Pemalite said:
teigaga said:
Ermm... Ok.
Until you care to elaborate in a meaningful response, I'm sticking with the widely akcnoweldged notion that higher resolutions are a GPU drain, hence PS4 Pro having 4.2Tflops versus 1.8, and rendering Horizon at 2x1080p instead of native 4k. Hence why several games utilising Pro have improved graphics at 1080p whilst having lower graphical settings in 4k (which isn't even native 4k)
|
Fine. The response is as follows.
What if a game on the PS4 is built completely in FP16 or double precision? Your flops as you state them aren't an accurate representation as it is not being used.
Not to mention that you excluded the Render Output Pipelines, Texture Mapping Units, Geometry units, Caches, various fixed-function units, Bandwidth and Memory amount... They all play a role at running a game at a higher resolution you know.
Not to mention that AMD and nVidia engineer their GPU's so that the performance hit from running at a higher resolution tends not to be a linear decrease, so you cannot just double your flops and expect double the resolution.
So I shall ask again, Can you explain to me how your flops is somehow tied to a specific resolution?
|
No one said X flops is tied to a specific resolution, I think thats your own asserstion. I made a generic statement regarding Scorpio's 6Tflops of compute power being drained by native 4k resolution demands & not leaving a lot of room to accomodate whatever PS5 sets as the graphical standards. If you just wanted to say flops alone are not a linear and accurate portrayal of a system capabilities, you could've saved a lot of time lol. Its called laymen. No one component will give you an accurate depiction of a systems performance benchmarks, but to pretend a zero correlation between AMD flops and the surrouding GPU components and overall performance, seems a bit disingenious, no? I mean even with more efficient achitecture Vega will offer over polaris, where are Scorpio's known specs going to offer a highly disproportionate difference suggested by the 30-40% tflop difference between it and Neo? Extrapolating on Teraflops performance here seems logical.
