Likely helping: Google (GOOGL) - Get Report announced on Tuesday morning that its public cloud services unit -- the Google Cloud Platform, or GCP -- has rolled out cloud computing instances that rely on AMD’s second-gen Epyc server CPUs (codenamed Rome).
The instances, which are currently in beta and go by the N2D label, are aimed at both general-purpose computing workloads such as business apps, databases and web servers, as well as high-performance computing (HPC) workloads in fields such as simulation, engineering and modeling. They follow the 2019 unveiling of Rome-powered instances for Microsoft Azure (MSFT) - Get Report and Amazon Web Services (AMZN) - Get Report.
Google’s announcement comes six months after it disclosed (at AMD’s Rome launch event) that it plans to offer Rome-powered instances via GCP, and also that it has begun using Rome for unspecified internal workloads. But while GCP was expected to eventually offer Rome instances, some of the performance claims it has made for the N2D instances have raised eyebrows.
Google claims that when an N2D instance was running the Coremark benchmark, which attempts to broadly test the performance of a processor core, it delivered 39% better performance than Google’s N1 instances, which are powered by Intel (INTC) - Get Report Xeon server CPUs. The N2D instance was also said to be 13% more cost-effective than comparable instances within Google’s N-series.
For HPC workloads -- a strong point for Rome to date -- Google claims N2D instances (aided by Rome’s high core counts and memory bandwidth) can outperform N1 instances by more than 100% on tests such as the Gromacs chemical simulation benchmark and the NAMD molecular dynamics simulation benchmark.
Also, Google’s N2D instances support up to 224 virtual CPUs (vCPUs), each of which represents a CPU core thread. Until now, N-series instances had topped out at 96 vCPUs.
For whatever reason, Google didn’t compare the performance of the N2D instances with that of its N2 instances, which are powered by Intel’s relatively new Cascade Lake Xeon CPUs. When launching N2 last August, Google claimed N2 instances “offer greater than 20% price-performance improvement for many workloads and support up to 25% more memory per vCPU” than N1 instances. That in turn suggests N2D instances could have somewhat less of a price/performance edge relative to N2 instances than relative to N1 instances.
Nonetheless, Google’s N2D performance claims do provide AMD with one more feather in its cap as it tries to wrest server CPU share from Intel. The company estimated it had a mid-single digit server CPU share as of the start of 2019, and has set a goal of having a double-digit share by mid-2020. In addition, CEO Lisa Su forecast on AMD’s Q4 earnings call that her company, which has guided for 28% to 30% total 2020 revenue growth, expects server CPUs to be its fastest-growing business this year.
With Intel’s Xeon line still quite dominant among traditional enterprises and many CIOs risk-averse when it comes to supporting new hardware platforms, cloud computing instances could prove an effective way for AMD to get its foot in the door at some Global 2000-type companies. Since cloud computing instances don’t require any hardware purchases and can be accessed on a pay-as-you-go basis, there’s little downside risk for companies to try out a Rome-powered instance, if they think it could deliver more bang for the buck for a particular workload.