Intel to also have large chiplet cache. It coming to processors for PCs are unclear though
AMD processors are now the performance leaders in games (people likely didn’t see this coming, back in the early Ryzen days), thanks in large part to 3D V-Cache technology boosting the L3 cache capacity to 96MB, which can hold a good part of game’s hot data. It makes sense to expect Intel to come up with something similar to catch up to AMD. After all, they once had something similar in the form of Broadwell processors with eDRAM.
And it turns out that Intel does have such plans, and they will even see implementation fairly soon. On YouTube, Der8auer and Bens Hardware spoke with Intel’s Tech Communications Manager Florian Maislinger and asked about the topic. He confirmed that Intel is also planning processors with massive cache capacities, but they will take on a different form. And unfortunately for gaming enthusiasts, this technology is not yet coming to PC processors, currently there’s only talk about servers.
Maislinger said that Intel is preparing large L3 cache in servers, so instead of competing with X3D Ryzens, it will be processors competing with 3D V-Cach Epyc CPUs (AMD refers to them as the “X” series). But because Intel does not use the same silicon in desktop and server processors unlike AMD, this larger cache feature will not be easily transferred to Core (or Core Ultra) processors.
Premiere in Clearwater Forest
The first server processor with “Intel V-Cache” should probably be the Clearwater Forest generation Xeon, scheduled for next year. These will be chiplet processors consisting of a relatively large number of compute tiles. They should be mounted on substrate chips, but while the substrate tile is more or less a passive layer and serves only for interconnect routing in Lunar Lake or Arrow Lake processors, in Clearwater Forest processors this substrate tile will also be used to place the cache. So it will be a 3D cache in the full sense of the word. The difference compared to AMD is that there is not a single dedicated cache chiplet for each CPU core chiplet, but instead the die with cache is shared by several CPU chiplets, which should give advantages in core communication and data sharing.
It is not yet clear how many MBs of cache will Intel be able to pack into Clearwater Forest this way. Clearwater Forest is to use Intel’s 1.8nm process node (Intel 18A technology) for compute tiles with CPU cores, and the chiplets will be interconnected using Hybrid Bonding technology. It seems that the core architecture could be Skymont, so it’s going to be “E-Core” Xeon.
Unfortunately, there are no plans to make a version of this technology available in mainstream processors for (gaming) PCs. Or at least anything such is not close enough to launch for Intel to be willing to announce it. However, there is evidence that Intel has at least considered implementing it in PC processors. Meteor Lake processors, which also have a carrier base tile, were also considered for having an extra cache being integrated into the base, apparently. Intel patented such a solution and information about these plans or studies leaked to the Internet. This technology was apparently codenamed Adamantine Cache.
Read more: Like Crystal Well? Intel Meteor Lake CPUs have L4 cache
However, Meteor Lake (and also Arrow Lake) ultimately did not pursue this option. Perhaps the potential performance improvements were not enough to justify the increased manufacturing costs. But it doesn’t have to be this way forever, so someday 3D Cache may appear in Intel’s chiplet processors in the base tile. The fact that Intel uses a an actual piece of silicon that needs to be in the processor anyway for the carrier layer is an opportunity, it could be an advantage against AMD’s current solution where the cache needs to add a completely new chiplet (which is not needed in standard processors and thus makes those cheaper, however).
Games, too small a market for a specialized CPU?
According to Maislinger, AMD’s 3D V-Cache processors are geared towards a very specific group while posing some compromises for other uses due to the reduced clock speeds (it has to be said that this has been greatly mitigated with Zen 5, though). His statements in the interview sound a bit as if the market for such specifically-gaming CPUs is not big enough in Intel’s opinion. It’s a question if that is really true because given the company’s market position, the potential number of customers for special gaming CPUs from Intel should always be higher than what AMD can count on, and for AMD the gaming market apparently is not too small to be worth pursuing.
- Read more: Ryzen 7 9800X3D: Second generation V-Cache for peak gaming performance
- Read more: AMD Ryzen 7 9800X3D: Top “gaming” CPU with MT perf boost (review)
Perhaps we can interpret this as Intel wanting their PC processors to stay universal (and the approach of deriving a gaming modification from such a universal processor, like AMD does it, doesn’t suit them for some reason). It could also be that the gaming market, while not too small, doesn’t have as much room for technologies that make the processor more expensive to produce requiring the result to be sold at a high price, whereas in servers it’s less of a problem and customers are more willing to pay extra for advanced feature.
Sources: Tom’s Hardware, Intel
English translation and edit by Jozef Dudáš
⠀