The main perk is if you can overclock since the cache is now under the cores. Being able to lock all cores to 5.4 would make me buy one.
Do you get that much performance anymore tho? What with AMD not leaving much performance on the table as is and letting the CPUs go as safely fast as your cooling can handle.
It will become 10% less over 7800x3d when Win update will introduce a bug that auto-enables/forces RECALL !!!
It's now an NPU right ?
I haven't seen Recall myself so I can't say if there's a visible (visible with your own eyes, not with benchmark tools) performance loss. I can definitely see everyone here disabling it even if the performance loss was 0.001%, though mostly because it creeped everyone out
So no magical uplift of productivity in new X3D CPUs as some have prophecised.
Not surprised there. These are not "productivity" CPUs, unless you have a productivity task that relies a lot on cache. Not denying they perform well enough for nearly all use cases, but the V-cache thing makes the X3D CPUs more of a gaming-oriented CPU.
Also, the 8-core 9800X3D vs the 14700K that has 20 cores isn't a fair comparison by Puget.
5.2GHz is a bit underwhelming. I imagined 5.4GHz would've been attainable, but I guess not?
AMD needs all processors of a certain SKU to be able to reach specific base clocks and boost clocks, and needs to be able to produce at least a certain amount of them, so this all lines up with the worst chiplets that AMD is willing to tolerate to use for the 9800X3D to reach that number of units produced. Anything that doesn't meet the criteria gets used in other SKUs.
I don't have any dual-CCD X3D chips, but isn't there now a driver to compartmentalize workloads between the 3D and non 3D parts of the CPU?
There's some software-side stuff but depends on Microsoft as well so... it's not optimal. Intel's Thread Director is hardware-based, for example, and actively participates in the scheduling process, at least according to what Intel says.
AFAIK, and let anyone else correct me if I'm wrong, AMD only passes the CPU information (number of cores/threads and little else) to the OS scheduler and lets the OS scheduler handle everything else, so the CPU doesn't really participate like it happens with Intel.