Intel again accused of faking true CPU performance and it's not AMD doing the blaming

We have all probably heard of the proverb "fake it till you make it" as a way to encourage confidence, and it looks like Intel is doing the same thing with some of its benchmark optimizations. It is no secret that the company has been struggling to keep pace with the efficiency of AMD's Ryzen.

When we reviewed the latest Intel 14th Gen desktop parts, the i9-14900K and the i5-14600K, we remarked that the performance was certainly there but at the cost of very high power requirements.

The company has been looking at ways to convince potential customers why Intel is the better choice by throwing shade at AMD. Recently, a marketing PDF dubbed "Core Truths" leaked out where the company was found comparing AMD's CPU naming schema to "snake oil selling." And back in 2017, Intel ended up calling AMD's Infinity Fabric interconnect as cores and chip components being "glued together."

Intel has now been accused of inflating benchmark scores by SPEC (Standard Performance Evaluation Corporation) as the latter found specific optimizations were integrated into the Intel oneAPI DPC++ compiler, which are affecting the "523.xalancbmk_r" and "623.xalancbmk_s" benchmarks. These benchmarks evaluate Xalan and the XSLT (eXtensible Stylesheet Language Transformations) output.

On a SPEC CPU 2017 Integer Rate Result for Intel's Xeon Platinum 8480+ (Sapphire Rapids), the compiler data notes (spotted by Phoronix):

SPEC has ruled that the compiler used for this result was performing a compilation that specifically improves the performance of the 523.xalancbmk_r / 623.xalancbmk_s benchmarks using a priori knowledge of the SPEC code and dataset to perform a transformation that has narrow applicability.

As such, SPEC has decided to cancel this optimization per rule number 14 of the SPEC 2017 Run and Reporting Rules, thus removing over 2600+ results:

In order to encourage optimizations that have wide applicability (see rule 1.4 https://www.spec.org/cpu2017/Docs/runrules.html#rule_1.4), SPEC will no longer publish results using this optimization. This result is left in the SPEC results database for historical reference.

In case you are wondering, here is what Rule 14 says:

SPEC is aware of the importance of optimizations in producing the best performance. SPEC is also aware that it is sometimes hard to draw an exact line between legitimate optimizations that happen to benefit SPEC benchmarks, versus optimizations that exclusively target the SPEC benchmarks.

Companies like AMD and Intel often use SPEC results to tout the IPC (instructions per clock cycle) of their processors. Perhaps this was an attempt by Intel to show its Xeon CPUs were superior to AMD's EPYC. It is likely because AMD's share keeps growing in data centers, indicating Intel's previous "glued together" remark did not work.

This isn't the first time Intel has been accused of this kind of misdeed. Popular benchmarking software like Cinebench version 11.5, PCMark 2005, apparently had code supplied by Intel that had a "cripple AMD" function in them as in when a "GenuineIntel" CPUID was not detected, the benchmark would choose to run on a slower code path that would take longer to complete.

And there is also the case of BAPCo, where both AMD and Nvidia walked out accusing the company of favoring Intel in its SYSMark benchmark.