I see. Well the M1 Macs have GPU’s that are more capable than the PS4 but less than the Series S. Maybe there is a version of the Razer 14 that matches, I honestly am not sure since my initial thought for the reply was in reference to the M1X (in a 16” MacBook Pro). To be frank, unless they redesign the GPU cores again I am not sure the M1X is going to have feature parity with the Series S, let alone the X, though I expect it to be faster than the S at least.
High performance Wintel graphics workflow has a serious bottleneck when it comes to dispatching work or getting returns from the GPUs ... requests are formatted in main memory, compressed, transmitted over PCIe, received by the GPU, decompressed into GPU memory, and executed. Return GPU compute tasks have the same problem going the other way.
All this takes real time clock tick overhead that is not directly contributing to getting pixels on the screen, or getting compute results.
Apple Silicon circumvents some of this with unified memory, where the CPU and all the IP blocks and GPU share the same memory so all that copying doesn't have to take place - meaning that with the unified memory model you get a lot more bang for the buck.
Comparisons of actual benchmarks aren't affected by this - after all, those overheads are already baked in.
It's discussions of the internal processing speed of discrete GPUs where this is a factor: which assume the request and data are already sitting in GPU memory ready to be processed. These are the speed discussions which fail to account for the Wintel graphic processing workflow overhead, and that is why faster discrete GPU's real and practical results can be surprisingly close to the 7 or 8 GPU cores of the M1 which eliminate those overheads.
Not sure how consoles handle the graphic workflow - though being PC-like I'd assume they follow the Wintel model. OTOH, at their price point maybe they use a unified memory model - though I doubt it.
CPU/GPU manufacturers are pretty set in their ways, and the economic model which creates this workflow is unlikely to change. These vendors are pretty set on optimizing their bottlenecks (faster PCIe buses, hardware compress/decompress) rather than revisualizing the process which is the basis for their bread and butter.
