AMD explains how the brand new 3D V-Cache improves over the unique

In context: AMD launched the Ryzen 9 7950X3D on the finish of final month and welcomed an enthusiastic response to its second-gen 3D V-Cache regardless of some blended opinions about its usefulness in a 16-core CPU. Now they’ve shared a number of the technical particulars that designate its efficiency.

AMD began mixing nodes in 2019 when it used the 7 nm node for the core complicated die (CCD) and the 12 nm node for the IO die of the Zen 2 microarchitecture. AMD just lately confirmed to Tom’s {Hardware} that Zen 4 steps it as much as three nodes: the 5 nm node for the CCD, the 6 nm node for the IO die, and the 7 nm node for the V-Cache.

The V-Cache sits over the center of the CCD and the eight cores flank the perimeters.

AMD defined a number of the challenges it confronted stacking one node onto one other throughout its current ISSCC presentation. Each the 7950X3D and the unique 5800X3D have their V-Caches positioned over their common L3 caches to permit them to be linked. The association additionally retains the V-Cache away from the warmth produced by the cores. Nonetheless, whereas the V-Cache matches neatly over the L3 cache within the 5800X3D, it overlaps with the L2 caches on the sides of the cores within the 7950X3D.

A part of the issue was that AMD doubled the quantity of L2 cache in every core from 0.5 MB in Zen 3 to 1 MB in Zen 4. But it surely labored across the further area constraints by punching holes by way of the L2 caches for the through-silicon vias (TSVs) that ship energy to the V-Cache. The sign TSVs nonetheless come from the controller within the middle of the CCD however AMD tweaked them too to cut back their footprint by 50%.

The Zen 4 L2 cache is bigger due to its bigger capability, but in addition as a result of it has TSVs passing by way of it.

AMD shrunk the V-Cache down from 41 mm² to 36 mm² however maintained the identical 4.7 B transistors. TSMC fabricates the cache on a brand new model of the 7 nm node that it developed particularly for SRAM. Consequently, the V-Cache has 32% extra transistors per sq. millimeter than the CCD regardless of the CCD being manufactured on the a lot smaller 5 nm node.

The entire refinements and workarounds AMD applied add as much as a 25% improve in bandwidth to 2.5 TB/s and an unspecified improve in effectivity. Not dangerous for 9 months between the primary and second generations of a supplemental chiplet. Hopefully it exhibits its worth when the Ryzen 7 7800X3D arrives in a month’s time.