We’re within the second day of Intel’s Innovation ’23 occasion, and our man Dave is on the scene. Intel held a Q&A session as we speak the place members of the tech press have been allowed to fling questions at CEO Pat Gelsinger. There was some fascinating dialogue within the Q&A, however arguably essentially the most fascinating query to our viewers was one about whether or not Intel would undertake a 3D cache strategy just like AMD and its “X3D” CPUs outfitted with 3D V-cache.
Intel did add-on CPU cache first, manner again in 2013.
Intel’s no stranger to chip stacking, nor to the concept of add-on cache for CPUs. The truth is, the corporate did it first, again in 2013 with Haswell CPUs that included the “Crystal Effectively” 128MB eDRAM cache die on the package deal. It labored like a System Degree Cache, in that it might cache reminiscence for any a part of the SoC. It improved the built-in graphics efficiency of CPUs so-equipped dramatically, however later, it additionally improved the CPU-only efficiency of the desktop Core i5-5675C and Core i7-5775C once they have been outfitted with discrete GPUs.
Gelsinger’s response to the query was lengthy, however right here it’s, lower barely for brevity:
“Whenever you reference V-Cache, you are speaking a couple of very particular know-how that TSMC does with a few of its prospects as nicely. […] That specific sort of know-how is not one thing that is a part of Meteor Lake, however in our roadmap, you are seeing the concept of 3D silicon the place we’ll have cache on one die, and we’ll have CPU compute on the stacked die on prime of it, and clearly utilizing EMIB and Foveros we’ll be capable of compose completely different capabilities.”
Gelsinger went on to say “we really feel superb that we have now […] benefits for 3D stacking for each little die in addition to for very massive packages for AI and high-performance servers as nicely. […] We’ll be utilizing these for our merchandise, in addition to presenting it to the foundry prospects.” This would appear to state outright that Intel plans to utilize stacked reminiscence applied sciences in each its CPU and GPU merchandise, together with the huge many-tile behemoths in its Knowledge Heart GPU Max vary.
We’re curious to see if these applied sciences and concepts will filter down into the consumer-facing Arc Graphics product line. AMD achieved a humongous efficiency and effectivity leap from RDNA to RDNA 2 by making use of liberal quantities of “Infinity Cache” to its GPU structure. It is going to be fascinating to see if Arc can shore up its middling effectivity in coming generations by making use of intensive caching.
