Today we launched the 4th generation Intel Xeon Scalable. We’ve been testing the new chips for months, and now it’s time to dig deep into what Sapphire Rapids is and why it’s important to the industry. Undoubtedly, the introduction of Sapphire Rapids is a big step forward for Intel. At the same time, the competitive environment has also changed. “How good is this new processor?” answer to the question. has changed significantly over the past decade from a black-and-white answer to shades of “it depends.” With that introduction, let’s get to it.
4th Gen Intel Xeon Scaling Sapphire Rapids: Video
If you want to learn more, watch the video for this piece.
We have more details in this article, but if you want to post it as a podcast later (you can even speed it up), feel free to get an accessible overview. As always, we suggest opening this video in another window, tab or application for a better viewing experience.
Intel Sapphire Rapids Xeon Context: Today’s Market
We wanted to take a moment to recognize where we are in the market today. About two months ago, we had the AMD EPYC 9004 Genoa presentation. While we’ve looked at some Genoa platforms, OEMs have been slower to acquire systems, mainly due to availability. Without a doubt, the AMD EPYC 9654 is the current king of per-socket x86 performance. AMD has 96 cores with decent clock speeds, while Intel leads this generation with 60 cores. However, 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 waging an asymmetric war with EPYC in this generation. Intel’s main bet is that by including overclocking in its processors, the relative performance of each core is higher than with higher clock speeds or x86 pipeline improvements.
To be clear, this is a risk Intel is taking on a number of fronts this generation. First, investing in on-chip transistor acceleration is an expensive burden for all customers if that acceleration is not adopted. Secondly, in workloads where acceleration is not used, those transistors are “dead weight”. Third, accelerators may offer new safety surfaces that should be scrutinized. Make no mistake, Intel is aware of these risks and feels the benefits outweigh the costs. As we’ll show, using these accelerators unlocks the relative performance of each core. In the world of per-core licensed software, this is Intel’s secret weapon, but also its biggest point of exposure.
You’ll hear the benefits of acceleration in the presentations discussed by Intel and others. This is Intel’s main advantage with this generation. Meanwhile, looking at three of the big four new accelerators, QAT, DLB and IAA, less than 45% of SKUs have them enabled. DSA is the only accelerator in every SKU, but only 27% of SKUs have a full DSA configuration, and most SKUs only have about a quarter of the accelerator capability. Perhaps the strangest part of the launch is that Intel has been touting acceleration on Sapphire Rapids for months, the company’s main competitive advantage, but more than half of the SKUs are either light on accelerators or don’t enable them.
One of the most interesting parts of this launch is really the common platform and what the server is. With this generation we get DDR5, PCIe Gen5, CXL and 50% more cores per slot along with these accelerators. This is a huge leap in platform capability and performance beyond just the CPUs themselves, so we’ll get into that in this article as well.
In this article, we will go into as much as we can, but there are market segments that we cannot cover. For example, in this generation Intel has up to 60 cores per socket, AMD “Genoa” has up to 96 cores, but if someone asks which company has more cores per system, it is actually Intel. While our presentation focused on single and dual-socket servers, Sapphire Rapids is a new 4-socket and 8-socket platform, and we were able to see an 8-socket platform about a month before launch. We were also unable to test the Intel Xeon Max series with HBM2e memory on board. Again, performance per socket, performance per core with/without acceleration, performance per node, etc. are valid ways to view servers, but there’s too much to cover in this review.
We usually started our journey by looking at the core and the platform. Product SKUs are so important to the discussion that we should look at them first. Without this context, it is difficult to understand the scale and impact of the systems. With that, on to the SKUs.