Shulou(Shulou.com)05/31 Report--

When Apple announced its M1 processor, the chip became the first design to adopt this hybrid arrangement of CPU cores, which are made up of larger, high-performance cores and smaller, lower-power cores. Of course, we already know that these big kernels perform well, but we haven't seen the small kernels pass the head-to-head benchmark. However, Eclectic Light Company's blog has now addressed this issue, so we are looking at how these smaller kernels perform in general tasks compared to their higher-power counterparts.

The Apple M1 chip consists of four high performance "Firestorm" cores and four low power "Icestorm" cores. The two IP work together, but each IP is designed to run certain types of code efficiently.

Smaller kernels are usually dedicated to basic operating system processing tasks and background code execution, giving the larger kernel the freedom to perform anything else-- especially high-performance work. However, when major tasks such as heavy floating-point computing arise, it is directed to the larger Firestorm kernel.

These tests place the small Icestorm kernel under the floating point product test suite. To ensure that only the IceStorm kernel executes the workload, the tester sets the quality of service (QoS) setting to 9, which triggers only these smaller kernels, not the larger Firestorm kernel, as you can see in the processor image below.

The equipment tested performance in four cases: "A tight loop of assembly language that uses mixed SIMD instructions on a 4-wide single-precision floating-point array; Apple SIMD (a relative of the accelerator library) is in two simd_float4 arrays, and a simple Swift; that uses nested loops using Swift; uses map and reduce's more" customary "Swift nested loops."

If we compare it with the reference large Firestorm kernel used as a 100% score benchmark, the percentage of time it takes for the Icestorm kernel to complete these tasks is as follows: 190% SIMD (acceleration) for 33% library functions, 280% for simple Swift,550% for "idiomatic" Swift processing. Note that the baseline is 100%, while a 200% result means that it takes twice as long to complete a particular task. The usual Swift code benchmark represents the worst-case scenario of the Icestorm core, which takes more than 5.5x longer than the Firestorm core.

However, this is not surprising, because the Icestorm kernel is mainly built for efficiency. This means that low power and low heat output while still providing good performance is a buzzword today. Because the performance of the benchmark FireStorm kernel is very good, it doesn't make much sense to be a few times slower considering the increased efficiency.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.