Memory bandwitch / TF: PS5 and SX are the same.
Yet the XSX GPU can read more data in parallel across the bus on a given cycle (40 bytes) compared to PS5 GPU (32 bytes). Which means for any same amount of data both systems are reading from RAM through the GPU, the XSX does it faster because of the wider bus.
In other words, XSX's GPU bus gives it a 4:5 ratio over PS5's GPU when it comes to every fifth consecutive bus access (4 x 40 = 160, 5 = 32 = 160) in data transmission by the GPU across the memory bus, which helps a lot when dealing with large-ish (or larger) data objects in memory to have the GPU process (I know this explanation's a tad sloppy but I'm trying to hurry up to take care of some work).
Yet on XSX each CU can make more accesses across the bus per frame for data in a memory amount equivalent to the CU compared to PS5. See above.
unified memory> divided memory.
What do you mean by 'divided memory'. Clarify. On XSX the pools are not physically split, the partitions are virtualized by OS management.
There is no required shadow copying of data between the two pools just because CPU-bound tasks may take to one pool and GPU-bound tasks to the other pool. It's not equivalent to split memory pools of system RAM/VRAM on PC, or other consoles like XBO (DDR3/ESRAM), PS3, and almost every console older than those.
only advantage of Xbox SX: use UP TO 10GB, which is impossible because the OS is in the remaining 6GB slow.
OS being in the 6 GB pool doesn't prevent the GPU from accessing the full 10 GB of its optimized pool, this is dumb. This is like saying on PC a GPU card can't use its full VRAM because the OS is in system RAM. That's not how it works.
The location of the OS in terms of the virtualized pools has no impact on the amount of physical memory the GPU can address because the OS is still managing entire system resource accesses. The GPU is treated as a co-processor , not a central processor with its own OS and memory partition like in a dual-boot system (Linux on PS3 for example prevented that OS from accessing full system resources).
isn't the "geometry engine" whatever pretty much sony's vrs equivalent?
Not necessarily. They may have implemented VRS equivalent features in that part of the pipeline per their customizations, but Geometry Engine itself (as well as Primitive Shaders) were already present on earlier RDNA1 GPUs, so chances are high that XSX also has a Geometry Engine, but maybe AMD have it under a different name in newer GPUs? Primitive Shaders were present in RX Vega, FWIW.
Basically, Geometry Engine and Primitive Shaders (both also referenced in the Eurogamer article) are not Sony-coined terms; AMD was already using them from the start with RDNA1 GPUs and at least are likely using Geometry Engine still to describe that part of their GPUs in RDNA2 (Primitive Shaders have been replaced with Mesh Shaders). The functions of Geometry Shader have little in common with VRS, though.
