In 2018, I distinctly remember attending the AMD EPYC Embedded 3000 Series Launch in London and sitting at dinner with Scott Ayler, who was the head of EPYC at the time, and we were asking when we would get EPYC Ryzen. Other executives at AMD are probably sick of me asking for a Ryzen-based EPYC for the past six years. Finally, I can stop asking since we already have the AMD EPYC 4004 series for entry level servers. With this launch, something we’ve discussed in several pieces over the years comes true, Intel needs to rethink its Xeon E series as it now pushes into the segment.
As you might expect, we’ve got a video about it, which you can find here:
As always, we suggest opening this in a separate tab, browser, or app for the best viewing experience.
AMD EPYC 4004 Series Overview
Let’s get to the new processors and here’s the review. AMD is using its AMD Ryzen 7000 series architecture, which scales up to 16 cores and with 3D V-cache models, to disrupt the entry-level server market.
We won’t go deep into the architecture because it is well known from the Ryzen side. We have Zen 4 cores and a TDP of 65-170W, ECC UDIMM support and even integrated graphics.
Indeed, these CPUs even look like someone etched something different on a Ryzen 7000 series heat spreader and denied it. And that’s not too far from what’s going on here, but as with the Xeon E series, the EPYC brand really adds value for validating a number of features and validating with server operating systems and software. As an example, AMD will lean on the fact that it has two 16-core SKUs and Windows Server 2022 is licensed in 16-core increments, making it now the de facto entry-level server SKU for those looking to use Windows Server.
When we say these parts are Ryzen-like, consider this. We put our CPUs into the ASRock Rack 2U1G-B650 2U AMD Ryzen GPU server we reviewed. For the lower-powered 6-core part, it was a straight swap.
For the 16-core SKU, we had to add a larger heatsink. We will have a follow-up with the larger version of the radiator, but it’s the same socket.
AMD previously had its AMD EPYC 9004 large socket series with Genoa, Genoa-X and Bergamo. Then there was the AMD EPYC 8004 “Siena” for up to 64 cores. Still, there was no real entry-level server product, so companies just made Ryzen-based servers. We’ve probably looked at about a dozen Ryzen server platforms over the years. At the same time, it’s always been a bit odd that the Ryzen server platform will run on hardware and software it hasn’t been validated for.
Here’s the SKU stack, ranging from the AMD EPYC 4124P at $149 to the 16-core SKUs at $699. Historically, the 65W Rzyen CPUs have been the absolute more efficient SKU, so the AMD EPYC 4464P is the one we’d focus on.
AMD uses the “P” to indicate that these are single-socket SKUs, as it does with the rest of the EPYC range. X stands for 3D V-cache. We wish there was an 8-core 3D V-cache SKU, as the AMD Ryzen 7800X3D is one of our favorite SKUs for Ryzen-based servers. This is an obvious gap in the SKU stack.
AMD pricing is also very strong. Its 65W TDP SKUs are priced between $36-41 per core, at least 20% lower than Intel’s offerings. Its higher TDP SKUs are under the $/core of even the best Intel Xeon E-2400 series parts. The 6-core Intel Xeon E-2486 is a part with a 30W higher TDP than the EPYC 4224P, but offers lower performance and costs twice as much.
Intel needs to massively adjust its pricing to stay competitive even on a core-to-core basis. The days of charging a premium for the Xeon E-2400 line are over and now it needs to adjust pricing to the reality of AMD’s entry into the market. With these tweaks, Intel could be more competitive, but what it really needs is an all-core Xeon E series that’s somewhere between the Intel Core i3-N305 and the Xeon 6 Sierra Forest.
Next, let’s move on to the performance of the new parts to see the performance difference at the socket level.
