Intel launched its 4th Gen Xeon processor, code-named Sapphire Rapids, today. This launch comes after some delay due to manufacturing issues, a change of leadership and other internal organization shifts. But IT consumers don’t much care about any of this—they care about quality, performance, fit for workloads, reliability and security. In that spirit, I will use this piece to share my thoughts on Sapphire Rapids and how it impacts Intel's position in the competitive datacenter market.
Please note that while the full product name is 4th Gen Intel Xeon Scalable processor, for convenience I will refer to it simply as Sapphire Rapids.
Before we start, a disclaimer about specs
If you are reading this review hoping to see a "head-to-head" versus AMD's EPYC or some negative positioning of Sapphire Rapids because of a delta in high-level specifications that have little relevance on the vast majority of workloads running in the datacenter—you can stop reading now.
Any IT professional worth their salary understands that fitness for the datacenter goes beyond gaudy specs. It's about a CPU architecture that can deliver the best value for the workloads that power the business. That value is a balance of optimized performance, economics, reliability and security.
Each CPU vendor brings a unique value proposition to the equation. And there are no incorrect choices. Let’s finally have conversations based on the real world and not billboard-type promotions?
Introducing Sapphire Rapids
A small gathering of analysts and press got to spend time with Intel's executives prior to this product launch, and from the very outset Intel preached the theme of "workload-first" design. What types of workloads are being deployed to power the modern business (or the business trying to modernize), and what are the specific needs of those workloads—computational, memory, I/O, network, security, etc.? Based on this analysis came Sapphire Rapids – a platform that addresses all of these distinct performance requirements in one CPU.
The Sapphire Rapids family is categorized into performance categories, targeting specific workloads. Below is a chart that helps explain how the Sapphire Rapids SKU stack is positioned for datacenter workloads:
This will look familiar if you’ve seen Intel's previous Xeon product lineups. Intel categorizes Xeon into groupings that make it easier for an IT person to identify where to begin the search when considering fit for a specific function. General virtualization? Begin your search in the "Xeon Gold" family and perhaps move up to Platinum if performance or consolidation ratios warrant. Likewise, HPC, AI? The new Xeon Max family is what you would consider.
As an ex-IT person, I like how Intel groups the Xeon SKUs, making it easier to begin my consideration/buying journey. IT folks don't live in the CPU world, and remembering product numbers—or even trying to track 3000 Series up to 9400 Series—can be a little daunting.
If you're keeping count, Sapphire Rapids has 52 SKUs that cover the needs of the datacenter, spanning everything from IoT to edge to the cloud.
The new addition to the equation is the Xeon Max Series category (9400 Series). This aims squarely at high-performance computing (HPC) and AI—machine learning, deep learning, inferencing—with up to 64GB of high-bandwidth memory (HBMe2) to support driving the most resource-intensive workloads. The addition of the Max Series is a good example of how Intel has looked at the unique needs of the workloads running the modern business and tailored Xeon to drive maximum performance.
How Intel attempts to differentiate
After sitting through Intel’s analyst and media workshop and reviewing its materials, here is how I interpret Intel’s positioning with Sapphire Rapids. The workloads populating and powering the enterprise require more than just cores. And more than just lots of cores. These workloads require architectural tweaks and assistance that enable a business to hit the vital time-to-value metric in the shortest time possible. That time-to-value could be a life sciences company working on the next big breakthrough in cancer treatment, or it could be a financial services company engaged in algorithmic trading, or it could be a new tech startup trying to better understand the needs of the marketplace it intends to serve.
In the above scenario, more cores can help. But only so much. The data transformation journey requires computational resources, memory resources, I/O resources and networking resources. And depending on the workload, the knobs used to accelerate these resources must be turned and tweaked differently.
Sapphire Rapids ships with four acceleration engines to drive workload performance. These acceleration engines are in addition to the software extensions (and software extension improvements) Intel has made. The graphic below gives a good overview of how Intel approaches driving performance in Sapphire Rapids.
I like what Intel is doing to differentiate itself from its competition. There are several ways to look at why the company has positioned Sapphire Rapids the way it has. The first is obvious: the company doesn't have leadership in terms of what I refer to as the "billboard promotion"—high-level specifications that are impressive but don’t necessarily speak to real performance. And I’m certain this was a factor.
