Dell server supports 205 AI agents in PT tests

Principled Technologies tested Dell PowerEdge XE9780LAP systems with Intel Xeon 6979P processors in CPU-only and NVIDIA B300 GPU-accelerated configurations across manufacturing, financial services and coding agentic AI workloads. The report says the server reached up to 205 concurrent agents and highlights how infrastructure needs rise as agentic AI workloads get more complex.

Why it matters: - Agentic AI workloads can require hundreds or thousands of interactions among models, tools, data sources and orchestration layers. - Infrastructure sizing becomes a core deployment decision as organizations move from early agent use to more complex multi-step workflows. - PT’s results show a single Dell PowerEdge XE9780LAP platform can scale from CPU-only use cases to GPU-accelerated agentic AI.

What happened: - Principled Technologies released a report on the Dell PowerEdge XE9780LAP server with dual Intel Xeon 6979P processors. - PT tested the server with eight NVIDIA B300 GPUs and in a CPU-only configuration. - The testing used a custom agentic AI benchmark covering manufacturing, financial services and coding workflows. - PT said the tests ran on realistic workloads rather than synthetic prompts.

The details: - With eight NVIDIA B300 GPUs, the server supported 205 AI agents on an early-stage manufacturing workload while meeting SLA thresholds. - The same configuration supported 62 agents on a late-stage financial services workload. - The server supported 14 agents on a complex code generation workload while meeting SLA thresholds. - The manufacturing workload reached up to 3,139 output tokens per second. - PT calculated cost per million tokens at $0.25 for the coding workload and $2.05 for a complex mixture-of-experts financial workload, using server purchase price and energy costs. - On CPUs alone, with Intel Advanced Matrix Extensions enabled, the two Intel Xeon 6979P processors supported 57 agents on a moderate financial workload within a 30-second SLA. - With Intel AMX disabled at FP32 precision, the CPU-only system could not sustain even one agent within the SLA. - PT tested eight manufacturing tasks, eight financial services tasks and 16 coding tasks. - The tasks included predictive maintenance, fraud investigation and autonomous bug fixes in a real code repository. - In the study, the Intel Xeon 6979P processors handled orchestration, tool invocation, policy enforcement and data movement, while the NVIDIA B300 GPUs handled model inference and token generation.

Between the lines: - The report suggests agentic AI capacity depends heavily on both workload complexity and accelerator choice. - PT’s CPU-only results indicate GPUs are not mandatory for every early deployment, but performance headroom drops fast as agent workloads become more demanding. - The report argues that organizations that trust agents to do more will need more hardware. - The findings also frame the Dell platform as a scale-up path for teams that want to start with CPU-only agentic AI and expand later.

What’s next: - PT points readers to the full report and infographic for more detail. - The report concludes that Dell PowerEdge XE9780LAP servers with Intel Xeon 6979P processors and NVIDIA B300 GPUs are positioned to support agentic AI infrastructure needs across organization sizes. - Principled Technologies says the CPU-only configuration can serve as a starting point for earlier-stage or lower-volume agentic AI use.

The bottom line: - PT’s tests show the Dell PowerEdge XE9780LAP can run from modest CPU-only agentic AI to much larger GPU-accelerated deployments, with capacity rising sharply as hardware scales.