There’s been a lot of noise in the datacenter CPU market over the last few months. Intel launched its 2nd Generation Scalable Xeon Processor (codenamed Cascade Lake). Arm is winning in the cloud and seems to be finding momentum with its N1 & E1 processors. And let’s not forget AMD —in addition to finding momentum in the HPC space, the dribs and drabs about its upcoming EPYC launch (codenamed Rome) look very enticing. Though there’s been less noise around it, there’s also IBM POWER to consider. The CPU steadily makes its way in the market, with lots of upside. Given all this, what does the competitive market look like, who is poised to win in the end? Over the next few paragraphs, we’ll look at each company and try to make sense of this “summer of silicon.”
Intel – The reigning champ
Despite the hits Intel has taken around security and its delays in 10 nm production, the company still dominates in terms of server CPU market share at around 95%. At its Data-Centric Innovation Day in April, Intel shifted and made a subtle, yet significant tweak to its narrative. The story is simple and logical—we live in a data driven world and the needs of the modern datacenter span beyond compute. The ability to generate, process, move and store data in a secure fashion is critical for enterprise IT. To align with these needs, Intel aimed its narrative at processing everything, storing more, and moving faster.
The result is a portfolio of products that Intel believes will address these very specific needs, across compute, memory, storage, networking, and workload accelerators (FPGAs). Additionally, the Intel roadmap shows further development of accelerators that further enable this data-centric world with ASICs and a datacenter GPU portfolio.
Intel gets enterprise IT, and its strategy of tweaking the narrative to speak directly to these changing needs (with a solution focus) is smart. I believe it is well-positioned to compete today and tomorrow.
The big question for Intel is this - can the company ramp 10nm to production volume in short order? And can it demonstrate parity with AMD’s 7nm process across performance and power? Despite what Intel achieved with its Data-Centric launch, the CPU is still its crown jewels and the market will judge the company on its ability to get that performance and process crown back.
Intel certainly has the infrastructure, GTM channels, and enterprise inertia to weather the storm, so to speak. However, it must do so with an understanding that the competition is legitimate, and the market is open to change. Furthermore, that competition is no longer just another x86 processor company.
AMD – The #1 contender
If one takes a longer view of the chip market, AMD nailed the launch of EPYC (codenamed Naples) back in 2017. The product established AMD legitimacy in the server market and won quite a few cloud deals in the process. The product didn’t return AMD to double digit share in the server market, but it didn’t need to. As the mark of AMD’s return to the server market, it simply needed to demonstrate Naples as a viable alternative to Xeon, and it did. It is fair to say that the market was enthusiastically cautious with Naples.
Sometime in the next few months, AMD will launch the next generation of EPYC (codenamed Rome), and this is where AMD can turn cautious enthusiasm into real share gains. I believe it will achieve this goal. Built on the Zen 2 microarchitecture, Rome promises to be a beast of a server CPU with up to 64 cores (128 threads) feeding up to 4TB across 8 memory channels. Throw in 128 PCIe Gen 4 lanes in a single socket configuration, and you have a server that can do it all.
Recent announcements indicate that Rome is not just a leader when it comes to specs. AMD was just awarded a next-generation supercomputing project, codenamed Titan, by Oak Ridge National Lab (read about this win by my colleague Karl Freund—it is quite impressive). AMD also won in the academic world with deals in the United States and Norway. While one can discount performance numbers on a public demonstration, the fact that those who understand CPU technology better than most are choosing Rome indicates this part is going to rock for workloads that benefit from high core counts and large memory capacity
Why does this matter? These wins (combined with AMD’s success in the cloud) should have a cascading effect on enterprise IT. AMD now looks like not just a viable alternative, but the CPU of choice for many IT admins who are looking to get the best bang for their buck.
AMD is kind of the inverse of Intel. From a product perspective, AMD has the advantage, and in terms of market momentum, the company has it in spades. The question I have is this: can AMD expand its reach from cloud and HPC wins to enterprise datacenters, and all the way down to Main Street?
