The Intel Xeon E series is something that we have been reviewing for over 10 generations at STH. A fun fact, I used to buy the Intel Xeon E3-1200 series at retail before the chips were “launched” over a decade ago when STH was much smaller. Still, in 2025, Intel has a really neat challenge on its hands compared to AMD in the lower-end single-socket market.
The Intel Xeon E Challenge in 2025
We are at a strange crossroads for the Intel Xeon E series. The Intel Xeon E-2400 series launched about 14 months ago. The “Raptor Lake” parts are aligned to the 14th Gen Intel Core processors. Since then, Intel has its new 100 and 200 series processors, yet Xeon E is still stuck in 2023. It only scales to 8 cores and 16 threads, albeit with maximum clocks above 5GHz.
High-speed cores are great for applications like databases and even EDA applicatons. I still remember seeing in 2016 the big racks of Intel Xeon E parts in an old Intel Santa Clara fab converted to a data center for EDA. These were in dense QCT systems that were very neat.
Let us put this into some perspective. The Intel Xeon E3-1230 launched in 2011 was a four core part and one of my favorite chips. At that time, the top-end Intel Xeon E5-2600 parts had a maximum of eight cores. Since then, the dual socket CPUs have gone to at least 128 cores (excluding the cloud-native segment.) While we have seen roughly 16x core count growth in the mainstream server segment, we have only seen 2x growth in the entry Xeon segment. Of course, both series have higher clock speeds, newer architectures, and newer faster memory. Still, the Xeon E-series has roughly a similar PCIe footprint while the mainstream servers have increased significantly.
Perhaps the obvious answer is that Intel could take its P-core/ E-core designs to the Xeon E series. One of the big challenges to that is introducing heterogeneous cores on the same chip. Software like VMware ESXi does not play as well with that type of architecture. While this can be fixed, and seems trivial at first, when SLAs are involved in enterprise environments, having heterogeneous cores adds another layer of scheduling challenges.
Intel’s main path to scaling, using heterogeneous cores, might not have been necessary five years ago, but it is a different world today. The AMD EPYC 4004 series launched in 2024 and that was the point when the entry server market was almost all Intel’s. Now there is a competitive part in the market.
Final Words
Part of this came about from the fact we are now working on our HPE ProLiant Microserver Gen11 review. This is an update to the HPE ProLiant MicroServer Gen10 Plus (and V2) that we looked at a few years ago and it was released just last year. I cannot help but wanting something a bit more from the CPU on the one hand. On the other hand, in this type of machine, if you just need storage and to run a few applications, the Xeon E-2400 series is still capable. This might have been a good segment for a Xeon D or even an Atom part years ago, but those have moved upmarket in terms of pricing.
It is a different market now than it was now that AMD has entered. This is not a segment where the dollars are huge from a market share perspective, but it is a segment where units shift. Maybe this is a long way of saying: Intel needs a new Xeon E. It feels like it is about time.
The unwritten assumptions in this article are that better cores will shift more chips and won’t undercut more expensive Xeons. I don’t think either are particularly true: customers in this segment are likely to be extremely conservative in their server choices (e.g. still more confident in Intel than AMD). Many of them will be SMBs running important line-of-business applications (usually existing ones: new ones would be more likely to be cloud-based). If they needed more CPU power they’d already be elsewhere; if they were larger they’d have consolidated to much larger rackmounted servers. Sun showed that you can last for decades with underpowered cores if the trust is there.
Maybe the bottom line is that Intel simply doesn’t have a better core to put into Xeon-E packages, on desktop most of the performance increases have come from increasing power consumption and cost balancing was done with combining P and E cores.
There likely isn’t silicon that would be cost/performance efficient for Intel in this market. For purposes where cost is the main driver there’s cheap desktop-like parts and for purposes where reliability and support are important Xeon-D / Atom parts are capturing most if not all available premium in that market. So I don’t really see a huge market opportunity for Intel that is currently not being captured.
