Dell PowerEdge XE7740 supported up to 74 concurrent AI agents in Principled Technologies testing

Test report highlights results from 4-, 6-, and 8-GPU configurations on real-world agentic AI workloads in manufacturing and financial services.

Principled Technologies (PT) has released a new report evaluating the agentic AI performance of the Dell PowerEdge XE7740, a purpose-built enterprise AI server. PT testing measured how three GPU configurations of the server, powered by Intel Xeon 6747P processors and NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, handled real-world agentic AI workloads in the manufacturing and financial services industries.

PT tested three configurations of the Dell PowerEdge XE7740: one with 4 GPUs, one with 6 GPUs, and one with 8 GPUs. On a financial services workload with a 30-second latency SLA, the 8-GPU configuration of the PowerEdge XE7740 supported up to 74 concurrent AI agents and sustained 1,179 tokens per second. On a manufacturing workload with a 90-second SLA, the 8-GPU configuration of the PowerEdge XE7740 supported 15 concurrent agents and delivered more than double the throughput of the 4-GPU configuration.

Agentic AI represents a significant change in enterprise AI infrastructure demand. Where a traditional chatbot query might involve a single inference call, a single agent task can spawn hundreds. That workload increase falls on both the GPU and the CPU, making server platform choice critical for organizations planning production deployments of agentic AI.

Also Read: AiThority Interview with Matej Bukovinski, Chief Technology Officer at Nutrient

Key takeaways from the report included:
• More GPUs mean more agents—increasing GPU count also increased the number of agents the server could support
• Agentic AI demands CPU as well as GPU power—PT saw a wide range of CPU utilization, peaking at 99% during task assignment and agent orchestration
• On-premises ownership has the potential to offer a cost advantage—compared to Amazon Bedrock, the PowerEdge XE7740 offered a lower cost per million tokens ($/Mtok)

To test the server, PT used “a custom agentic benchmark that runs realistic AI-agent workflows end-to-end on a computing solution.” As PT notes in the report, “Because agentic AI agents can do a wide range of work, much of it industry- and company-specific, we built the benchmark to simulate a range of realistic industry-specific workflows. For this study, we used the financial services and manufacturing workflows, each of which incorporates eight workflow scenarios.”

Also Read: ​​AI systems – Interoperable AI systems: Connecting models across platforms