Revolutionizing Engineering Workloads: Clarios’ Cloud HPC Success with Azure and AMD

When Clarios—the global leader in advanced energy storage—set out to overhaul its high-performance computing (HPC) strategy, the destination was clear: faster insights, increased agility, and the resilience to keep innovating. Yet the road there was anything but certain. Like many manufacturing giants, Clarios’ critical R&D workloads were historically anchored to meticulously tuned, on-premises infrastructure. Simulation tools such as Ansys Fluent and LS-DYNA had shaped process and culture, powering the company’s ability to engineer one of every three vehicle batteries produced worldwide. Could cloud really match, or surpass, the relentless demands of electrochemical modeling and the need for rapid, iterative design? Clarios’ journey with the Microsoft Azure HPC Innovation Lab—developed in concert with Oakwood Systems Group, AMD, and Microsoft—offers compelling answers for every engineering-driven enterprise deciding between legacy certainty and digital reinvention.

The Legacy Barrier: Why On-Premises HPC Became a Bottleneck​

For years, Clarios thrived by keeping compute close to the hardware. HPC clusters—painstakingly configured and tightly controlled—ensured simulations ran at optimal performance, handling everything from cell chemistry to crash dynamics. But progress brought complexity. Exploding data volumes and proliferating design variants rendered even robust, modern hardware a constraint rather than a catalyst.
Scaling compute, testing next-gen architectures, and pivoting to rapid “what-if” analysis became difficult in traditional environments. Resource contention delayed workloads, hardware refresh cycles lagged behind simulation innovation, and costs escalated as capacity sat idle during workflow lulls. Worse, the operational friction discouraged experimentation and lengthened development timelines—an acute risk in fiercely competitive markets where speed and flexibility are everything.
The cloud appeared promising, but Clarios’ engineers had no illusions about its challenges: Would cloud-native solutions (especially those underpinned by AMD EPYC processors in Azure) deliver equivalent performance, transparency, and control? Could simulation workflows migrate without breaking, and would licensing or provisioning create new headaches rather than alleviate old ones?

The Azure HPC Innovation Lab: Redefining the Cloud HPC Experience​

Microsoft, AMD, and Oakwood’s response was both ambitious and practical—a dedicated “HPC Innovation Lab” for risk-free, hands-on cloud exploration. Instead of theoretical promises, the Lab delivered browser-based, preconfigured test environments mapped to “t-shirt sizes” (small, medium, and large), all running in Microsoft Azure on HBv4 virtual machines equipped with up to 176 AMD EPYC 9V33X cores and 3D V-Cache technology.
Key features included:

• Immediate, license-free access to popular simulation solvers like Ansys Fluent and LS-DYNA.
• Simple, intuitive user interface: engineers uploaded files, launched jobs, and analyzed results without negotiating with IT, managing licenses, or configuring infrastructure.
• Integrated telemetry and job analytics: each simulation produced detailed logs, resource charts, and performance summaries for transparent evaluation.
• Zero-commitment: trial runs carried no financial or operational risk.

This approach stripped away complexity, letting engineers concentrate on scientific questions—not logistical constraints—and laid the groundwork for radical performance benchmarking.

Performance Benchmarking: Disproving the Cloud Skeptics​

Perhaps the most consequential outcome of Clarios’ Innovation Lab engagement was the empirical, side-by-side benchmarking between legacy on-premises and cloud HPC. In one standout comparison, simulations that once monopolized physical clusters for 13–14 hours completed within a few hours on Azure HBv4 VMs—a reduction of over 85%. This leap shifted the innovation calculus: what was once a multiday cycle of simulation-debugging-restaging became a near-continuous design feedback loop.
Several factors drove this advantage:

• AMD EPYC Genoa-X processors with 3D V-Cache: These chips, exclusive to Azure’s latest HB-series, pair high core counts (up to 176 per VM) with massive on-chip cache. This design is particularly potent for memory-bound workloads common in computational fluid dynamics (CFD) and finite element analysis.
• Vertical scalability: Engineers could dial up compute “size” instantly, matching simulation needs to hardware with no wait for procurement or physical setup.
• Next-gen cloud fabric: Each Azure HBv4 node benefits from up to 350 GB/s memory bandwidth and high-throughput InfiniBand networking, confining latency to a minimum for distributed, data-intensive workloads.

