Extremely anticipated: Hypothesis round AMD’s new 3D V-Cache know-how has swirled ever since Dr. Lisa Su gave us a sneak peek at Computex 2021. Since then, AMD and tech lovers have remained cautiously optimistic concerning claims that the brand new chiplet-stacking method can yield substantial efficiency beneficial properties with minimal impression to latency, responsiveness, and total performance. A current check of an EPYC processor with V-Cache is giving early indication that AMD’s efficiency uplift claims may maintain true.
Nobody was fairly certain what to anticipate when AMD introduced their 3D V-Cache technology at Computex final summer season. Whereas some lovers noticed the substantial improve in cache as an thrilling improvement, others in the neighborhood discovered themselves upset that the brand new choices wouldn’t supply substantial will increase in clock velocity, enhancements in energy draw, and so forth. Final Friday tech information outlet Chips and Cheese revealed results of their preliminary testing with certainly one of AMD’s new Milan-X processors with 3D V-Cache, the server-oriented EPYC 7V73X. And to this point, issues look promising.
In response to the location’s abstract, AMD has managed to considerably improve a processor’s cache measurement (768MB) compared to the earlier Milan household of processors (256MB). Testing by Chips and Cheese studies spectacular efficiency from the stacked CPU and far bigger L3 cache with out incurring any vital improve to cache and reminiscence latency. Preliminary testing reveals the latency penalty protecting the rise someplace between three to 4 cycles.
If these preliminary findings maintain true for AMD’s upcoming AM4 and AM5 releases, such because the Ryzen 7 5800X3D, then the chipmaker will undoubtedly proceed exploring the probabilities and advantages related to 3D chip stacking.
AMD’s present 3D stacking know-how entails bonding a single V-Cache chiplet to a processor’s present core advanced die (CCD) and cache. Because the know-how matures It might be doable for future architectures to additional develop their L3 cache capabilities utilizing extra chiplets.
We’ll have to attend and see what the long run holds, but when the EPYC-based outcomes are any indication of what’s doable, then AMD may ship one other sizeable efficiency improve with their subsequent spherical of CPUs.