By Key Points
- AMD claims its Epyc Zen 4 cores outpace Nvidia’s Grace CPU Superchips.
- AMD’s testing suggests 2.75x efficiency over Nvidia’s offerings.
- Nvidia’s Grace CPUs are primarily for HPC and GPU support.
- Performance varies based on workloads and specific use cases.
In a bold move, AMD has asserted that its Epyc Zen 4 processors are significantly more powerful and efficient than Nvidia’s Grace CPU Superchips. But is this a straightforward win for AMD, or is there more to the story?
AMD’s Epyc Zen 4 Takes the Lead
AMD recently released its benchmarks, claiming that its Epyc Zen 4 processors outperform Nvidia’s Grace CPU Superchips by a wide margin. According to AMD, their processors are not only twice as fast but also up to 2.75 times more efficient.
These claims come after rigorous testing against Nvidia’s 2022 Grace CPU Superchip, a device featuring 72 Arm Neoverse V2 cores per CPU die, connected with a 900GB/sec NVLink interconnect, and supported by up to 960GB of LPDDR5x memory. However, AMD’s tests reportedly used the 480GB version of the Grace Superchip.
AMD claims Nvidia’s Grace CPU Superchip, Arm are no match for its Epyc Zen 4 cores https://t.co/ta0X0Ob7kW pic.twitter.com/HqS0YXRb7q
— Eric Vanderburg (@evanderburg) July 23, 2024
Impressive Benchmarks
In a series of ten different workloads, including general-purpose computing, transactional databases, and high-performance computing (HPC), AMD’s single- and dual-socket systems, equipped with Epyc 4 Genoa (9654) and Bergamo (9754) processors, claimed a performance edge ranging from 1.5x to 4x over Nvidia’s Grace CPU Superchips.
Specifically, in the SPECpower-ssj2008 benchmark, a single 128-core Epyc 9754 demonstrated approximately 2.5 times better performance per watt compared to Nvidia’s chip, with dual Bergamo Epycs boosting that advantage to 2.75 times.
AMD also challenged the notion that Arm systems are inherently more energy-efficient, showing substantial performance-per-watt gains in various tests.
The Bigger Picture: Nvidia’s Focus on HPC and AI
While AMD’s performance claims are noteworthy, they might not paint the complete picture. Nvidia’s Grace CPU Superchips are predominantly designed for high-performance computing (HPC) and AI applications, often acting as a support system for GPUs rather than serving as standalone compute units.
Workload-Dependent Performance
Nvidia’s benchmarks reveal a different story, with Grace CPU Superchips achieving anywhere from 90% to 2.4x the performance of dual 96-core Epyc 9654s in various cloud and HPC services.
The Grace Superchip is optimized for feeding data to GPUs, making it highly effective in specialized environments like AI and data analytics.
This is evident in systems such as the UK’s Isambard-3 and Isambard-AI supercomputers, where Grace CPUs handle data processing for large-scale AI models.
Moreover, Nvidia’s Grace-Hopper (GH200) configuration, which pairs a Grace CPU with a powerful H100 GPU, is designed for maximum computational efficiency.
This setup is particularly advantageous for applications that require intense data processing, such as those used by Mistral AI and Moore Threads, the Chinese GPU manufacturer.
Strategic Deployment
Nvidia has strategically positioned its Grace CPUs within the broader scope of its GH200 Superchips, which emphasizes the synergy between CPUs and GPUs.
This design philosophy reflects Nvidia’s focus on AI and HPC applications, where the GPU’s computational prowess is paramount, and the CPU’s role is to efficiently manage data throughput.
This approach was evident at Nvidia’s GTC developer conference, where the company highlighted its next-generation GB200 Superchips.
In essence, while AMD’s Epyc Zen 4 processors may outperform Nvidia’s Grace CPU Superchips in certain benchmarks, the overall effectiveness of these processors depends on the specific workload and use case.
Nvidia’s Grace CPUs are not intended to compete head-to-head with AMD’s offerings in general-purpose computing but are instead tailored for specialized tasks in HPC and AI.
AMD’s recent claims showcase the impressive capabilities of its Epyc Zen 4 processors, but the true measure of performance lies in the specific applications and environments in which these processors are deployed.
For general-purpose and high-efficiency computing, AMD’s Epyc processors offer significant advantages. However, in the realm of HPC and AI, Nvidia’s Grace CPUs, with their optimized data processing and GPU support, remain formidable contenders.
To learn more about Nvidia’s ventures, check out our coverage on Mistral AI and Nvidia’s Revolutionary 12B NeMo Model, Moore Threads: The Chinese GPU Manufacturer, and Nvidia Faces Supreme Court in Crypto.
