- AMD is pushing to corner more of the data center server market, with an eye on enterprises
- It's pitching upgrades as a boon for efficiency and a way to make the most of pricey software licenses
- But its focus on efficiency has the potential to push prospective customers into the arms of cloud providers, which offer cheap ARM-based silicon
GPUs have been getting all the love lately – so much so that Dell’Oro Group noted in December that hyperscale cloud companies were on track to deploy 5 million accelerators in their data centers in 2024. If you ask AMD though, there’s something to be said for CPU upgrades as well.
We’ve noted before that updated servers have the potential to save power in an energy-constrained data center environment. But Madhu Rangarajan, AMD’s CVP for Epyc Products, pointed out that more modern chips can also help enterprises make the most of their software licenses as well.
Wait, what? That was our reaction, too. But here’s how it works.
Say you have 100 users and they have their jobs queued up on an old CPU. If you have a set number of licenses, it will take a certain amount of time for that CPU to get through all those jobs. But with a more modern chip doing the work faster, you can either serve those same 100 users twice as fast or serve double the number of users in the same amount of time with the same number of licenses.
Basically, “fewer or the same number of licenses go a much longer way,” Rangarajan explained.
And in a world that has largely moved from perpetual licenses to subscriptions, saving on licensing costs is a big deal.
Server market smackdown
Of course, AMD has a horse in this race. The company has long been chipping away at Intel’s market share and Rangarajan outright stated that AMD’s goal is to “be the CPU leader in the market.” As of Q3 2024, Mercury Research data (as reported by PCWorld) indicated AMD held 33.9% market share for server revenue.
Rangarajan said AMD has already locked down some major clients like Meta and What’sApp. Now, it’s about scaling out to enterprises large and small – including the “long tail” of those with just 50-100 servers. “We want to get a lot more enterprises running on AMD,” Rangarajan said.
But there’s a catch to AMD’s big push around efficiency.
As AvidThink’s Roy Chua explained, “If you push too hard on the efficiency argument, it could flip them from x86 architecture to an ARM architecture.” The former is used by Intel and AMD, while the latter is used by cloud hyperscalers like AWS and others like Apple, which makes its own silicon.
“Efficiency is a good argument against Intel” for on-premises workloads that can’t easily be moved to the cloud, Chua said. But for those that can, as well as for new developments, ARM-based silicon will likely end up being the most cost effective.
(For what it’s worth, during AWS Re:Invent last month, CEO Matt Garman said the company’s Graviton processor is used by almost every customer and is still “growing like crazy.” Today, he said, “there’s as much Graviton running in the AWS fleet as there was all compute in 2019.”)
Chua acknowledged that switching workloads from x86-based chips to ARM-based ones takes work – mostly in the form of recompiling code. But he noted that process has gotten easier in recent years and is relatively easy for standardized business software.
And if Apple was able to move a massive consumer ecosystem from x86 to ARM, then enterprises – which are playing in a much smaller ecosystem – can do the same.
Meanwhile, in on-premises server land, Jack Gold said AMD faces a few other hurdles besides making the efficiency argument. For one thing, he noted that while both Intel and AMD use x86 architecture for their server chips, “they’re not 100% compatible” thanks to the incorporation of proprietary software. So, that alone – the possibility that a transition could spark issues – could be cause for hesitation.
There’s one other differentiator that’s a factor in the equation: customer service. Those eyeing a switch may question how well they’ll be supported by AMD compared to Intel, which is a much larger company, Gold said.
The takeaway? AMD isn’t wrong that efficiency is important, but there’s plenty more to look at beyond just power and licensing costs. And some of those factors could work in Intel and cloud providers’ favor.