A series of leaked documents seem to confirm that AMD is planning to bring back a key consumer feature to its highest-end lineup. server microprocessor So far, EPYC Genoa-X.
3D V-cache launched on Ryzen 7 5800X3D and has now been rolled out to three more consumer-focused processors – with four 3rd generations EPYC The processor also adds 768 MB 3D L3 cache in 2022.
This technology allows more cache to be allocated to CPU cores. This is particularly useful in a number of applications: AMD deals with computational fluid dynamics (CFD), finite element analysis (FEA), electronic design automation (EDA), and analysis. Structures are the use cases that are likely to get the most improvement.
AMD leak
A leaked spec sheet mentions 100-000000892-04, a 96 core/192 thread processor likely called the EPYC 9684X, which resides in the SP5 socket and can operate in a dual CPU configuration .
Total available cache capacity (L1, L2 and L3) is more than 1.25 GB; 10x more than Intel’s flagship product xeon, Platinum 8490H can deliver and will be a new world record. Other non-3D members of the Genoa family, like EPYC 9654, have about a third of the L3 cache.
The other three SKUs are expected to have a potentially TDP of 400W (up from 280W for current “ROME” processors and up from 360 for EPYC 9654 non-3D caching. Turbo speeds remain the same compared to the EPYC 9654 non-3D caching. with the following processor and a 200 MHz boost over the 64-core EPYC 7773X.
L4 in the future?
The role of caches in modern computing cannot be underestimated: of the many levers available to processor designers to improve performance, this is one of the most important and one of the most important. no wonder AMD might consider rolling out an L4 cache in the future. In general, more cores means slower speeds, and the need for fast memory (aka cache) closer to these cores to keep them running (rather than idling). It also introduces more latency and other compromises to the CPU floor plan.
With 96 cores and a 128 core version slated to come out in the near future, that might come sooner rather than later. ONE patent 2020 called “Direct Card Support in Virtualized Environments” shows AMD’s willingness to explore exotic solutions like multi-gigabyte L4 caches.
Of course, all this is nothing new. Others before AMD did it: IBM’s Z series for example had 128MB of L4 cache, and as reported by Anandtech returns in September 2021Big Blue is working on a new chip (Telum, now in z16), where each individual L2 cache can hold its own virtual L4 memory (note the use of virtual words).
