In reviewing CPU and server benchmarks, you have undoubtedly seen that testing covers each single-core and multi-core efficiency. This is the distinction.
By way of uncooked efficiency, each are equally vital, however single- and multi-core have areas of use the place they shine. So when selecting a CPU, it is vital to contemplate your specific workloads and consider whether or not single-core or multi-core greatest meets your wants.
There are nonetheless numerous functions on the market which are single-core restricted, equivalent to many databases (though some, like MySQL, are multicore).
Efficiency is measured in a few methods. Clock frequency is the large one; the upper the frequency the sooner apps will run. Additionally vital is the width of execution pipelines, and the broader the pipeline, the extra work can get carried out per clock cycle. So even when an app is single threaded, a wider pipeline can enhance its efficiency.
Multi-core benchmarking usually entails operating a number of apps in parallel relatively than bringing a number of cores to bear on a single software. Every app runs on a separate core with out having to attend its flip as it will with a single core.
Many chips concentrating on cloud suppliers and enormous enterprise have 96 (AMD Epyc “Genoa”) to 128 (Ampere AltraMax) cores. The extra customers and extra digital machines, the extra cores to deal with the load.
Per core pricing
These very giant chips are usually used to run multi-teneant workloads, together with containers and digital machines, mentioned Patrick Kennedy, president and editor of Serve The Dwelling, an impartial testing web site for SMB to enterprise server gear.
As a result of chips are licensed per core, enterprises ought to search the very best efficiency per core in an effort to reduce license charges, he mentioned. Loads of the demand for single-core efficiency is to get round these charges.
Cores Getting Some Assist
After years of AMD lagging behind Intel in each single- and multi-core efficiency, the 2 are actually equal in each benchmarks, Kennedy says. “I would say Intel and AMD are very a lot interchangeable in most functions. However I believe that there is in all probability that 10%-15% instances the place they’re simply vastly completely different,” he mentioned.
For instance, in any state of affairs the place reminiscence bandwidth was restricted, he would use AMD Epyc processors over an Intel Xeon as a result of Epycs have monumental caches, and going to cache is quicker than going to fundamental reminiscence.
“For a basic function, enterprise workload, I believe realistically, you might use both [Intel or AMD]. However I’d typically inform individuals, at this level, it is in all probability value attempting one in every of every, and making a choice based mostly in your workload,” mentioned Kennedy.
The efficiency of CPUs alone is not the deciding issue. Servers are more and more being augmented by accelerators like GPUs, FPGAs, and AI processors that offload duties from the CPU in an effort to pace up the system as a complete.
For instance, in something having to do with VPN termination, Kennedy mentioned he would “100%” use an Intel processor with a QuickAssist crypto/compression offload card as a result of it lifts a giant load off the CPU. However, if he was doing one thing that was reminiscence bandwidth restricted, he would use AMD Epyc chips as a result of Epycs have monumental caches.
Copyright © 2022 IDG Communications, Inc.