ARM Servers: Hype vs. Reality

b283e604-cfcf-11e3-a3f9-12313d1c3a13-mediumAs the ARM server market began to emerge in press and powerpoint, it was not hard to separate the hype from reality:  it was a lot of hype.  Spread by well-meaning advocates trying to change the world and give Intel a run for their money, these myths created unrealistic expectations on whether ARM chips are worthy of server applications, when they will ship, and how hard they will be to use. I applaud the early leaders including APM and AMD for their early efforts on 64-bit products.  While they have tried to balance their excitement and the uncertainty of semiconductor development schedules, there are nonetheless a few myths that need clearing up.  Here are six common ones:

1. ARM 64-bit V8 is available. Only on PowerPoint and very early samples for developers like Canonical, RedHat, Citrix, and Facebook, plus the hardware designers like HP.  ARM V8 chips are just now coming out from AMCC and AMD. Nobody else has even taped-out an A0 part to my knowledge. Production silicon is not yet available, and real server systems built on production silicon are at least 9 months away.

2. The first ARM V8 chip going to be a game changer.  Not from a volume standpoint. The first generation of anything is just that: the first step. Expect many, many revisions. Even some mis-steps.  I expect ARM SOCs to become a real threat to Intel on 2nd generation chips  built on 16nm FINFET (2016).  Not before.

3. ARM is low power. SOC’s are low power. ARM itself is not low power. No instruction set is inherently lower power, its the implementation that matters. Research conducted by IBM and others shows that an instruction set does not make a difference in power consumption. What matters are the SOC and system design. The SOC can aggregate several different system technologies onto one low power device. As Intel ramps up their own SOC capabilities, ARM does not necessarily have a sustainable power advantage. To be clear, ARM’s advantage lies in IP availability, partner innovation and differentiation, price, and lower development cost.

4. ARM has no software. False. For example, Calxeda moved its website to ARM servers in two hours, all spent just copying files and configuring the server. For the many websites just running LAMP stack applications, it’s that easy. Java is still pretty slow, but hopefully Oracle and/or RedHat can address that soon.  In fact, Java 8 was just released and is said to provide some significant improvements to Java performance on ARM.

5. Intel doesn’t get it. Wishful thinking by some. The most recent Atom chipset (Avoton, or Atom C2000) shows that Intel does get it. Although Centerton was an embarrassment, Intel released Avoton with aggressive pricing and very low power on 22 nanometers. Despite a roadmap designed for upsell, memory limitations, and questionable performance claims, Intel has a viable low-power SOC product today, vs. PowerPoint slides and claims for most ARM vendors.

6. ARM won’t make it. It will, but it won’t happen overnight.  64-bit ARM could be a very competitive part. The initial performance gap against Intel may slow early ARM adoption, but the performance difference will shrink and diminish over time as the 2nd generation chips are built and are fabricated on more modern process nodes. (2016)

In summary ARM servers are coming, but this market is going to take time to develop.  I suspect we will see a two-humped camel shape to the Gartner Hype Cycle when it comes to ARM servers.  More on that later….

 

Comments

  1. Why ARM servers ? Why not MIPS etc ?

    • Karl Freund says:

      Good question! MIPS could have stepped in, but they focussed on networking, so their volume and ecosystem are much smaller than the 6 Billion-with-a-B ARM units enabled by mobile and embedded. I think if anyone has a shot at challenging ARM, its IBM POWER.

  2. “1. ARM 64-bit V8 is available. Only on PowerPoint and very early samples for developers like Canonical, RedHat, Citrix, and Facebook, plus the hardware designers like HP. ARM V8 chips are just now coming out from AMCC and AMD. Nobody else has even taped-out an A0 part to my knowledge. Production silicon is not yet available, and real server systems built on production silicon are at least 9 months away.”

    There’s one that’s been shipping in large volumes for almost a year now – albeit not in servers. :-)

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Follow

Get every new post delivered to your Inbox.

Join 989 other followers

%d bloggers like this: