SPEC Notice: Intel oneAPI DPC++compiler issues have resulted in over 2600 SPEC CPUs being invalidated in 2017 benchmark tests

SPEC ruled that these results use the compiler to perform a narrow range transformation through prior knowledge, specifically improving the benchmark performance of 523. xalancbmk_r/623. xalancbmk_s.

In order to encourage more widely applicable optimizations, SPEC will no longer disclose the results of using this optimization, but these data will be retained in the SPEC results database for historical reference.

The affected compiler versions are Intel oneAPI 2022.0 to 2023.0, so there are no issues with the data for the latest versions after 2023.0 and versions before 2022.0.

SPEC modified approximately 2600 records, including SPEC CPU 2017 results for Intel CPUs, and indicated that the results for Intel Xeon processors were invalid.

Michael Larabel from the Phoronix website believes that Intel's specially designed compilers can increase speed by up to 9%, while SPECint can increase speed by about 4%.

ServeTheHome stated that the results of the fourth generation Intel Xeon Sapphire Rapids were most affected when checking the relevant records of SPEC, and Intel's latest 2023.2.3 version has removed this special optimization.

SPEC was founded in 1988 and is a global third-party application performance testing organization composed of dozens of well-known universities, research institutions, and IT enterprises worldwide, including Stanford University, Tsinghua University, and Microsoft. It is committed to establishing, modifying, and certifying a series of standards for evaluating server application performance.

SPEC CPU 2017 is its latest set of CPU subsystem testing tools, which includes four categories and a total of 43 tests, covering multiple aspects such as integer operation speed, floating-point operation speed, SPECspeed 2017, SPECrate 2017 Integer, SPECrate 2017 Floating Point, etc.