Centriq Launch Poster
This week, Qualcomm
officially jumped into the server chip market with the launch of its Centriq 2400 SoC (System on a Chip) in a server market that has seen a steady stream of product launches from Advanced Micro Devices
. Though Qualcomm
announced Centriq some time ago, this launch represents a significant shift in the industry as we see the first performant Arm based server chip. For years, the question has been “can Arm find a place in the datacenter?” Based on what Qualcomm
delivered, I believe the answer is “yes.”
What is Centriq?
Centriq is Qualcomm
’s 48-core server processor comprised of 24 core pairs connected via a 250GB/s ring topology. This represents the highest number of cores per socket available in the server market. It was designed on the ARMv8 architecture to support scale-out datacenter architectures commonly seen in cloud environments.
Centriq is manufactured in Samsung foundries, using a 10nm FinFET process. What does this mean? Better performing transistors packed into a smaller space (18 Billion transistors in 398 mm2
). Moreover, what does this really mean to an IT consumer? Better performing cores that consume less power.
The Centriq lineup will have three shipping SKUs:Centriq lineup
Centriq will only support single socket server motherboard designs. Moreover, from the demonstrations shown at Qualcomm
’s launch event – it appears half-width “brick designs” are what the market can expect to see from hardware vendors. This would make sense as Qualcomm
focused its entire event on demonstrating cloud affinity.
This is not your average Arm chip
Over the last five years, several Arm vendors have come and gone from the server market. Calxeda, Applied Micro, and even AMD
attempted a run at the ARM market with its A1100 launch. Even Samsung made a run at the server space before pulling the plug before the product development was even complete. Perhaps most ironic was the failed Arm play from Broadcom
with its “Vulcan” processor.
The reason why companies failed in the past came down to performance. Arm server chips just couldn’t match the performance of their x86 counterparts. Even in the low end of the market, Intel
was able to stave off Arm with the position of Xeon D.
Things are different this time around. Qualcomm
claimed pretty impressive performance numbers during its launch event. When running SPECint_2006
, the Centriq 2460 saw about 7% better performance than the Intel
Purley Platinum 8160. Granted, Qualcomm
is comparing its 48 cores against 24 Skylake cores, but it is a competitive socket.
However, cloud providers do not just look at raw performance. “Price/performance” is a key measurement for success in the cloud as this articulates amount of productivity per dollar. This is where Centriq has a real story, claiming up to a 4x advantage over Purley.
Centriq is tailored for cloud environments
If it has not yet been obvious, Qualcomm
is targeting cloud providers with Centriq. Not just Azure, Amazon.com
and the like, but also the tier two players who provide “as a Service” to consumers. This seems like a good fit for a few reasons:
Is there a Centric ecosystem? Is there an Arm server ecosystem?
- Density –The half-width form factor should make Centriq racks highly compute dense.
- Price/performance – It appears Qualcomm will be able to claim bold price/performance leadership. This is especially attractive to cloud providers who live and die by OpEx.
- Security – By implementing Arm TrustZone, Centriq delivers an immutable root of trust for servers. This prevents attacks at the lowest levels from penetrating cloud datacenters.
- Power – A 120W TDP is not impressive. Being able to drive down idle power consumption to 8W is. This is what Qualcomm has achieved with Centriq.
- Loyalty (or lack thereof) – Economics drive cloud providers. Not legacy. Not loyalty. If Qualcomm can really achieve a price performance advantage they claim – cloud providers will jump. Not only to lower their operational costs. However, to apply pricing pressures to their other CPU vendors.
One of the more impressive takeaways from the Centriq launch was the lineup of ecosystem partners demonstrating their wares on Centriq. NoSQL, Bioinformatics, video processing, AI, cloud infrastructure. There was a breadth of technology partners that spanned the entire ecosystem on display. Qualcomm
was smart to recognize the importance of this robust ecosystem and smarter still to develop those relationships.
Centriq software ecosystem
The key for Qualcomm
will be driving value from these partnerships in the Centriq ecosystem. What does this mean? Developing an understanding of where Centriq can win and working with partners to build the compelling evidence that can move a potential customer through the inertia of adopting an alternative architecture.
Will Centriq find success?
Centriq is not a match for every workload and every use case. For workloads that require high “per core” performance, look elsewhere. I do not expect to see an SAP
HANA server reference architecture based on Centriq.
However, that is not where Qualcomm
is aiming. Qualcomm
has its sights set on cloud scale-out architectures – with an emphasis on scale out
. They designed and built a CPU that should support those workloads that benefit from many, well-performing cores—highly threaded applications such as many big data and analytics distributions.
For Centriq to find success, there are a few things that have to happen:
- Continue to build the ecosystem. Qualcomm needs a rich hardware ecosystem. Having HPE on stage promoting Cloudline was a good start. Qualcomm must now go after the other major server vendors such as Dell EMC , Lenovo , Huawei , Sugon and those “white box” server vendors that have found success in the cloud space.
- Be self-aware. Qualcomm must know where Centriq shines. Moreover, where it does not compete. Build the partnerships, solutions, and expertise to establish a beachhead. Once that beachhead is secured– find the next.
- Publicize the wins. Qualcomm has to show the world that cloud customers have moved beyond the pilot and into production. If it can get that one customer that quantifies the savings achieved with Centriq (and allows Qualcomm to publish), others will fall in line. I would have expected more deployment commits at this stage of the game, but I will remain patient. The previous Arm “wins” from Applied Micro and Calxeda ended up being tire-kickers, and Qualcomm needs to move beyond that.
- Keep the foot on the gas. At the Open Compute Conference in March, Qualcomm dominated headlines due to its collaboration with Microsoft on Project Olympus. It has once again garnered headlines with the launch of Centriq. Qualcomm must maintain momentum through the development of proof points and campaigns that educate potential customers.
Qualcomm has all the ingredients for success: a compelling part, a strong (and growing) ecosystem, and the right relationships with key customers. It will be exciting to watch how this disruption plays out in the market, but it is mandatory for Centriq to move far beyond its Arm predecessors.