Search results for Signal65
Industry
Chipmakers publish their own AI agent performance metrics. Here is why
Chipmakers are introducing their own performance evaluation standards tailored to the AI agent era, saying theoretical compute speed and single benchmark scores do not capture real workloads completed through multiple steps. They are also disclosing measurement conditions and methods to leave room for verification. Qualcomm tested a new GPU and AI benchmark tool designed to show itemised results. Arm published a CPU methodology for agentic AI and released rack-level comparison results, though the new metrics remain in a validation phase.
Industry
Arm unveils CPU benchmarking methodology for agentic AI
Arm has introduced new CPU performance evaluation criteria for the agentic AI era and released a benchmarking methodology developed with Fujitsu, Meta and OpenAI. The approach measures systems by the amount of work agents actually complete while maintaining responsiveness. Arm also disclosed independent validation results for its Arm AGI CPU by Signal65, including modeled rack-level performance comparisons with AMD EPYC Turin and sandbox creation latency measurements.
Industry
CPU revives as AI agents spread, AMD and Intel rebound
Rising adoption of personal AI agents is increasing demand for server CPUs, putting AMD and Intel back in focus. As agents run autonomously for hours or days, some AI infrastructure workloads are expected to shift from GPUs to CPUs. Meta’s Muse and OpenAI’s Dots were shown running on AMD-based systems, though both companies use multiple CPU suppliers. AMD said data centre revenue more than doubled year-on-year in the June-ended quarter.