The "Standard Performance Evaluation Corporation" organizes standardized tests of, among other things, CPUs.
“Up to 9 percent better performance in certain tests”
Now they believe that Intel is cheating with the company's "oneAPI DPC++" which in turn affects the test results of the tests "523.xalancbmk_r" and "623.xalancbmk_s." We are not talking about consumer CPUs, but among other things the Xeon Platinum 8480+ which costs over NOK 200,000 (so a competitor to AMD EPYC intended for data centers):
“SPEC has determined that the compiler used for this result performed a compilation that specifically improves the performance of the 523.xalancbmk_r / 623.xalancbmk_s tests by using a priori knowledge of the SPEC code and dataset to perform a transformation that has limited applicability.”
SPEC has therefore chosen to remove over 2,600 results because Intel allegedly broke the rules surrounding SPEC's "2017 Run" tests. SPEC's tests are important because they use the results to boast about the IPC (“Instructions Per Clock Cycle”) strength of the products. Intel's changes, according to SPEC, resulted in up to 9 percent and 4 percent better performance, depending on the test.
Recently, Intel was in trouble with an internal marketing presentation against AMD that was intercepted online, but later deleted:
"Intel's claim is that AMD uses misleading product names for its CPUs in order, according to the competitor, to trick customers into buying something that is slower than they think. According to Intel, this means, among other things, starting a SKU with the number 7, instead of 4. The former reveals the year of production, 4 reveals which generation it is in question. As an example, Intel writes that "Ryzen 5 7520U is built on dated Zen 2 architecture released in 2019!"