Arm recently announced the launch of its next-generation Neoverse family of IP, and its intentions are clear – to be the architecture of choice in the cloud-driven enterprise. So, what is next-generation Neoverse, and what is the direct impact on enterprise IT organizations? And how will/should the x86 players respond? I’ll try to cover this in the following few paragraphs.
First – what is Neoverse?
Arm’s Neoverse microarchitecture focuses on supporting the needs of the datacenter, This includes everything from high throughput switching to low power radio access networks, from general-purpose computing to high-performance computing (HPC) and artificial intelligence (AI). Think of Neoverse as taking the underlying ArmV9 architecture and tailoring it for the datacenter through modifications and specific instructions.
For high-performance needs (e.g., AI, HPC), V-series is the Arm microarchitecture. The V-series has the highest performing cores and microarchitecture to move and process more data faster.
If looking for more traditional infrastructure, N-series is Arm’s IP. The N-series can deliver high core counts that scale well to meet the needs of many of the workloads that one would usually consider more “traditional.” An excellent example of this would be AWS’ ECS M6g instances, powered by Graviton2.
In cases where throughput is critical, the Neoverse E-series IP is the best fit. The E-series also delivers the most significant power efficiency—efficiency from a company known for its power-efficient designs.
Finally, Arm’s Coherent Mesh Network (CMN) is an open, highspeed interconnect that enables Neoverse CPUs to access memory and I/O devices more quickly – and more openly.
So, what is Neoverse, and what does it mean to an IT organization? Neoverse is the building block to many of the solutions powering modern enterprise. This enables cloud providers, hyperscalers, and others to deploy solutions fast and securely and at a lower cost than legacy infrastructure solutions. And for IT organizations, this translates into cloud services that can perform many tasks better than the traditional instances, at a lower cost.
Second – what was announced
Arm announced the launch of its second-generation N-Series microarchitecture, N2. As previously mentioned, N2 is based on the recently announced Armv9 architecture and, as would be expected, delivers greater levels of security, performance, and power efficiency. When comparing against N1, N2 shows a 40% single-threaded performance improvement at the same power.
What does this mean for cloud providers? The ability to drive down costs while increasing levels of service to customers. And the beneficiary of this should be enterprise IT cloud consumers who will have lower-cost cloud options without sacrificing performance or consistency of performance.
With the introduction of V1, Arm signals a commitment to driving a microarchitecture suited squarely at the high-performance computing space. There are a couple of ways to look at V1 – performance and design flexibility.
From a performance perspective, V1 looks to be a beast, with a 1.8X uplift in vector workload performance, a 4X uplift in machine learning (ML) performance and a 2X increase in floating-point performance. Understanding that different HPC workloads have other computational profiles, V1 addresses each of those compute needs quite well.
From a design perspective, V1 enables the Arm silicon ecosystem to quickly tailor SoCs for target workloads, applications, and frankly, even customers. Don’t overlook the importance of this openness, as this, along with CMN-700, enables Arm partners to quickly bring tailored solutions to market to meet the needs of the next killer app in the datacenter.
Third – Arm’s journey is a little unconventional – and really smart
While the traditional enterprise datacenter is filled with racks and racks of (standard) x86 powered servers, the enterprise increasingly relies on the cloud and cloud-native principles. This is where Arm’s journey to the enterprise begins. Rather than supplant these traditional rack servers supporting a lot of legacy applications and workloads, Arm appears to be focusing on two segments: cloud/cloud-native and high performance.
Arm’s moves in the cloud have led to much success, with providers like AWS rolling out more and more Arm-based instances (via Graviton2) that go beyond general-purpose compute. When looking at the range of workloads AWS targets with Graviton2, it is hard to distinguish between what Arm can do relative to its x86 competition.
If you’re wondering how Graviton2 is working for AWS, consider this. Almost half of the new AWS instances introduced last year were on Graviton2, and those instances are available in 70 of 77 regions. It’s fair to say that AWS sees Graviton2 as a success.
Arm is also approaching the enterprise datacenter with its play in high-performance computing (though maybe more specifically in those workloads with high compute requirements). This is an area where Arm’s partners have shown an ability to deliver higher performance levels at a fraction of the TDP and a fraction of the cost. As V1 hits the market through companies like Nvidia (Grace), Fujitsu and SiPearl, expect to see Arm represented quite well on the supercomputing TOP500 list.
I like the approach Arm is taking to gain traction in the enterprise. The company and its partners are looking for pockets of opportunity where a joint solution can have success. These include new markets (edge), markets without a particular architectural requirement (cloud) and markets where customers are open to change for the sake of performance (HPC, AI). As success is found in each of these areas and software continues to evolve into architectural agnosticism, Arm and its partners should find opportunity in the more traditional markets, where x86 has a stronghold.
Fourth – Here’s why Arm is so interesting
Forget about performance. Forget about power. Forget about paths the enterprise. Put all that aside for a second. Here’s what I find most compelling about Neoverse and the Arm ecosystem. In a software-driven world, the typical development lifecycle for CPUs is in the three-year range (from concept to launch). That means silicon is typically lagging software support. As a result, the software market perpetually waits for the right infrastructure to exploit its capabilities to the fullest.
With Arm’s model, that changes. Arm partners can quickly develop and roll out infrastructure tailored for that next software trend. Silicon partners can utilize architectural licenses to build uniquely tuned platforms for that critical workload in a much faster timeline. The result is obvious–faster adoption of technologies that can lead to real change in the world.
Just a few short years ago, Arm’s server business looked like a dead end. Chip companies developing Arm chips were starting and failing at a rapid pace, and it felt like x86 had retained its stranglehold on the server market.
When Arm announced Neoverse in October of 2018, the company said all the right things and showed all of the right images, but I believe the market was still getting over the likes of Calxeda, AMD (yes, AMD), Applied Micro and Qualcomm. At best, the market seemed to take a “wait and see” approach.
The market failed to note that while these companies may have come and gone, a robust software ecosystem was built around Arm in servers. The open-source community fully embraced Arm and building for it accordingly became standard practice. What was needed was the “killer app” to drive mainstream adoption.
Three years later, that killer app has turned into killer apps–cloud, edge, and high-performance computing workloads. Arm is finding much success in each of these areas. As the enterprise relies more and more on both the cloud and workloads with unique compute requirements, Arm will continue to find success.
The modern enterprise datacenter is no longer four walls with racks and racks of servers and gear. It spans the edge to the cloud to the core. Arm will continue to grow its footprint in these markets. Make some room, x86 – you have company.