Google has revealed its cloud customers will get first crack at Intel’s next-generation “Skylake” Xeon processors. The company is targeting the new cloud instances for HPC workloads in healthcare, media and entertainment, financial services, and other industries.
The announcement was made in a blog posted by Urs Hölzle, senior vice president of Google Cloud Infrastructure. In his post, Hölzle explained the since the Skylake Xeon platform includes the Advanced Vector Extensions (AVX-512) that doubles the floating point performance from the previous generation, the new processor is “ideal for scientific modeling, genomic research, 3D rendering, data analytics and engineering simulations.”
Up until now, AVX-512 was only available in Intel’s Knights Landing Xeon Phi processors, which were designed specifically for HPC workloads. The inclusion of the instruction set in the new Skylake parts means that HPC applications not well-suited to the Xeon Phi’s manycore architecture now have a multicore alternative. With the Skylake Xeon, you get better single-threaded vector processing performance balanced against the disadvantage of less threads per processor.
No information was forthcoming with regard to the specific attributes (core count, clock speed, cache size, etc.) of the new Xeon parts. It’s even possible the processors have been customized for Google’s cloud platform. If you fill out the form the company has provided for customer inquiries, you may be privy to more detailed information.
The fact the Google will have the Skylake processors first is somewhat of a coup for the web giant. Google is in a fierce competition with Amazon Web Services (AWS) and Microsoft Azure for cloud customers, and currently trails both its rivals in market share. Earlier this week, the company announced it was adding NVIDIA GPUs to its cloud platform, making this the second successive roll-out of hardware aimed at users looking for lots of floating point performance. Whether this is part of a large strategy at Google to capture the high-performance end of the cloud market remains to be seen, but it’s an interesting development nonetheless.
The debut of Skylake Xeon on the Google cloud is also somewhat of a departure from a normal server processor roll-out, which is traditionally done in concert with major system vendors like HP Enterprise, Dell EMC, Supermicro and others. In the past, the earliest deployment of these server chips often ended up in supercomputers, which represented the most high-profile way a chipmaker could introduce a new platform. The fact that a hyperscale company like Google is able to get first dibs on a new processor suggests a shift in the way Intel, and perhaps other chipmakers, now perceive the server market.
Google has made the Skylake processors available in five of its most important regions, namely, the Western US, the Eastern US, the Central US, Western Europe and the Eastern Asia Pacific. If you want your Skylakes in-house, however, you’ll have the wait a few more months. Intel is expected to make the new Xeon processors generally available sometime in the second half of the year. It’s a fair bet they will be officially launched at ISC High Performance in late June.