Google seems very keen to continue using Intel virtual machines • The Register


Google Cloud really, really wants businesses to keep using its Intel-powered virtual machines. This week, it unveiled a “white glove” service to convince customers that silicon is worth paying extra for.

It’s no secret that Google Cloud’s Intel-based instances have a premium over those using Ampere’s AMD Epyc or Altra processor cores. A fairly modest eight-threaded 32GB N2-Standard instance based on Intel’s Xeon Scalable Platform, for example, will run you about 25% faster than an Ampere Altra-based instance with twice the physical cores. . Meanwhile, AMD’s Epyc instances fall somewhere in the middle.

Despite the higher cost of Intel-based instances, Google says that depending on your workload, opting for a cheaper option may mean leaving performance on the table, and so you may get more bang for your buck from powered instances. by Intel. Google also spent a lot of money on these Xeons and damn it wants you to use them all.

Intel has long touted the artificial intelligence and cryptographic accelerators built into its Xeon Scalable processors. It really comes down to its AVX 512 vector extensions, Deep Learning Boost, and cryptographic engines. These can be used to accelerate a variety of workloads ranging from transcoding, database, and inference, without relying on GPU instances. But that’s only if the client’s software is optimized to take advantage of these features.

And that’s pretty important if Google and Intel are willing to work one-on-one with selected “high-growth enterprise” customers to refactor deployments and even modify their application code base in search of better performance.

At launch, Google is explicitly targeting customers running database, analytics, AI inference, Nginx, runtime language, and/or media transcoding workloads.

Under the three-phase program, engineers from Google and Intel will review a customer’s performance goals and identify opportunities for improvement. Based on this information, the web giant and titan x86 teams will take resource usage into account and provide a performance report detailing recommended changes to the customer’s environment, we are told.

Equifax, one of the first companies to join the program, said the optimizations had doubled the throughput and halved the latency of its deployments.

While the program is made available for free, customers must request the service. This suggests that Intel and Google could be blocking access to the program to avoid inadvertently sending customers to show up for the competition.

It’s worth noting that customers who rely on AVX 512 instructions to speed up their workloads won’t be stuck on Intel for long. As we saw with the launch of AMD’s Zen 4, Intel’s rival will finally roll out support for large vector extensions to its Epyc processors this fall. ®


Comments are closed.