Intel and Google Collaborate on Computing ASIC for Data Centers
A collaboration between the two tech giants has resulted in improved data center compute hardware.
As cloud services grow in popularity, computing developers are reimagining the data center. In light of the slowing of Moore’s Law, many are turning to heterogeneous computing and hardware acceleration as the answer.
Google is releasing its C3 virtual machine and Hyperdisk to improve its computing infrastructure. Image used courtesy of Google
To address the needs of the modern data center, Google announced the release of its new C3 machine series, which features a custom infrastructure processing unit (IPU) developed in collaboration with Intel. In this article, we’ll look at the new IPU and how it provides a custom computing infrastructure for Google’s C3 machines.
Google and Intel Create Custom Computing IPU
At the heart of Google’s new C3 machine series is Intel’s IPU E2000, a custom infrastructure computing ASIC that Google and Intel developed together.
One of the key requirements for the IPU was programmability; the chip needed to be flexible enough to be optimized for any given computing requirement. To achieve this high level of programmability, Intel drew inspiration for E2000’s architecture from the design philosophies and successes of its FPGA SmartNICs. The result is a highly programmable packet processing engine that will be used to offload packet processing tasks from the C3’s CPU at line rates up to 200 Gb/s.
Intel’s IPU E2000. Image used courtesy of Intel
Beyond programmability, Intel highlighted other key features of the E2000, including an NVM Express storage interface, an advanced crypto engine, and compression acceleration blocks. Other hardware features include up to 2 x 100 GbE connectivity, 16 Arm Neoverse N1 cores, and 48 GB of onboard DRAM.
About Google's C3 Machines
Google's unveiled the E2000 ASIC in the context of its new C3 machine series, a custom architecture designed to power the future of Google's cloud services.
The C3 machines are powered by the 4th Generation Intel Xeon Scalable processor and supported by the E2000. Along with these hardware advancements, the C3 machines will feature Hyperdisk, a newly-released block storage solution from Google said to yield 80% higher IOPS per vCPU than other hyperscalers. This translates to a 10x higher IOPS in C3 machines than C2 machines.
In addition, C3 virtual machines will include 200 Gbps low-latency networking powered by the E200 IPU alongside PSP-based line encryption. Google claims that certain customers have seen performance improvements of over 20% compared to C2 machines.
Supporting the Cloud
As the strain on the cloud continues to grow, new hardware and software offerings are necessary to ensure that the data center can keep up. With the new C3 machines from Google, the company hopes to provide their cloud customers with greater security and performance while ensuring scalability as their cloud demands grow.
According to Google, the SoC hardware architecture introduced in C3 VMs can enable better security, isolation, and performance. In the future, Google may expand this purpose-built architecture to develop a richer product portfolio—for instance, designing products that support native bare-metal instances.