Cornelis today announced an expanded collaboration with AMD to help HPC and AI for science customers design, validate and deploy systems pairing AMD EPYC™ CPUs and AMD Instinct™ GPUs with Cornelis® networking. The program combines AMD compute platforms and engineering with the Cornelis fabric, so customers get more choice in compute, systems validated on their own applications, and a direct path to buy and deploy through the OEMs and system integrators they already use.
“Every AMD EPYC generation puts more cores in a node, and every one of those cores needs its own path to the network,” said Lisa Spelman, CEO of Cornelis. “When the network can’t keep up, you pay for compute that sits waiting. The fix is to pair the network to the workload, because no two codes stress the fabric in the same way. Our expanded collaboration with AMD does that for EPYC and Instinct platforms, validated together and available through the OEMs and integrators customers already trust. GWDG shows what this looks like in production, across thousands of research codes.”
Global design and support for customers’ workloads
AMD and Cornelis are bringing engineering, field and partner teams together for each customer engagement, from architecture design and benchmarking through acceptance testing and production. Customers buy the combined solution through their preferred OEMs and system integrators and can run their own code at the AMD HPC Center of Excellence in Munich.
The program reflects AMD’s open approach to HPC and the way Cornelis works with system partners. Customers choose their compute, their network and their system builder, backed by AMD engineering and the AMD ROCm™ open software stack .
Cornelis CN5000 is qualified and in production on AMD EPYC 9005 Series platforms, validated on more than 25 platform configurations, and deployed at university, research, government and industrial sites across North America and Europe. Cornelis has also extended CN5000 qualification to AMD Instinct MI350X Series GPUs.
The program targets scientific and engineering workloads where compute and networking must scale together. These include computational fluid dynamics, climate and weather, life sciences, physics, materials science and other FP64-intensive codes. Joint pursuits span national laboratories, university research computing, industrial and automotive engineering, and sovereign supercomputing in Europe. AMD and Cornelis will publish joint reference designs and application performance data for leading ISV software and extend the program to each new generation of AMD EPYC and AMD Instinct platforms.
“AMD is the open platform behind the world’s leading supercomputers, and our strategy is simple: give customers leadership compute, leadership precision and the freedom to build around it with the partners they trust. EMMY P4 shows what that looks like in production, and with Cornelis we’re giving more HPC customers a validated path to scale, from traditional simulation to AI for science,” said Scott Tease, Corporate Vice President, HPC and AI Business Development, AMD.
Deployed at GWDG
At GWDG in Göttingen, Germany, the new EMMY P4 supercomputer runs approximately 300 dual AMD EPYC 9745 CPU nodes on a Cornelis CN5000 Omni-Path® fabric. It serves NHR@Göttingen, one of nine centers in Germany’s National High Performance Computing (NHR) Alliance, and supports thousands of research codes across life sciences, physics, quantum computing, AI and the digital humanities. EMMY P4 adds capacity to an environment that runs approximately 3.5 million HPC jobs a year.
EMMY P4 joins a group of large European research systems built on AMD processors, including LUMI in Finland, Eni’s HPC6 and HPC7 in Italy, Hunter at HLRS in Germany, and the Alice Recoque exascale system in France.
“GWDG exists to give researchers the computing they need to advance science, whatever their field. On one shared system, we run thousands of different codes, from life sciences and physics to AI and the digital humanities. Each one puts different demands on the system. AMD EPYC processors and the Cornelis fabric give us the compute density, low latency and consistent performance to serve them well, so our users can focus on their research,” said Prof. Dr. Julian Kunkel, Deputy Head of High-Performance Computing at GWDG.
Leadership performance across HPC workloads
AMD EPYC processors with Cornelis CN5000 have demonstrated strong performance and scaling across scientific and engineering applications including CONVERGE CFD, Ansys Fluent, OpenFOAM, Siemens Simcenter STAR-CCM+, and WRF. Cornelis has published its application-level performance results across these workloads, providing customers with additional data as they evaluate production configurations.
