Introducing Compute Optimized VMs Powered by AMD EPYC Processors
Over the last six months, Google launched 3rd Gen AMD EPYC™ CPUs (formerly code-named “Milan”) across our Compute Engine virtual machine (VM) families. It introduced the Tau VM family, targeting scale-out workloads. Tau VMs are the leader both in terms of performance and workload total cost of ownership (TCO) from any leading provider available today. We also refreshed our general-purpose N2D instances with 3rd Gen AMD EPYC processors, providing a 30% boost in price-performance.
Today, Google are excited to announce the General Availability of the newest instance series in our Compute Optimized family, C2D, also powered by 3rd Gen AMD EPYC processors.
“AMD EPYC processors continue to showcase their capabilities for HPC and compute-focused workloads. Whether that’s running drug simulations for the latest vaccines, exploring the cosmos, or helping design critical hardware and electronics for the future of the industry,” said Lynn Comp, corporate vice president, Cloud Business, AMD. “The Google Cloud C2D instances with AMD EPYC processors show the continued growth of the AMD and Google Cloud collaboration, by now offering some of the highest performance instances for demanding, performance-intensive workloads.”
New larger machine shapes for the Compute Optimized Family
C2D instances take advantage of advances in processor architecture from the latest generation AMD EPYC™ CPUs including “Zen 3” core. C2D supports Persistent Disks, Advanced Networking, Compact Placement Policies, and soon-to-follow Sole Tenant nodes. Instances are configurable with up to 112 vCPUs (56 cores), 896 GB of memory, and 3 TB of Local SSD. C2D is available in standard, high-cpu and high-mem, each with seven machine types for optimal memory-to-core ratio, to better align with your workload.
Improved performance for a wide variety of workloads
The Compute Optimized VM family is ideal for customers with performance-intensive workloads. C2D instances provide the largest VM sizes within the Compute Optimized VM family and are best-suited for memory-bound workloads such as high-performance databases, gaming, and high-performance computing (HPC) workloads, such as electronic design automation (EDA) and computational fluid dynamics (CFD). C2D high-cpu and standard instances serve existing compute-intensive workloads, including high-performance web servers, media transcoding, and AAA Gaming. C2D high-mem machine configurations are well suited for workloads such as HPC and EDA that require higher memory configurations. For optimal HPC workload performance, check out Google’s best practices for running tightly-coupled HPC applications on Compute Engine.
We’ve illustrated below how C2D with 3rd Gen EPYC compares against N2D with 2nd Gen EPYC (formerly code-named “Rome”) in GCP’s preferred set of benchmarks to measure compute intensive performance, media transcoding, and gaming benchmarks.
We worked with AMD engineers to benchmark some key applications in the HPC industry. The improvements in the Compute Optimized family are clear when C2D is compared directly to AMD’s previous generation of EPYC processors, specifically the n2d-standard-128 machine shape, closest to C2D’s 112 vCPUs. We first compare performance on industry-standard measures of memory bandwidth (STREAM Triad) and floating-point performance (HPL).
Compared to the N2D VM’s baseline performance, the C2D’s 3rd Gen EPYC processor improvements, including higher L3 cache sizes per core and full NUMA exposure, have a direct benefit in memory performance. This is empirically observed through the 30% improved STREAM Triad results. C2D’s floating-point improvements can also be seen in the 7% performance increase in the HPL results, despite being run with 12.5% fewer cores than the previous-generation EPYC processor.
Looking at application benchmarks across some key areas of focus in HPC, we can see that C2D VMs provide material gains for representative benchmarks in areas such as weather forecasting (WRF CONUS 2.5km), molecular dynamics (NAMD), and CFD (OpenFOAM).
What customers are saying
Not only is the c2d-standard-112 machine shape faster overall in the above workloads, but it’s also ~6% cheaper than the baseline n2d-standard-128 machine shape. It’s no wonder that customers are choosing it for their memory-intensive and HPC workloads. Here’s a sampling.
AirShaper is an cloud-based CFD platform that helps designers and engineers to easily run aerodynamic simulations to improve the performance and efficiency of cars, drones, motorbikes — even athletes themselves.
“Getting the best performance helps us drastically reduce run times, improving user experience and cutting costs at the same time. By running our CFD simulations on C2D, we’ve been able to reduce our costs by almost 50% and reduce simulation times by 30% compared to previous generation high-performance computing instances. Also, compared to our on-prem instances we’ve been able to reduce our simulation times by more than a factor of three.” – Wouter Remmerie, CEO Airshaper
Clutch’s Integrated Customer Data and Marketing platform delivers customer intelligence and personalized engagements for brands to identify, understand and motivate each segment of their customer base. Clutch offers solutions for CDP, Loyalty, Offer Management, Marketing Orchestration and Stored Value that use embedded machine learning to increase the lifetime value of each customer.
“We moved our compute and memory intensive Data Analytics platform to Compute Optimized on AMD EYPC Milan instances. The C2D instances provide a sweet spot of memory and CPU performance.” – Ed Dunkelberger, SVP Technology
Google Kubernetes Engine support
Google Kubernetes Engine (GKE) is the leading platform for organizations looking for advanced container orchestration, delivering the highest levels of reliability, security, and scalability. GKE supports C2D VMs, helping you get the most out of your containerized workloads. You can add C2D 3rd Gen EPYC CPU-based VMs to your GKE clusters by choosing the C2D machine type in your GKE node pools.
Confidential Computing (coming soon)
Confidential Computing is an industry-wide effort to protect data in-use including encryption of data in-memory — while it’s being processed. With Confidential Computing, you can run your most sensitive applications and services on C2D VMs.
We’re committed to delivering a portfolio of Confidential Computing VM instances and services such as GKE and Dataproc using the AMD Secure Encrypted Virtualization (SEV) security feature. We’ll support SEV using this latest generation of AMD EPYC™ processors in the near term and plan to add more security capabilities in the future.
Get started with C2D today
C2D instances are available today in regions around the globe: us-central1 (Iowa), asia-southeast1 (Singapore), us-east1 (South Carolina), us-east4 (North Virginia), asia-east1 (Taiwan), and europe-west4 (Netherlands), and in additional regions in the coming months. C2D instances are available via on-demand, as Spot VMs, and via reservations. You can also take advantage of further cost savings by purchasing Committed Use Discounts (CUDs) in one- and three-year terms. To start using C2D instances, simply choose the C2D option when creating a new VM or GKE node in the Google Cloud Console.
Need help to get started?