Intel Bumps Xeon Memory Speeds

Coming soon, faster DDR5 and MRDIMMs

Jul 20, 2026 by

Intel LogoIntel is doing the spec upgrade by opinion blog thing again, this time it is Xeon memory speeds. SemiAccurate likes the tech of the upgrade but the format is… meh.

We can’t link the original because it isn’t up when this is being written, but it is basically three bullet points done over 2+ pages. The bullet points are that current Xeon 6 and 6+ (no clue what those are actually) will soon support DDR5-8000. Then comes the tease for MRDIMM-8800 in the upcoming 6900P CPUs in Q1/27. Lastly we have the fun fact that Intel considers high core counts to be 72-128 cores. If you heard that too, it is AMD engineers giggling. Lets take a look at these things in a bit more detail.

The memory speed upgrade on the 6/6+ series from 6400 to 8000MT/s is a good thing, period. Sure it tends to necessitate high quality PCB material and lots of layers but that is the case for everyone, AMD and ARM included. The 25% speed bump is welcome, Intel’s wrapping it in AI BS is not, it is just a simple speed upgrade albeit a hard one to engineer. 1DPC is here to stay, why didn’t Intel and AMD jump on board IBM’s memory buffer train, it is a better way.

With MRDIMMs going from 8000 to 8800MT/s we also go from MRDIMM Gen 1 to Gen 2. There are a few things here, first off that given that MRDIMMs are effectively rated at ~2x the clock speed of the chips, the current 8000MT/s devices run at much lower clocks. The net result is about a 35-40% improvement in actual transfers. Moving from 8000 to 8800MT/s is a lower percentage bump than DDR5 going from 6400 to 8000MT/s. That said it is still welcome.

More troubling is the claim that the 8800MT/s parts are ‘Gen 2’. Why? Because there is no info on what Gen 2 is. Did they bump the clocks? Is it a spec change? Compatible? Will you need new CPUs? New boards? Who knows? This is why you don’t do spec announcements via fluff blog posts, people like me ask questions and mock you. See the next paragraph for more mocking.

Here is the more mocking. Intel calls Granite Rapids CPUs in the 72-128c range ‘high core counts’ which they seem to be using separately from HCC devices from what we can tell. Intel has unquestionably better memory support but is a bit light on memory channels. AMD is currently at 192 2-thread cores per socket, the small Intel socket caps out at 86c with 8-channel memory while be big socket has up to 128c with 12-channel memory. AMD has one socket with 12-channel memory and one set of boards and related infrastructure. 128c vs 192c isn’t a fight and when Venice debuts really soon we will have 256c with 16-channel memory too. Sometimes you should just leave things with the abbreviations, see HCC for more.

So in the end, noxious ‘because AI’ fluff aside, we have a 25% bump in DDR5 speeds from 6400 to 8000MT/s. This is a good thing when it comes in September, possible infrastructure upgrades aside. Same with the MRDIMM boost, 8000 to 8800MT/s is smaller but still welcome next year. Again the question of possible infrastructure upgrades and incompatibility with ‘Gen 2’ aside. All this, and the mocking could have been avoided with a technical brief, even a short one.S|A

The following two tabs change content below.
My Twitter profileMy Google+ profileMy LinkedIn profile

Charlie Demerjian

Roving engine of chaos and snide remarks at SemiAccurate
Charlie Demerjian is the founder of Stone Arch Networking Services and SemiAccurate.com. SemiAccurate.com is a technology news site; addressing hardware design, software selection, customization, securing and maintenance, with over one million views per month. He is a technologist and analyst specializing in semiconductors, system and network architecture. As head writer of SemiAccurate.com, he regularly advises writers, analysts, and industry executives on technical matters and long lead industry trends. Charlie is also available through Guidepoint and Mosaic. FullyAccurate