These HPC codes exchange large volumes of small messages, so performance depends on network latency and message rate as much as on compute. On an AMD EPYC 9755 CPU system, Cornelis testing found that STAR-CCM+ scaled better than linearly, running on average nearly nine times faster on eight nodes than on one. Across 132 CONVERGE CFD workloads, CN5000 delivered 20% higher average throughput than NVIDIA NDR InfiniBand at four nodes. For customers, that means faster time-to-results and more simulations from every node they deploy.[1]
Ready for the next generation of AMD EPYC and AMD Instinct
The collaboration extends to the next generation of both companies’ technology. AMD EPYC 9006 Series CPUs is the first PCIe 6.0 server CPU, and the CN6000 PCIe 6.0 x16 interface gives an 800G adapter the host bandwidth that today’s HPC and AI applications require. CN6000 can give every core in a 512-core, dual-socket server its own dedicated connection to the network, so applications can use every core without cores competing for network access. It delivers an industry leading 1.6 billion messages per second and, in Cornelis measurements, 23% lower MPI latency than NVIDIA XDR InfiniBand.
The codes this program targets run in double precision, and AMD Instinct GPUs support FP64 in hardware, with the AMD Instinct MI430X GPU design to deliver up to 288 TFLOPS of native FP64.[2] Cornelis is also validating CN6000 with next-generation AMD Instinct platforms. OEM systems pairing CN6000 with AMD EPYC 9006 Series platforms are expected beginning in Q1 2027.
Getting Started
Customers can engage with the program through the OEM or system integrator they already work with, who can scope a validated AMD and Cornelis configuration for their workloads. Organizations that want to talk with the joint team directly can reach Cornelis at AMD@cornelis.com or AMD at HPCAI@AMD.com. Reference designs, application performance data and program details are available at www.cornelis.com/partners/amd. Cornelis will show AMD EPYC platforms with Cornelis networking at SC26 in Chicago, Nov. 15–20, at booth #1419.
Performance notes: [1] Application throughput and parallel efficiency figures are from Cornelis white papers A01521 (August 2026) and A01562 (October 2026). STAR-CCM+ and CONVERGE CFD results in the body are from the same papers also found at https://www.cornelis.com/resources/white-papers. All CN5000 and CN6000 figures are solely based on internal measurements and projections by Cornelis Networks. [2] MI430X figures are AMD peak theoretical specifications, FP64 peak, as published by AMD. Results may vary.
About Cornelis
Cornelis delivers high-performance, scale-out and scale-up networking solutions that accelerate AI and HPC workloads. Cornelis technology enables lossless, congestion-free networking that reduces training time, improves inference, and maximizes compute utilization. From foundation model training to complex climate modeling and real-time analytics, Cornelis solutions power the most demanding workloads across commercial, academic, government, and cloud environments. With a focus on performance, scalability, and efficiency, Cornelis helps organizations achieve faster insights and greater return on infrastructure investments. Learn more at www.cornelis.com.
AMD, the AMD Arrow logo, EPYC, Instinct, ROCm and combinations thereof are trademarks of Advanced Micro Devices, Inc. Other names are for informational purposes only and may be trademarks of their respective owners.
FAQ
What are Cornelis and AMD announcing? Cornelis and AMD are expanding their collaboration with a joint go-to-market program to help HPC and AI for science customers design, validate and deploy AMD EPYC CPU and AMD Instinct GPU platforms with Cornelis networking.
What does the program offer customers? The program brings together AMD and Cornelis engineering, field and partner teams for platform design, benchmarking, validation and deployment. Customers can work through their preferred OEM or system integrator and run their own code at the AMD HPC Center of Excellence in Munich. The collaboration builds on more than 25 validated AMD platform configurations with Cornelis networking.
Where are AMD and Cornelis technologies deployed today? GWDG’s EMMY P4 supercomputer uses approximately 300 dual AMD EPYC 9745 nodes with Cornelis CN5000 networking to support NHR@Göttingen and thousands of research codes across scientific disciplines.
Which Cornelis products work with AMD platforms? Cornelis CN5000 runs in production on AMD EPYC 9005 Series platforms today, including GWDG’s EMMY P4. Cornelis CN6000 is designed for AMD EPYC 9006 Series ‘Venice’ and PCIe 6.0.
Will the collaboration include CN6000? Yes. Cornelis is validating CN6000 with AMD EPYC 9006 Series and next-generation AMD Instinct platforms ahead of OEM system availability beginning in Q1 2027.
View source version on businesswire.com: https://www.businesswire.com/news/home/20261110694456/en/
Media gallery
