Less cores and “more” cache than the competition.
Intel’s upcoming generation of scalable Xeon processors, known as Granite Rapids-SP, is scheduled for a late 2024 release. After a challenging 2023 that saw multiple generation releases of data center processors, the coming months will be more demanding.
Launching a refresh with modest increases in cores and cache is a different challenge compared to a comprehensive upgrade involving an actual new core architecture on cutting-edge processing nodes with a more advanced package is critical for the company to stop market share bleeding to superior AMD EPYC processors.
The release of Intel’s next-gen Granite Rapids-SP will serve as a test of the company’s roadmap execution under the leadership of CEO Pat Gelsinger.
As per the latest Intel SDE (Software Development Emulator), Granite Rapids is set to boost its L3 cache to an impressive 480 MB, which is 1.5 times more than the 320MB found in its predecessor, Emerald Rapids, which “launched” in the middle of December 2023 featuring up to 64 Raptor Cove (RPL) cores and two massive “modular” tiles compared to four tiles.
For comparisons sake, AMD’s latest crop of EPYC / Threadripper processors max out at 384 MB of L3 cache featured amongst top of the line EPYC processors and the Threadripper PRO 7995WX.
But of course, AMD doesn’t simply play by the rules of conventional computing, thanks to 3D stacked V-Cache, Genoa-X processors dramatically increase the overall cache capacity from 384 MB to 1.125 GB of L3 cache in the form of the EPYC 9684X processor.
To put it into more simplistic terms, that’s 48 MB of L3 cache per core. Try as Intel might, they still won’t succeed.
The increased cache capacity of Granite Rapids aligns with Intel’s strategy to enhance its Xeon processors for generative AI, catering to the increasing demand for large cache reserves as training models become more extensive, but if their latest several generations of server processors are anything to judge by Intel won’t manage to recapture the performance crown anytime soon.
The flagship variant of Emerald Rapids-SP increased the L3 cache from 112.5MB featured with Sapphire Rapids to 320MB, coupled with the addition of various accelerators, hopefully making it a formidable AMD competitor when it comes to data center applications.
This move reflects Intel’s ongoing efforts to make Xeons more attractive to FAM (Finance, Automotive, and Manufacturing) sectors, as well as the continuous influx of AI startups, no doubt Intel would be trying to venture out into new unexplored markets considering just how severely they are bleeding market share in the enterprise sector with each passing quarter.
The upcoming Xeons based on Granite Rapids will incorporate Lion Cove cores, much akin to the unreleased Lunar Lake. Which hopes to leverage Intel’s own N3 process node with immensely sophisticated chiplet packaging.
Recent leaks suggest a regressive core count of 56, if that’s actually the case then rest assured that this platform will be certainly dead on arrival in all performance fields outside of AI.
Granite Rapids is anticipated to be launched in the latter half of 2024, preceded by Sierra Forest. Which is expected to feature up to 144 Crestmont “E” cores and 108MB of L3 cache.
Both processors are designed to utilize the LGA4710 socket with a TDP of 350W, and with early samples operating between 1.2 to 2.6GHz, this isn’t exactly a decent indication that the Intel 3 processing node showcases a resurgence in efficiency for Intel, more rather it’s likely that their Core architecture is what’ll be holding them down and making their processors hot, hungry and slower as a result.