However, I believe the work Intel has done in workload acceleration is more significant than increasing core counts in many ways. Building acceleration capabilities through software and acceleration engines is a foundational capability that drives great value through improved performance, lower overall power consumption and lower TCO when considering support for workloads at the platform level.
Something else is important to consider in what Intel is doing, and that is the enablement of the software ecosystem. Architectural innovations in the CPU are useless if there is no upstream enablement. Frameworks, tools and applications can’t take advantage of these features if architects and developers don’t support them at the most fundamental levels. And this is an area where Intel has worked hard to engage successfully over the years.
In the end, it’s all about performance. And the question is, does all this focus and effort around acceleration translate into real-world value? Based on the numbers—like the ones in the graphic below—the answer is a resounding "yes."
Intel has also built out another area of differentiation for Sapphire Rapids through the CPU’s on-demand capabilities, named Intel On Demand. Through this service, users can enable CPU features as workload requirements change. Want to add an acceleration engine? But also need the budget to purchase a new server? Not a problem; just pay a fee and unlock these capabilities in Sapphire Rapids.
It is important to note that this capability is limited in support at the time of launch. And I can understand why—this is a big undertaking, and I believe the company will want to crawl/walk/run with this feature to ensure it lands smoothly in the marketplace. With this said, On Demand could evolve into something big in the industry across the value chain, from silicon providers to server vendors to IT consumers. Imagine being able to upgrade your server’s capabilities without having to buy a new server. Cost, sustainability, overall TCO, etc. – wow.
Lastly, Intel did a lot to shore up Xeon's security capabilities with Sapphire Rapids. This is an area where the company was competitively disadvantaged in previous generations. Applications and data are now protected through larger and improved SGX and VM isolation capabilities in the cloud to deliver a secure and robust environment with minimal impact on workload performance.
I'd like to see a couple of things added to Intel's security solution. First, notably absent from the list of cloud providers supporting VM isolation is AWS. I'm still trying to figure out why this would be, but I expect to see the AWS logo on Intel’s slides soon.
Separately, I don't believe Intel should limit VM isolation support to cloud providers. Enterprise IT organizations can also greatly benefit from this support. While I don't have inside knowledge about this (nor could I disclose it if I did), I suspect Intel is working with the likes of VMware, Nutanix, Red Hat and others in the IT ecosystem to enable such a capability. But we’ll see.
Addressing the elephant in the room
Let’s talk about the question everybody will be asking. How does Sapphire Rapids compare to Genoa? And does this do anything to slow down the momentum AMD has seen in the server market?
Let's address the comparison question first. At the time of writing, I have no performance comparisons relative to Genoa. In fairness to Intel, given Genoa’s launching in November and availability some weeks after, I don't think Intel had the opportunity to do any real benchmarking. I did see comparisons to Milan (the previous generation of AMD’s EPYC line), and those numbers were favorable. From this, I could make some guesstimates, but I don’t think that is a good practice.
The short version is that I believe Sapphire Rapids will perform well in those workloads where Intel has worked to deliver performance.
Regarding slowing the momentum of EPYC in the server market—this is the wrong way to look at things. AMD has done a great job with EPYC and has brought competition back to the datacenter. I believe Sapphire Rapids is a strong response from Intel that shows the company has sharpened its focus and has developed an architecture and platform that will deliver real-world value today—and tomorrow.
IT executives have asked me if the launch of Sapphire Rapids signals that Intel is back. My response has been, "Was Intel ever gone?" This company has the resources—human, IP, manufacturing and financial—to shape and reshape the industry repeatedly. And with Pat Gelsinger at the helm, and Lisa Spelman as the GM of Xeon, I think the company is in a good position.
The long-awaited 4th Gen Xeon Scalable processor has finally arrived. There are going to be some who focus on slipped schedules and timelines and use this as the lens through which they evaluate the product and the health of Intel.
Intel vs. AMD—it's almost a religious argument. While I don't believe Sapphire Rapids will convert the AMD faithful, I think it will give Intel customers a strong reason to stay with Xeon. As I previously mentioned, there is no wrong choice for IT consumers.
Now let’s move on to Emerald Rapids, coming to a datacenter near you in 2024.