Google cosies up to AMD for high-performance scale-out Tau VMs – but makes eyes at Intel and Arm, too

New 60 vCPU VMs come with some bold price-performance claims, but AMD needs to stay on its toes


Google's cloud arm has hooked up with AMD, tapping up its latest EPYC processors for a new family of virtual machines, Tau VMs, aimed at scale-out applications - but the company isn't keen on tying itself down to just one chip-slinger.

"The name 'Tau' also stands for the golden ratio," Sachin Gupta, VP and GM of Compute at Google Cloud, told The Register in an interview. "This is like the balance that you need in the design, with the latest CPU technology from our partner AMD, in order to achieve the best outcome.

"We've taken the latest generation of the CPU and put it into a server that's optimised for performance, rather than for some of the other things and other flexibilities like we have for large-scale databases. Working with key customers like Snap and Twitter, we talked to understand what are the things that we should shift in the server design, in the machine itself, to optimise for this kind of scale-out architecture," Gupta added.

"We have entered a high-performance computing megacycle led by the accelerated digital transformation of businesses and industries that is reshaping cloud computing," claimed AMD President and CEO Dr Lisa Su of the partnership.

"We work extremely closely with strategic partners like Google Cloud to ensure our AMD EPYC processors are ideally suited to meet the growing customer demand for more compute, more performance, and more scalability.

"The new Tau VM T2D Instances take advantage of the performance and efficiency leadership of the 3rd Gen AMD EPYC processors to provide Google Cloud customers with the latest compute capabilities to meet their scale-out workloads."

Part of the Google Compute Engine (GCE) portfolio, though also launching with day-one support for Google Kubernetes Engine (GKE), the Tau VMs are based on third-generation AMD EPYC processors with up to 60 vCPUs per virtual machine and up to 4GB of memory per vCPU. Compared to rival offerings from Amazon Web Services – both Intel-powered x86 and Arm-powered Graviton VMs – Google is claiming a 56 per cent performance hike translating to a 42 per cent boost in price-performance ratio.

"We looked at scale-out architectures, we compared with multiple offerings from the leading cloud providers," Gupta claimed. "The only thing that is fair and apples-to-apples is the list price for a standard on-demand VM that would be always persistent and on – and the same config on both sides in terms of vCPUs and memory.

"After that you have all the savings plans and offerings, like every cloud vendor has, different options there to get to a price with the customer. But that [price], you can't guess. And so we just looked at the overall list price for a corresponding VM."

While the Tau VMs won't be ready for general availability until later this year, Google has been trying them out with some pre-selected customers – and boasted of impressive results. "Both Snap and Twitter, in their initial testing, saw double-digit improvement," Gupta claimed, though Twitter Platform Lead Nick Tornow and Snap Senior Engineering Manager Cody Powell both clarified these initial tests had merely shown the potential for a possible double-digit performance gain in the companies' real-world workloads.

"We have become the high-performance partner of choice," Forrest Norrod, SVP and GM for the Datacentre and Embedded Solutions Group at AMD, claimed of the partnership, "instilling confidence from key partners like Google Cloud. We will continue to raise our own bar by consistently executing with competitive solutions available today, and strategically planning for leading products we will deliver in the future.

"For the Tau VMs, we worked with Google Cloud to optimise our third-gen AMD EPYC processor for the scale-out workloads that the Tau VMs will support. Because of the tight collaboration with Google, we were able to generate even better performance than initially expected. This means that Tau VM customers with scale-out workloads, can now do more, faster. For cloud-native companies that rely on scale-out, that equates to more users, more transactions, and better experiences."

Does that mean Tau VMs are synonymous with AMD, and Google is wedded to the company forever? Not so fast, said Gupta: "We're not saying 'hey, it's only going to be one processor type' either, because the Tau VM family is focused on scale-out applications," he admitted.

Could the future hold Tau VMs powered by parts from AMD's long-standing rival Intel, then? "Absolutely, yeah," said Gupta. "It could be Arm, as well." ®

Similar topics


Other stories you might like

Biting the hand that feeds IT © 1998–2021