AMD GPU Appears to Leave Room for Future 3D V-Cache
Semiconductor engineer Tom Wasick discovered He attributes it to the 3D V-Cache feature of one of AMD’s best GPUs, the RX 7900 XT. The engineer looked inside the 7900 XT’s die with thermal imaging and he discovered a 3D V-Cache connection point, the same type used in AMD’s Zen 3 and Zen 4 architectures. Wassick found the connection of the MCD die.
Wassick doesn’t know if these TSV connection points will be used specifically for caching purposes, but AMD currently has plans to expand its 3D packaging capabilities beyond vertically stacked caches. So it looks like these connection points are used with some sort of 3D cache in mind to improve gaming performance or computing performance.
This discovery comes after many unconfirmed rumors that AMD would add 3D V-Cache technology to its GPUs.
3D V-Cache has so far been used in AMD’s Ryzen and EPYC CPUs with great success. The technology relies on hybrid bonding techniques to fuse an additional 64MB cache slab on top of a Ryzen or EPYC compute die to increase L3 cache capacity. Now, with this 3D stacking technology, AMD is doubling the amount of L3 cache available on its desktop Ryzen 9 7900X3D and 7950X3D parts, and tripling it on its Ryzen 7 5800X3D, 7800X3D consumer chips and EPYC Milan-X server processors. I can.
The performance benefits from this technology are impressive, and the 3D-V-Cache chip delivers a full generational performance boost in applications that greatly benefit from large chunks of cache. A good example of this is the Ryzen 7 5800X3D, which saw a 28% improvement in gaming performance over the Ryzen 9 5900X and 7% faster performance than the Core i9-12900KS.
AMD’s server-ready offerings are even more impressive, with AMD and Microsoft’s Milan-X benchmarks showing well over 50% performance improvements over standard Milan parts. However, this technology cannot magically improve performance at will. This type of behavior is seen only for cache sensitive workloads.
What else do you see? A linear array of ‘spots’ very similar to the keep out zone in X3D, on the same 17-18 μm pitch. Any chance you’re looking at a stacked MCD feature (or something else)?January 27, 2023
I’m not sure how 3D V-Cache works in GPU applications. In theory, however, the main principles of 3D V-Cache still apply. More cache capacity means the GPU has to make fewer trips to his slower GDDR6 memory, so it can speed up processing of cache-sensitive workloads.
AMD’s Infinity cache for the RX 6000 series allowed AMD to use slower GDDR6 memory and maintain the same performance as Nvidia’s RTX 30 series GPUs. An infinite cache that keeps feeding data to the GPU.
However, I’m not sure if the same behavior applies to 3D V-Cache. This all depends on how sensitive AMD’s GPU architecture is to additional cache capacity, and how many applications will benefit from it.
Another issue AMD has to deal with is heat. This issue is prevalent in AMD’s Ryzen X3D processors. Additional cache slabs impede heat dissipation, resulting in lower CPU frequencies and higher temperatures (compared to non-X3D parts) at the same time. It’s likely that AMD will address the same issue with its 3D V-Cache GPUs, forcing them to lower clock speeds to keep temperatures down.
Nonetheless, it’s cool to see AMD possibly considering the idea of adding a 3D-Vache to their GPUs. We can see AMD’s next silver bullet to “magically” boost gaming performance.