Critically, Clarios’ team found that Azure’s streamlined job queueing, flexible resource allocations, and integrated solver images eliminated common on-prem bottlenecks around dependency management, software versioning, and file system contention.

Usability, Transparency, and Debugging: Engineering Buy-In​

One of cloud HPC’s traditional hurdles is cultural: engineers are often deeply invested in established workflows, toolchains, and validation rituals. Here, the Innovation Lab’s transparency was pivotal. Journal files, iteration snapshots, and real-time resource monitoring made it easy to validate results against historical data. Engineers independently confirmed solver parity, examined new speedups, and iterated rapidly.
Furthermore, Clarios and Oakwood collaborated on workload telemetry to prioritize cloud migration. Not every job is equally suited to cloud: tightly coupled MPI runs (historically sensitive to network latency) remained candidates for future optimization, while bursty high-compute simulations moved immediately to cloud bursting—a strategic approach that balanced risk, cost, and performance.

From Proof of Concept to Strategic Transformation​

What began as a technical test morphed into a cultural shift. The Lab’s zero-risk framework didn’t just prove that Azure could meet (and exceed) on-premises performance; it instilled organizational confidence. Engineers embraced the agility to explore new architectures, experiment with more variables per run, and react to product changes at speed.
Operationally, this shift empowered IT to reallocate resources, eliminate hardware maintenance overhead, and focus on optimizing cloud consumption. Financially, the ability to scale resource usage precisely—with no need to over-provision for peak demand—presented compelling cost-saving potential.

Azure HB-series VMs: Inside the Technology That Made It Possible​

At the core of this innovation wave lies the Azure HBv4 and HBv5 VM series—environments purpose-built for engineering and scientific applications demanding both raw throughput and memory bandwidth.

Technical Deep Dive: HBv4 and HBv5 Capabilities​

• CPU and Memory: HBv4 VMs feature up to 176 AMD EPYC 9V33X “Genoa-X” cores, each equipped with substantial 3D V-Cache, while HBv5 (leveraging custom AMD 9V64H CPUs with up to 352 Zen 4 cores) advances the memory conversation further by supporting up to 450 GB of HBM3 per VM—around 9 GB per core. This architecture is particularly game-changing for CFD, FEA, and other memory bandwidth-constrained workloads.
• Memory Bandwidth: HBv5 offers up to 7 TB/s per node—roughly eight times higher than typical enterprise server nodes. This enables CPUs to maintain full compute speed, with memory flows robust enough to keep pace with processor throughput.
• Networking: 800Gbps InfiniBand links—standard on new HB-series—mean distributed workloads can scale across nodes without bottlenecks, even in tightly coupled tasks.
• Security and Single-Tenant Design: By disabling simultaneous multithreading (SMT) and relying solely on physically isolated resources, Azure ensures that sensitive industrial workloads benefit from maximum security and predictable performance.

These technical advantages, engineered jointly by AMD and Microsoft, distinguish Azure from mainstream public cloud offerings, particularly in memory-bound and large-scale engineering applications.

A Critical Perspective: Strengths, Risks, and What IT Leaders Need to Know​

Notable Strengths​

• Performance at Scale: The Azure HBv4/5 series and exclusive AMD Genoa-X CPUs bring unparalleled speed to complex engineering workloads, with empirical results confirming improvements far beyond incremental gains.
• Operational Flexibility: Instant scaling, seamless access to latest solvers, and browser-based lab environments dramatically reduce friction and learning curves.
• Transparent Validation: Integrated logging, telemetry, and comparable job debugging build trust among engineering teams, addressing a perennial concern around reproducibility.
• Cost Effectiveness: Pay-as-you-go compute without capital investment or over-provisioning delivers new operational models for R&D—enabling businesses to run more experiments, more often, for less.
• Platform Ecosystem: Deep integration with Azure native tools (storage, AI/ML, analytics) future-proofs investments and opens doors to further automation and workflow orchestration.

