Intel is leading the AI PC revolution by achieving full NPU support in the industry-first MLPerf Client v0.6 benchmark.

Intel today released the first standardized benchmark results measuring client NPU performance in the new MLPerf Client v0.6 evaluation.

In this evaluation, Intel® Core™ Ultra series 2 processors demonstrated industry-leading performance in CPU, GPU, and NPU-based AI computing.

In particular, it demonstrated powerful performance, generating output that exceeds the typical human reading speed on both GPU and NPU.

“Intel is leading the industry in implementing NPU acceleration for AI workloads on client PC platforms,” said Daniel Rogers, general manager of Intel PC Product Marketing. “This achievement is the result of Intel’s ongoing efforts and software optimizations to drive innovation in AI PCs.”

Intel's Core Ultra Series 2 processors redefine the AI PC experience by delivering unprecedented AI compute performance across CPU, GPU, and NPU.

MLPerf Client v0.6 evaluates four content creation and summarization tasks based on the Llama 2 7B model, with Intel's NPU and built-in Intel® AIndustry-leading performance with rc™ GPUs.

Intel achieved a record-breaking NPU response time, taking just 1.09 seconds to generate the first token, enabling instantaneous response.

It also recorded a processing speed of 18.55 tokens per second, demonstrating its ability to support real-time AI interactions without interruption.

In terms of GPU performance, Intel also achieved a faster first token creation speed than its competitors, solidifying its leadership in AI acceleration technology utilizing NPU and GPU.

MLPerf Client v0.6 expands the evaluation scope beyond the existing GPU-centric tests to include NPU-specific benchmarking.

This benchmark is a test jointly developed by MLCommons consortium companies including Intel, AMD, Microsoft, Nvidia, and Qualcomm, and presents a new standard for evaluating AI computing performance.

Intel was the only processor to fully meet the NPU evaluation requirements in the final benchmarks, based on collaboration between the NPU hardware team and the OpenVINO™ software team.