AMD Claims SYSmark Benchmark is Biased

In the technology industry, reviewers rely on accredited benchmarks to easily distinguish between the latest hardware and provide scores which help quantify the difference in performance. Generally, the majority of respected websites, and YouTube channels opt for the same software. For example, 3DMark is an essential tool to gauge GPU performance at various resolutions. Companies tend to fixate on benchmark numbers because if gives the consumer a direct comparison between their product and the nearest competitors. Today, AMD released a scathing video about SYSmark and claims the benchmark is biased, and inaccurate.

During the video, John Hampton, director of computer client products and Tony Salinas, engineering manager at AMD, said “SYSmark does not use everyday real world workloads”. In the software, an Intel i5 CPU attained a score of 987, while a “comparable AMD platform powered by an FX processor” managed 659. This is a delta of 50%. In comparison, PC Mark 8 accelerated scores are 4199 on the Intel platform and 3908 on AMD, a difference of 7%. According to AMD, there’s an excessive amount of stress being placed on the CPU and doesn’t accurately measure other key system components. PC Mark 8 provides a more well-rounded indication of real world performance.

To clarify this, AMD created a set of custom scripts and measured the start to end end time in the office suite and reached a time stamp. The Intel platform took 61 seconds to complete the task while the AMD system finished in 64 seconds. This is a delta of  6-7% which coincides with PC Mark 8’s accelerated results. Obviously, these are bold claims made by AMD and some might argue it’s just sour grapes for creating an inferior product. However, the data does suggest there’s a smaller gap than people might expect in real world scenarios. Despite this, SYSmark is recording the CPU’s performance so it’s bound to show a marked difference compared to a more general benchmark.

Do you feel AMD is right to criticize SYSmark’s benchmarking algorithm?