Intel on ARM servers: Wimpy cores niche at best
An Intel executive weighed in on the recent chatter about ARM and Atom low-powered processors taking over the server game. The upshot: Low powered processors have a role, but aren't going to make much of a dent in the server market.
Speaking at a Morgan Stanley technology and telecommunications conference, Kirk Skaugen, general manager of Intel's data center unit, talked about ARM in servers as well as Atom. Skaugen also talked about the prospects for Atom-powered servers via the likes of SeaMicro, a startup that has been garnering a lot of press of late.
Here's what Skaugen had to say to counter ARM Holdings CEO, who recently had a much more optimistic take about his company's intellectual property prospects in servers.
We've been out talking about Atom and servers for several years to our customers under NDA. And candidly, there hasn't been a lot of interest in that architecture in a broad sense. I could see if you go out four to five years maybe 10% of the total market, give or take a couple percent, could be interested in such an architecture...if Atom is the best micro architecture we'll embrace it. And so we're constantly in the labs going out and asking about these workloads.
From there, Skaugen detailed a Google white paper that concluded that in most cases brawny chips (Xeon) outperform wimpy ones (Atom, ARM).
That said, Skaugen acknowledged there were some cases where Atom servers work in specialized environments.
What SeaMicro has done is they've put 512 Atoms into a 10U form factor. So if everybody in the world took a Xeon and bought an Atom because their servers were underutilized that would be a bad thing for Intel and our OEMs. That's not what we hear from the customers on what they're interested in doing. What they're interested in doing is getting, for example, for dedicated hosting -- let's say they have $140,000 to spend, they're wanting to know how many hosting nodes or how many customers they can host. So what SeaMicro has done is they've said, hey, I'm going to sell a $148,000 Atom server, they put 512 Atoms into a 10U and they say you can buy either I think 89 one-socket Xeons for the same price, 1U pizza box machines, so you can buy 89 1U's or you can buy the single system which has 512 nodes in it.
So what they've done is they've built a blade and that blade has eight Atoms on it and they put this into a 512 socket system. Well for Intel, our share of wallet goes something from 10% dedicated CPUs in that 89 Xeon example to 35% in the SeaMicro example -- 35% share of wallet. Why? Because in the 1U pizza box example you're buying 89 pieces of sheet metal and Intel doesn't make sheet metal, you're plugging in two power cables into the back of every one of those, we don't make power cables. You have fans on every one of those 1Us, we don't make fans. So instead they're just packing more compute into a much more dense form factor. So the reason I'm giving you that detailed explanation is Atom makes good margin for Intel; if the workload actually works that's incredibly good.
From there, Skaugen said ARM has some issues today.
Now what's the challenge that ARM has in that same form factor? Well, it has an instruction set issue. So if you're going to do hosting what application do you host? And what is an application porting effort -- we did application porting with Itanium, it took us about 10 years. Hundreds and hundreds of millions of dollars to port about 14,000 applications. ARM has to port for hosting all those applications over. Second challenge is the A9 and the A15 as we know it are 32-bit processors. Microsoft only supports 64-bit operating systems today. So I'd encourage you to go ask Microsoft what their position is on 32-bit operating systems. But I think they're pretty firm on their 64-bit. So it's an instruction set issue as well as a 64-bit issue. Everything we do in servers for real servers will be 64-bit.
Related: SeaMicro: Welcome to the world of $148K Atom servers