MobbleOpen in Mobble ⇢
Technology · Software & cloud · published 2026-11-21 · via Tom's Hardware

AMD and Nvidia Showcase Next-Gen HPC Tech at SC26 Conference

Image via Tom's Hardware
Image via Tom's Hardware

At the SC26 supercomputing conference, AMD and Nvidia highlighted advancements in high-performance computing, with AMD promoting its EPYC Venice CPUs for agentic workloads and Nvidia defending its Vera CPU as purpose-built for AI. Nvidia’s Ian Finder clarified that the company did not predict the data center boom but leveraged design expertise to build the processor. The event underscored ongoing competition in AI and HPC infrastructure.

Expanded Detail

At the SC26 supercomputing conference in November 2026, AMD and Nvidia unveiled new hardware targeting the growing intersection of high-performance computing (HPC) and artificial intelligence. AMD highlighted its EPYC Venice CPUs, emphasizing their strength in agentic workloads, and reported that its technology now powers over 40% of the new TOP500 supercomputers—up 14% from the previous year. The Lux supercomputer, developed with the U.S. Department of Energy, debuted at position 14 on the list, powered by EPYC CPUs and Instinct MI355X GPUs. Meanwhile, Nvidia introduced the Rubin HPC GPU, boasting over 50,000 CUDA cores—nearly double the standard Rubin—and positioned it for agentic workflows and scientific simulations. The company also defended its Vera CPU, with chief Ian Finder stating that predicting the data center boom was “absurd,” instead crediting design expertise and memory bandwidth as key to its development. Nvidia acknowledged that software optimization tailored to its Olympus-based Arm cores can boost performance by up to 39% in certain workloads, though the full extent of gains beyond matrix multiplication remains unclear.

The competition reflects deeper industry shifts: while AI-driven data centers are consuming vast compute capacity, AMD maintains that traditional HPC workloads—such as high-precision simulations—remain essential and will not be fully replaced. Both companies are investing in heterogeneous compute stacks, suggesting a future where CPUs, GPUs, and specialized accelerators work in tandem. What remains unresolved is how widely these new chips will be adopted outside of niche HPC environments, and whether software ecosystems will evolve fast enough to unlock their full potential. The lack of public benchmarks for Rubin HPC and limited details on Vera’s real-world performance in non-optimized scenarios mean their long-term impact is still uncertain.

Context

The developments at SC26 signal a broader trend where AI and HPC are converging, reshaping how institutions and corporations deploy computing resources. As AI clusters absorb much of the new hardware, traditional HPC centers may face challenges in securing funding and infrastructure, potentially affecting scientific research in climate modeling, materials science, and nuclear simulation. The dominance of AMD in the new TOP500 list suggests growing confidence in its ecosystem, particularly in government and defense applications, while Nvidia’s push into CPU design reflects a strategic pivot to own more of the compute stack. This could strengthen its position in AI-driven data centers but may also raise concerns about vendor lock-in and software fragmentation. A comparable precedent is the early 2010s shift toward GPU acceleration in HPC, which initially faced skepticism but eventually became standard. The current situation may follow a similar arc, with performance gains driving adoption despite early uncertainty. What to watch for is whether open standards or cross-platform compatibility will emerge to prevent fragmentation, and how quickly software developers can adapt to new architectures like Vera’s Olympus cores.

Expanded detail and Context are AI-generated analysis; the linked article remains the authoritative source.
Read the full article at Tom's Hardware →
Related stories
Nexalus and Renda Unveil Liquid-Cooled AI Server for Retrofitting Existing Data Centers · Artificial intelligence
PowerCompute Trials AI and HPC Workloads at Oklahoma Cryptomine Facility · Artificial intelligence
This summary is Al-enhanced to contain extended analysis and broader social context. The original is {NAME); the linked article is the authoritative source. Original headline: “This week on Tom's Hardware Premium: 21 November 2026.” Browse more stories.