Today we have the launch of the 4th Generation Intel Xeon Scalable. We have been testing the new chips for months, and it is now time to go in-depth into what Sapphire Rapids is, and why it is important for the industry. Without a doubt, the Sapphire Rapids launch is an enormous step forward for Intel. At the same time, the competitive landscape has changed. The answer to “how good is this new processor?” has changed significantly over the last decade from a black-and-white answer to shades of “it depends.” With that introduction, let us get to it.
4th Gen Intel Xeon Scalable Sapphire Rapids: The Video
If you want to learn more, here is the video for this piece.
We have a lot more detail in this article, but if you want to put that one on as a podcast (you can even speed it up) for later, feel free to get an accessible overview. As always, we suggest opening this video in another window, tab, or app for a better viewing experience.
Intel Sapphire Rapids Xeon Context: Today’s Market
We wanted to take a moment and just recognize where we are in the market today. About two months ago, we had the AMD EPYC 9004 Genoa launch. While we have reviewed some Genoa platforms, OEMs have been slower to get us systems mostly because of availability. There is no doubt the AMD EPYC 9654 is the current king of per-socket x86 performance. AMD has 96 cores with decent clock speeds, while Intel is topping out this generation with 60 cores. With that said, the heart of the server CPU market is in the 16-64 core space.
Intel has a very interesting value proposition: acceleration. As you will see in this review, Xeon is fighting an asymmetric war with EPYC in this generation. Intel’s key bet is that by embedding a high degree of acceleration in its processors, the relative per-core performance becomes much higher than if it pursued higher clock speeds or x86 pipeline improvements alone.
To be clear, that is a risk Intel is taking in this generation on a number of fronts. First, investing transistors in on-chip acceleration, if that acceleration is not adopted, is an expensive burden to all customers. Second, in workloads where the acceleration is not used, those transistors are “dead weight”. Third, accelerators can offer new security surfaces that must be diligently vetted. Make no mistake, Intel is aware of these risks, and feels like the benefits outweigh the costs. As we will show, when these accelerators are used, the relative per-core performance pops. In a world of per-core licensed software, that is Intel’s secret weapon but also the biggest point of exposure.
In the launch materials Intel shows, and others discuss, you will hear the benefits of acceleration. That is Intel’s key benefit with this generation. At the same time, looking at three of the big four new accelerators, QAT, DLB, and IAA, less than 45% of SKUs actually have these turned on. DSA is the only accelerator on every SKU, but only 27% of the SKUs have the full DSA configuration, and most SKUs have only one-fourth as much acceleration capacity. Perhaps the strangest part of the launch is that Intel has been discussing acceleration in Sapphire Rapids for months, it is the company’s key competitive edge, yet more than half of the SKUs are either light on accelerators or do not have them enabled.
One of the most exciting parts of this launch is really the overall platform, and what a server is. With this generation, we get DDR5, PCIe Gen5, CXL, and 50% more cores per socket along with those accelerators. This is a massive jump in platform capability and performance beyond just the CPUs themselves, so we are going to get into that in this article as well.
In this article, we are going to get into as much as we possibly can, but there are market segments we will not be able to cover. For example, Intel has up to 60 cores per socket in this generation, AMD “Genoa” has up to 96 cores, but if someone asks which company has more cores per system, that is actually Intel. While our introduction will focus on single and dual-socket servers, Sapphire Rapids is a new 4-socket and 8-socket capable platform, and we were able to see an 8-socket platform about a month before launch. We also did not get to test the Intel Xeon Max series with onboard HBM2e memory. Still, performance per socket, performance per core with/ without acceleration, performance per node, and so forth are all valid ways to look at servers, but there is only so much we can cover in this review.
Usually, we would start our journey by looking at the core and platform. The product SKUs are so important to the discussion we must look at those first. Without that context, it is hard to understand the magnitude and impact of the systems. With that, let us get to the SKUs.