Intel is not sitting back idly. AMD has to build a volume and channel business that can contribute to the company if and when it loses the performance and/or process crown. In my opinion, success for AMD lies in building that vibrant volume business while it continues to chalk up wins in cloud, HPC, and large enterprise. Under the leadership of Dr. Lisa Su, I believe AMD will do so.
Arm – the (not so) Dark Horse
With so much hype and focus on the battle taking place in the x86 world, Arm has been (not so quietly) establishing itself in the server market. Let’s not forget that Arm already has a healthy datacenter footprint, powering many networking devices and appliances.
The hype around Arm-based servers started a few years back when companies like Calxeda, Applied Micro, AMD, Cavium and others made a run at the market (and let’s not forget Qualcomm’s big push with Centriq back in 2017). Arm gained serious player status with the launch of its Neoverse portfolio in late 2018 as it signaled the company’s commitment to developing a datacenter focused portfolio. Quickly following the Neoverse launch was Arm’s launch of its compute products focused on datacenter (N1) and networking/edge (E1). These products represent Arm’s first “ground up” server first designs, and I believe the ecosystem is responding in kind.
The Neoverse product family has a strong value prop. Standard speeds and feeds are on par with its x86 competition. Performance numbers look compelling, as does security. What’s also compelling is the potential for organizations to build device-to-datacenter solutions on platforms that maintain architectural consistency. This should lead to performance, management, and security optimizations that give Neoverse further market distinction.
AWS’s launch of Graviton, Packet’s support for Arm, and growth in China should all help add gravity to Arm’s momentum in the cloud space. The big question is how quickly the ecosystem (from silicon to platforms to ISVs) will turn Neoverse into solutions that resonate across the device-to-edge-to-datacenter continuum.
It certainly feels like the company is taking the right approach, and its partners are responding.
What about POWER?
As the hype swells around the x86 market and Arm continues to capture wins with the likes of AWS, let’s not forget a very strong player in the server space: IBM POWER. It is a powerful architecture that is well positioned in HPC and AI/ML. When one looks at the specifications and performance numbers, POWER rocks, particularly in accelerated workloads. So, why hasn’t it taken more market share from x86? Part of it is a function of the market, and part of it could be tied to the IBM brand.
First, let’s address the market challenges. POWER faces the same challenge as Arm in that x86 has dominated the server market for decades. During this time, the IT solutions landscape has been populated with vendors (hardware and software) that have an x86 first (and sometimes only) mentality. Even in the age of open source, the IT consumer is resistant to change and still looks at the world through the lens of Intel and AMD. I believe this viewpoint is shifting, but it takes time.
IBM is also a victim of its own success to some degree. Mention IBM to many enterprise IT admins and the first word that comes to mind is mainframe. What better example of an innovation that has stood the test of time? Unfortunately, it also represents a technology that is unfamiliar to IT consumers who are architecting next gen, cloud-native applications.
I believe IBM has an opportunity to leverage its acquisition of Red Hat to expand the reach of POWER in the datacenter, but it’s a delicate balancing act that, if overplayed, can hurt the Red Hat brand. My money is on IBM finding the right balance to advance the POWER brand (and market share) while Red Hat continues to thrive in the enterprise.
What’s it all mean?
Who’s going to win in the end? The answer is everybody. IT will be the biggest beneficiary of this highly competitive market. Choice will drive innovation. Innovation will drive products that enable IT solutions providers, from OEMs to ISVs, to better meet the needs of their customers.
Each chip company should find success as well. For some, like AMD, the success will be obvious—it comes by way of market share gains. Even companies like Intel, though, should come out on top. The renewed competition from AMD, the Arm ecosystem, and IBM forces Intel to crank its innovation engine into overdrive. This should lead to better technology and newly discovered opportunities that can only be found through a very strong competitive landscape.
It’s going to be a fun few quarters as this chip war continues to heat up. Check back for periodic updates.