As financial services enterprises integrate AI into their operations, the demands on their compute infrastructures grow. For many organizations, this growth leads to higher cloud costs. Organizations now realize that cloud solutions might not meet the performance consistency or budget predictability necessary for their crucial workloads. This report, commissioned by Dell Technologies, examines whether on-premises servers—specifically Dell™ PowerEdge™ servers equipped with 4th Generation AMD EPYC™ processors and NVIDIA® GPUs—can deliver a more secure, better-performing, and more cost-efficient alternative to cloud-hosted environments for real-world AI workloads.
Prowess Consulting assessed the total execution time, cost per job, and anticipated total cost of ownership (TCO) of using a hybrid machine learning (ML) and AI pipeline. The pipeline used in this assessment ranges from classic sentiment analysis to innovative applications of transformer-based large language models (LLMs), and it is similar to pipelines used by banking firms for customer analytics and digital marketing. The pipeline mirrors enterprise applications such as customer experience management tools, social media solutions, and semantic search technologies. To conduct this assessment, we compared a Dell PowerEdge R7615 server and a comparable single-GPU instance in Microsoft Azure®. Our analysis measured raw performance and price-performance, while also evaluating operational efficiency, system versatility, and the ability for enterprises to optimize workloads and manage data governance.
Our findings indicate that Dell PowerEdge servers have the potential to surpass cloud-based equivalents in AI workload performance, even with less-powerful GPU resources. Furthermore, our study underscores the architectural strengths of on-premises servers. These strengths include lower latency, consistent resource availability, and tailored hardware configurations that can help avoid the expenses of cloud overprovisioning.
TL;DR
On-premises Dell™ PowerEdge™ servers featuring 4th Gen AMD EPYC™ processors and NVIDIA® GPUs deliver faster AI workload execution and dramatically lower costs compared to cloud-based alternatives. In a real-world financial services marketing pipeline, Dell PowerEdge R7615 servers achieved up to 16% faster completion time, 88% lower cost per run, and 58% lower three-year TCO versus Microsoft Azure®. These savings stem from tailored hardware configurations, reduced overprovisioning, and predictable budgeting. The study highlights the architectural flexibility and operational control of on-premises infrastructure for AI-driven analytics.
Evidence: see Figures 1–3 and “Study Methodology” in the source.
FAQ
Q: How does Dell™ PowerEdge™ performance compare to cloud for AI workloads?
A: Dell™ PowerEdge™ R7615 servers completed the AI pipeline up to 16% faster than a comparable Microsoft Azure® instance. This performance gain is due to optimized hardware configurations and reduced latency. See Figure 1 in the source.
Q: What are the cost advantages of running AI on-premises?
A: On-premises Dell™ PowerEdge™ servers delivered up to 88% lower cost per run compared to cloud. This is largely due to avoiding premium cloud GPU pricing and overprovisioning. See Figure 2 in the source.
Q: What is the total cost of ownership (TCO) difference over time?
A: Over a three-year horizon, Dell™ PowerEdge™ servers showed up to 58% lower TCO versus cloud, even with modest daily usage. See Figure 3 in the source.
Q: Why is financial predictability better with on-premises infrastructure?
A: On-premises deployments offer fixed hardware costs and eliminate variable cloud billing, giving finance teams better control over AI budgets. See “Financial Predictability” section in the source.
Q: How does hardware flexibility impact AI workload efficiency?
A: Dell™ PowerEdge™ servers allow right-sizing of components like GPUs, avoiding unnecessary costs. For example, the NVIDIA® L40S GPU used on-premises matched workload needs without the premium of an H100. See “Architectural Advantages” section.
Q: What practical AI use case was tested in this study?
A: The study used a real-world financial services marketing pipeline involving sentiment analysis and transformer-based embedding across 100,000+ documents. See “Real-World Use Case” section.
Explore more research from Prowess Consulting: https://prowessconsulting.com/resources/