Yes, there is only one cache available.
Yes, there is only one cache available.
They rely on chiplets that work separately. If they were to exchange cache, both the CCDs and the 3D V-Cache would need to be built together rather than remaining modular. Sharing the 3D V-Cache requires constant communication between cores to keep data consistent. Currently, CCDs can't talk directly to each other. One must connect to the IO die, which then forwards the request to the other CCD. Unlike Intel, AMD doesn’t use a silicon interposer; instead, signals travel straight through the PCB, which greatly raises latency.
This would imply that in gaming, relying on two CCDs wouldn't be effective or offer low-latency benefits compared to using eight cores, no matter what the V-Cache is.
I don't get the point you're making. Using just one CCD with 8 cores is better than splitting it across two with 16 cores each. With 3D V-Cache, it's still preferable to run on a single CCD rather than connecting two different ones.
Ya thats what I meant. 1 CCD will always be better for pure gaming than having 2 CCDs, regardless if there is 1 V-cache, or 2. How is the 9950x right now with games? Is scheduling good enough to run all games on first CCD, or we still need lasso, etc?
I believe the scheduling process functions properly once it operates, but the main issue is its complexity. The inter-CCD latency challenge existed from the start of Ryzen, though few paid attention until AMD's processors accelerated and performance gaps became clear. Especially noticeable was the speed difference between single CCD X3D units and dual CCD ones. In the Ryzen 7000 series, they introduced scheduling, but getting it to work has been problematic for everyone. Check out this video. Using the Wintel concept makes me wonder why Microsoft hasn't included this in the kernel. Linux seems to handle core parking well, and benchmarks with dual chiplets—particularly X3D chips—effectively bypass the other CCD.
The impact seems unlikely in games because introducing the CCD adds a latency penalty that would nullify any advantages from using two CCDs, though it might help certain productivity tasks.
Accepted, though keeping the identical "layout" across both CCDs simplifies voltage regulation, power handling, and overclocking, while reducing performance fluctuations when switching from a non-x3D core to a x3D one.
The 2 CCD design might continue to challenge AMD for a long time, particularly in gaming. Intel's CPUs perform much better here, offering strong gaming and workload capabilities without major trade-offs—just some differences in temperature and power usage.