Hewlett Packard Enterprise (HPE) wants to make it easier for GreenLake customers to scale compute workloads through a new collaboration with Intel. HPE CEO Antonio Neri announced GreenLake Silicon on Demand, which enables users to add additional processor cores and memory with the press of a button, during his Discover 2021 keynote today.

“This first-of-its-kind feature, developed in partnership with Intel, gives you the ability to activate and pay with a click to add capacity at the processor core and persistent memory level. Nothing to order or install, just turn it on,” he said.

The service, which is available now, is actually quite simple. Rather than shipping GreenLake appliances with lower core-count processors that have been right-sized for a particular workload, HPE instead uses a high-core-count chip from Intel and disables cores and memory channels unless the customer licenses additional compute resources.

Neri noted that HPE’s implementation, which is handled at the the silicon level, is different than allocating additional cores or memory via a hypervisor. This has distinct advantages for customers, according to Keith White, GM of GreenLake cloud services. By physically turning off cores, customers only pay for what they’re using — especially when running software with core-based licensing models, he explained. “With Silicon on Demand, we only turn cores on as they’re required, and thus the customer only has to pay for that core-based license when that’s turned on,” he said. “So, we see dramatic savings for our customers.”

It should be noted that this kind of functionality is built into most PCs and servers and commonly used for troubleshooting. HPE’s trick, however, ties this functionality into its GreenLake management plane.

You can think of this kind of like selling a car with a V8 engine, but only four cylinders actually work unless you pay an additional subscription fee.

“If the car has a V8 but only four cylinders are used, it’s not an optimal design,” said Baron Fung, research director at Dell’Oro Group, about HPE's Silicon on Demand announcement.  “The car with the V8 will be heavier and use slightly more fuel than one with just four cylinders. This car will be heavier, so it will accelerate slower and have poorer balance.”

And this logic applies to HPE’s implementation as higher core-count chips tend to operate at lower clock speeds than chips with fewer cores. Because of this, a 32-core Xeon with half its cores disabled could perform significantly worse than a server with a 16-core Xeon, despite the identical number of active cores.

If this sounds familiar, it’s not the first time Intel has attempted something like this. In 2010, the chipmaker launched the Intel Upgrade Service, which enabled consumers of consumer hardware — mostly PCs with low-end processors — to unlock additional cache and enable simultaneous multithreading for an additional fee. The short-lived program was met with stiff criticism and was canceled after a year in 2011.

So, it’s no surprise that HPE’s reimagining of this idea for the server space has been met with some skepticism from analysts.

“This may help lower the barrier to entry for modernized infrastructure as the upfront and initial cost to run is fairly low,” said Zeus Kerravala, principal analyst at ZK Research. However, the success of the program will depend on HPE’s ability to ensure the utilization of the product increases quickly to get customers to add more cores, he added.

“If it can do this, it can be a viable go to market for HPE,” Kerravala said. “If not, it's going to put margin pressure on a company that already has paper-thin margins. If this isn’t executed on correctly, it could lead to disaster, and HPEs track record on execution has been spotty at best.”

Fung adds that while scalability on demand is nothing new, especially when it pertains to storage, applying this logic to CPU cores and memory won't necessarily scale linearly in the same way.

Despite these challenges, Fung notes that standardizing on a single CPU and memory loadout should help HPE streamline its product stack and reduce the number of server SKUs at the expense of higher overall component costs.

Given the current climate, including the ongoing semiconductor shortage, stocking fewer SKUs should help alleviate supply chain challenges, he added.