I wonder if the limited growth of core counts on the E-series is a result of Microsoft’s licensing framework for Windows Server. 16 cores is both the minimum number of cores you can license and the maximum number permitted with a base Windows Sever license. Since E-series Xeons are typically used in entry level servers, the majority of end users may be price sensitive to spending more on Windows Server licensing for higher core counts.
Windows Server licensing is less of a concern for end users of more powerful Xeon CPUs, because those chips are in far more expensive servers already.
In a row of bad management at Intel, this is what they get for canceling Bartlett Lake.
It was there on the roadmaps as a 12C/24T all P-core chip based on raptor Cove with the proven intel7 manufacturing.
It would have been all Intel needed to counter Epyc 4004 and give Xeon E3 a new lifeline. Hell, even for some Consumer markets (halo gaming) this would have been better than the 285k.
I am still amazed how little ground AMD has made in this segment despite having technically no competition. Intels ecosystem still draws in a lot of sale wins and that keeps them afloat.
I’m with Jeremy. We’ve switched to AMD because we deploy so many Windows Server systems on customer sites. Even if Intel has higher clocks and per core performance, AMD at 16 cores lets us use an entire license.
I just wish AMD did RDIMM instead of big speeds on memory so we could get 256GB with 16 cores
I’m curious whether Xeon E is really a viable niche now that AMD is a serious threat in the datacenter and Intel can’t just name their price and expect you to suck it up.
For essentially any smallish server/appliance application the included NIC features in Xeon D/server Atom parts are genuinely attractive added value/reduced BoM; so there’s a real reason why you might do that vs. Xeon E or Epyc 4004 and bolting on all the NICs you need over PCIe; but Xeon E doesn’t have that; it’s pretty much just the desktop lineup with ECC support and a nontrivial delay.
If socketed vs. BGA is important I’d imagine that Intel can socket a Xeon D; but it just seems hard to justify the continued existence of the Xeon Es when they lack the integrated networking that makes Xeon D servers more attractive and trail the desktop parts that they mostly don’t outperform.
Im in the belief that these need cheap networking built in as well.
I also wanna say that I love when Patrick does these
The current Xeon E line up carries the same baggage as the desktop Raptor Lake parts with potential instability issues. These have since been addressed but are not a sign of confidence. Raptor Lake’s successor, Arrow Lake, has been a mixed bag on desktop with some things being faster while others are slower. The biggest thing for Arrow Lake has been a solid improvement on the E core side of things and the applications that can leverage them full alongside the P cores have shown the biggest gains. It isn’t clear how this would play out for entry level servers. There is an alternative on the horizon with Bartlett Lake supporting up to 12 P cores without any E cores. This would be an ideal Xeon E chip and side step the issues when mixing P and E cores. The downside for Bartlett Lake is that the core design is the same as Raptor Lake and clock speeds on the same level as Raptor Lake. Thus aggregate performance may not move at all for Bartlett Lake, just cleanse the platform from the Raptor Lake issues and heterogenous issues.
They’ll probably name the next one something silly like Xeon 6300
i’m loving the commentary STH
@Ivan K,
I was puzzled by your comment,
“I just wish AMD did RDIMM instead of big speeds on memory so we could get 256GB with 16 cores”.
I have 256GB (4x 64 DDR5 RDIMM) right now in my Threadripper workstation.
@Syed
The topic of discussion is Xeon E3 and the related EPYC 4004 series on the AMD side.
I.e. re-badges of the respective desktop platforms. Hence UDIMM only.
That said, on the higher end the 16C EPYCs are not much more expensive on the CPU side. It is the platforms which are at a different level. Both cost, idle power consumption and mostly the form factor.
If you want more/faster cores in the Intel UP segment, you can always use a workstation chipset with a 12th/13th/14th gen Core i-series processor. You get ECC and all the management features; you just don’t have the Xeon branding.