GCP – C3D VMs with 4th Gen AMD EPYC Processors now GA, with 45 percent performance increase

Today, we are excited to announce general availability of the C3D machine series powered by 4th Generation AMD EPYC™ Processors (code-named Genoa) to Google Compute Engine (GCE) and Google Kubernetes Engine (GKE) customers.

The C3D machine series is a general-purpose VM that offers the enterprise-grade performance and reliability of AMD Genoa. Based on our testing performed in October 2023, and compared to our previous generation N2D, web-serving applications such as NGINX can see up to an 54% improvement in performance, relational databases such as Postgres or MySQL up to 62%, in-memory databases such as Redis up to 60%, development workloads up to 33%, and data analytics such as SparkSQL up to 21%.

C3D VMs are optimized to provide consistent performance and maintenance experience with minimal disruptions for general-purpose workloads that can’t tolerate performance variance. These are oftentimes workloads that are end-user facing, interactive, low-latency, or business-critical, including: web, app, and ad servers, databases and caches, streaming applications, or data analytics.

C3D scales up to 360 vCPUs and 2.8 TB of DDR5 memory across three memory configurations: highcpu (2GB/vCPU), standard (4GB/vCPU) and highmem (8GB/vCPU), with up to 12TB of Local SSD on the standard configuration.

Powered by Titanium

C3D VMs are built on Titanium, Google’s system of purpose-built custom silicon, security microcontrollers, and tiered scale-out offloads. The end result is better performance, lifecycle management, reliability, and security for your workloads. Titanium enables C3D to deliver up to 200 Gbps of fully encrypted networking, 3x faster packet-processing capabilities than prior generation VMs, up to 350,000 IOPS and 5 GB/s throughput with Hyperdisk Extreme (see below), near-bare-metal consistent performance, integrated maintenance updates for the majority of workloads, and advanced controls for the more sensitive workloads.

Hyperdisk storage

Hyperdisk is the latest generation of block storage in Google Cloud. Hyperdisk leverages Titanium to deliver significantly higher levels of performance, flexibility, and efficiency by decoupling storage processing from the virtual machine host. With Hyperdisk, you can dynamically scale storage performance and capacity independently to efficiently meet the storage I/O needs of data-intensive workloads such as data analytics and databases. Now, you don’t have to choose expensive, large compute instances just to get higher storage performance.

Certified for SAP NetWeaver

C3D VMs are certified by SAP to run SAP NetWeaver and SAP workloads on non-SAP HANA databases. SAP NetWeaver is the foundation layer for deploying, managing, and scaling SAP’s enterprise applications, including: SAP S/4HANA, SAP Business Suite (ERP, CRM, SCM, SRM), SAP Business Warehouse, SAP BW4/HANA, and SAP Solution. For more information on the SAP certification, visit: Certifications for SAP applications on Google Cloud and Google Cloud Certification SAP Note 2456432 (login required).

Accelerated performance for AI workloads

The 4th Generation AMD EPYC™ processors support AVX-512 with bfloat16, Vector Neural Network Instruction (VNNI) extensions and high-performance DDR5 memory that’s 50% faster than DDR4. These features allow for acceleration of AI inference on CPUs for common deep-learning use cases. You can take advantage of C3D’s features and further improve performance when using an inference runtime such as Neural Magic’s DeepSparse. Based on Google internal benchmarking, natural language models such as BERT and YOLO saw 2x higher throughput, and computer vision models such as ResNet saw 3x higher throughput on C3D with DeepSparse over N2D, as of October 2023.

What our customers are saying

“Operational and cost efficiencies are more important than ever as organizations move forward with AI adoption. As models continue to grow in size, the computational complexity and specialized hardware requirements can be overwhelming. Neural Magic’s work with Google Cloud and AMD helps companies deliver AI anywhere they want with readily available infrastructure they already know how to manage.” – Jay Marshall, VP of Business Development, Neural Magic

“After having significantly sped up our workflows and kept our costs low with last-generation AMD VMs, we eagerly anticipated the next-gen EPYC processors. With C3D, our job server workloads experience 25% higher peak performance over N2D and 20% over C2D. This sped up operations that cannot be further parallelized while offering much larger shapes for massively parallel tasks.” – Dimitrios Kechagias, Principal Developer, SpareRoom

“The Google Silicon team develops custom silicon solutions that provide differentiated user experiences in Google hardware. By leveraging Compute Engine C3D VMs, we saw a 40% reduction in simulation runtime compared to C2D. The faster runtimes can increase engineering and schedule efficiency as we build the next generation of devices.” – Rajat Bhargava, Senior Director, Silicon Engineering at Google

Get started with C3D today

C3D VMs are available today in the following regions: us-central1 (Iowa), us-east1 (S. Carolina), us-east4 (North Virginia), europe-west1 (Belgium), europe-west4 (Netherlands), and asia-southeast1 (Singapore). To start using C3D instances, select C3D under the General Purpose machine family when creating a new VM or GKE node pool in the Google Cloud console. Learn more at the C3D machine series page. Stay up-to-date on regional availability by visiting our regions and zones page or contact your Google Cloud sales representative for more information.