Better, faster, and more efficient chips are driving down cloud operating costs and pushing prices lower, according to research from IT infrastructure standards and advisory group, the Uptime Institute.
With each generation of processor family, cloud pricing has trended downward with one notable exception, Owen Rogers, research director for cloud computing at Uptime Institute, explained in a write-up this week.
The research tracked Amazon Web Services (AWS) pricing across six generations of AMD and Intel CPUs and three generations of Nvidia GPUs using data obtained from the cloud provider’s price list API. While Rogers acknowledged AWS’ Graviton series of Arm-compatible CPUs, they weren’t included in testing.
All tests were conducted on AWS’ US-East-1 region, however, Rogers notes his findings should be similar across all AWS regions.
Of the eight AWS instances Rogers tracked, the majority saw a steady decline in customer pricing with each subsequent CPU generation. Pricing for the AWS m-family of general purpose instances, for example, dropped 50 percent from the first generation to present.
Some instances — AWS’ storage optimized instances in particular — saw even more precipitous pricing drops, which he attributed to other factors including memory and storage.
It comes as no surprise that CPU performance in these instances tends to improve with each generation, Rogers noted, citing the various performance and efficiency advantages to architectural and process improvements.
For example, AMD’s third-gen Epyc Milan processor family and Intel’s Ice Lake family of Xeon Scalable processors claim a 19-20 percent performance advantage over previous-generation chips. Both families are now available in a variety of AWS instances, including a storage-optimized instance announced last week.
“Users can expect greater processing speed with newer generations compared with older versions while paying less. The efficiency gap is more substantial than simply pricing suggests,” he wrote, adding that it is plain to see in AWS’ pricing.
In other words, while intuitively you may think instances based on older processor tech should be less expensive, more modern, more power efficient instances are often priced lower to incentivize their adoption.
“However, how much of the cost savings AWS is passing on to its customers versus adding to its gross margin remains hidden from view,” he wrote.
Some of this can be attributed to customer buying habits, specifically those that favor cost over performance. “Because of this price pressure, cloud virtual instances are coming down in price,” he wrote.
The GPU pricing anomaly
The exception to this rule are GPU instances, which have actually become more expensive with each generation, Rogers found.
His research tracked AWS’ g-and p-series GPU-accelerated instances over three and four generations, respectively, and found that the rapid growth of total performance alongside the rise of demanding AI/ML workloads have allowed cloud providers — and Nvidia — to rise prices.
“Customers are willing to pay more for newer GPU instances if they deliver value in being able to solve complex problems quicker,” he wrote.
Some of this can be chalked up to the fact that, until recently, customers looking to deploy workloads on these instances have had to do so on dedicated GPUs, as opposed to renting smaller virtual processing units. And while Rogers notes that customers, in large part, prefer to run their workloads this way, that may be changing.
Over the past few years, Nvidia — which dominates the cloud GPU market — has, for one, introduced features that allow customers to split GPUs into multiple independent virtual processing units using a technology called Multi-instance GPU or MIG for short. Debuted alongside Nvidia’s Ampere architecture in early 2020, the technology enables customers to split each physical GPU into up to seven individually addressable instances.
And with the chipmaker’s Hopper architecture and H100 GPUs, announced at GTC this spring, MIG gained per-instance isolation, I/O virtualization, and multi-tenancy, which open the door to their use in confidential computing environments.
Migration migraines persist
Unfortunately for customers, taking advantage of these performance and cost savings isn’t without risk. In most cases, workloads aren’t automatically migrated to newer, cheaper infrastructure, Rogers noted. Cloud subscribers ought to test their applications on newer virtual machine types before diving into a mass migration.
“There may be unexpected issues of interoperability or downtime while the migration takes place,” Rogers wrote, adding: “Just as users plan server refreshes, they need to make virtual instance refreshes a part of their ongoing maintenance.”
By supporting older generations cloud providers allow customers to upgrade at their own pace, Rogers said. “The provider doesn’t want to appear to be forcing the user into migrating applications that might not be compatible with the new server platforms.” ®