.
- Clock Speeds: With clock speeds reaching up to 5GHz, EPYC processors significantly outpaced Nvidia Grace’s 3.1GHz base frequency, offering superior raw performance for compute-intensive applications.
- Resource Utilization: AMD’s Simultaneous Multithreading (SMT) technology enhanced resource efficiency further, allowing EPYC processors to handle more tasks simultaneously and effectively.
These benchmarks showcase EPYC’s suitability for diverse enterprise AI applications, including real-time recommendation systems, predictive maintenance, and vision and language processing tasks.
Ryzen AI MAX+ 395 Sets New Consumer Laptop AI Performance Standards
AMD also introduced the Ryzen AI MAX+ 395 processor (“Strix Halo”), positioned as the most potent x86 APU for premium thin-and-light laptops. It combines 16 Zen 5 CPU cores, an XDNA™ 2 Neural Processing Unit delivering over 50 TOPS, and integrated graphics with 40 AMD RDNA™ 3.5 Compute Units.
AMD’s Benchmarks show the Ryzen AI MAX+ 395 significantly outperforming Intel’s Copilot+ processors across various AI workloads. In token throughput tests, AMD’s new processor delivered up to 2.2 times the performance of laptops equipped with Intel’s Arc 140V. The latency improvements are particularly impressive: smaller models like Llama 3.2, which has 3 billion parameters, ran up to four times faster. Mid-sized models with 7 to 8 billion parameters experienced up to 9.1 times quicker response times. In comparison, larger models with around 14 billion parameters achieved latency reductions of up to 12.2 times compared to Intel’s Core Ultra 258V.
The Ryzen AI MAX+ 395 also excelled in vision-based AI tasks, running IBM’s Granite Vision models (3.2 billion parameters) up to seven times faster and Google’s Gemma Vision models (4 billion parameters) up to 4.6 times faster. Even larger vision models, like Google’s Gemma Vision with 12 billion parameters, saw performance boosts up to six times compared to Intel’s offerings.
Beyond raw performance, AMD’s new processor supports an impressive 128GB of unified memory, surpassing the 32GB maximum competitors offer. Additionally, AMD’s Variable Graphics Memory technology allows laptops to dynamically allocate up to 96GB of this memory as VRAM. This flexibility means users can comfortably run large-scale AI models directly on their laptops, including Google’s Gemma Vision with up to 27 billion parameters.
The Ryzen AI MAX+ platform dramatically enhances user experiences by allowing individuals to run advanced local language and vision models directly on their laptops, eliminating the need for technical expertise, as demonstrated by applications like LM Studio.
Versal AI Edge XQRVE2303 Adaptive SoC Achieves Spaceflight Qualification
In another groundbreaking announcement, AMD revealed that its Versal AI Edge XQRVE2303 adaptive System-on-Chip (SoC) has completed Class B spaceflight qualification, becoming AMD’s second radiation-tolerant Versal device qualified for space missions.
The XQRVE2302 is designed to bring advanced AI inferencing capabilities to space with its enhanced AMD AI Engines (AIE-ML). These engines deliver double the INT8 and 16 times the BFLOAT16 performance compared to first-generation AI Engines, with reduced latency and improved memory bandwidth. Its compact 23mm x 23mm form factor makes it the smallest adaptive SoC for space applications, offering high performance while reducing power consumption and board space.
Equipped with a dual-core Arm® Cortex®-A72 application processor, a dual-core Arm Cortex-R5F real-time processor, DSP blocks, and FPGA programmable logic, the XQRVE2302 is ideal for edge processing tasks such as image classification, autonomous navigation, and sensor data processing. These capabilities enable applications like wildfire detection, vegetation monitoring, and cloud detection for earth observation satellites.
AMD partnered with Alpha Data to develop a radiation-tolerant reference design for the XQRVE2302. The ADM-VB630 design platform facilitates rapid prototyping and cost-effective development of next-generation satellite systems.
The Versal XQR Series is designed for complementary roles in space systems. While the larger XQRVC1902 handles heavy-duty signal processing, the smaller XQRVE2302 focuses on command and control functions, AI inferencing, and edge computing. Both devices support unlimited reprogramming during development and post-deployment in space’s harsh radiation environment.
Conclusion
AMD’s latest announcements underscore its strong positioning in both enterprise and consumer markets. With EPYC processors clearly outperforming Nvidia’s Grace CPUs across critical benchmarks—especially in data centers and AI workloads—and the Ryzen AI MAX+ 395 setting new standards in consumer laptop AI performance, AMD demonstrates its ability to deliver robust, efficient, and high-performance computing solutions across diverse applications.
Additionally, the Versal AI Edge XQRVE2303 adaptive SoC achieving spaceflight qualification highlights AMD’s commitment to pushing technological boundaries even further. By offering powerful, versatile, and radiation-tolerant AI processing capabilities for space missions, AMD is expanding possibilities in aerospace, making advanced computing accessible even in challenging environments. These innovations collectively position AMD at the forefront of AI and high-performance computing advancements.
Engage with StorageReview
Newsletter | YouTube | Podcast iTunes/Spotify | Instagram | Twitter | TikTok | RSS Feed