Potential Risks & Cautions​

• Workload Suitability: Not every legacy simulation is cloud-ready. Some tightly coupled MPI jobs—where latency is mission-critical—may still perform best on-prem or require careful tuning and future cloud network enhancements.
• Data Egress/Compliance: As with any cloud migration, teams must address data governance, IP protection, and regulatory mandates for sensitive or regulated industries—a non-trivial consideration when computations span global cloud regions.
• Vendor Lock-In: Deep reliance on unique Azure hardware, software, and APIs (including custom AMD silicon) could raise future concerns about cross-platform portability and exit strategies if business needs change.
• Cultural Change Management: While Lab environments simplify pilot adoption, full production deployment requires upskilling, change management, and careful cost governance to avoid unanticipated expenses.

Industry Trends: Why This Matters Beyond Clarios​

Clarios’ cloud HPC transformation arrives as part of a broader wave. From semiconductor firms accelerating chip verification with EDA tools on Azure NetApp Files, to automotive giants simulating advanced driver-assistance systems, to climate scientists operating at petascale, the cloud’s “memory wall” is cracking.
Microsoft’s partnership ecosystem, exemplified by Oakwood and AMD, is rolling out pre-validated, industry-targeted blueprints that remove most of the legacy risks and unknowns. Programs like the HPC Innovation Lab aren’t just technical showcases—they’re confidence builders and cultural accelerators for cloud transformation.

Strategic Recommendations for Windows Ecosystem Decision-Makers​

For Engineering Leaders​

• Start With a Cloud Lab: Hands-on, risk-free labs offer the quickest path to technical and organizational buy-in. Run your actual workloads, benchmark real performance, and involve skeptical users from day one.
• Analyze and Prioritize: Not all workloads benefit equally. Use telemetry and performance analytics to identify which simulations are best suited for cloud bursting versus those requiring further optimization.
• Validate and Scale: Focus on proof of parity and pilot projects that can be iterated quickly. Once initial targets are migrated successfully and cost metrics are established, scale adoption to other engineering and R&D teams.

For IT Architects and Security Teams​

• Assess Data Flows and Compliance: Engage legal and security stakeholders early. Understand where data will reside and map compliance requirements to cloud provider capabilities.
• Plan for Hybrid: The future is likely hybrid; keep critical tools and applications portable and make cloud bursting a complement, not just a replacement, for on-premises investments.

For Finance and Operations​

• Monitor Actual Usage: Cloud’s “pay-per-use” model brings agility but can also mask creeping costs. Use dashboards and alerts to track spending by project and optimize VM types to workload profiles.
• Leverage Reserved Capacity and Storage Tiers: Take advantage of cool storage tiers and reserved instance savings, especially for simulations with large but infrequently accessed data sets.

Conclusion: Cloud HPC Arrives—On Engineering’s Terms​

Clarios’ bold step into cloud HPC, catalyzed by the Innovation Lab, signals a pivotal shift. Advanced manufacturing and engineering organizations can now access performance parity—or better—in the cloud, unlocking new speed, flexibility, and resilience without losing the precision or control that made their legacy systems world-class. Yet this isn’t just a technology story: it’s a template for digital reinvention, led not by IT, but by empowered engineers and scientists redefining what’s possible in the next wave of innovation.
For IT leaders, the Clarios case is a beacon: start with real pilots, demystify the cloud, validate results openly, and chart a future where shrinking R&D cycles and expanding design freedom become the new normal. As Microsoft, AMD, and the broader Azure ecosystem continue to push boundaries—adding more cores, faster memory, and increasingly sophisticated workflow tools—the journey from legacy constraint to continuous innovation has never looked more attainable.
Ready to pilot your own transformation? The answer may not lie in the datacenter you already own, but in the next simulation job you run—in the cloud, on your terms.

---

Source: HPCwire Clarios Accelerates Cloud HPC Adoption Through Innovation Lab on Microsoft Azure - HPCwire