| HollyGamer said: Ok so the point is, Polaris are on equal terms with High End GPu from previous year |
That is generally what should happen. A high-end parts performance level should filter down into the mainstream at some point, usually with the next generation or shrink.
But that hasn't happened with Polaris meeting Fury's performance level just yet as Fury still even out-benches the RX 590... And it's almost 4 years old! Which just reaffirms my prior points of how AMD has blatantly stagnated.
| HollyGamer said: Why the hell bring up flat earthers argument in here. Both are different matter. |
Er. Because both have youtube videos? I believe in empirical evidence, not opinion pieces from a youtube video.
| HollyGamer said: False??? yup , you are the one who arefalse, you don't get the entire big picture on how and low iGPU Navi can be as powerful as mainstream GPU . Imagine how powerful dedicated mainstream desktop GPU is? |
I don't think you fully comprehend how bad AMD is in terms of performance/watt when compared to any of nVidia's efforts... At the moment it is almost impossible to recommend AMD's GPU hardware... Only their CPU's. - Which is a complete reversal from a few years ago where you couldn't recommend their garbage FX CPU's, but you could recommend their GPU's.
At the end of the day... AMD's most powerful GPU ever is Vega 7. - That isn't going to change when Navi releases. Simple as that.
And Vega 7 does leave a user wanting considering how prohibitively expensive it is to manufacture.
drkohler said:
I don't think you understand what Permalite is saying (or in what troubles AMD is with its design philosophy). |
Indeed. With Vega 64 AMD spent the bulk of it's 3.9~ billion transistors over Fiji in adding additional stages to each cores pipelines.. And used transistor space to add "dark silicon" between areas to reduce leakage/noise and introduced mechanisms to lengthen a cycle so they can clock the GPU up significantly.
It's also what nVidia did with Pascal, hence why Pascal GPU's tend to have higher transistor counts over Maxwell for a similar amount of functional units (Remember Pascal isn't a big deviation from Maxwell from a functionality perspective either!)
Contrary to popular belief... Driving up clockrates on processors isn't just about dumping more voltage into the chip or building the chip on a smaller node.
Granted AMD did introduce some extra half-arsed features into Vega as well which took up some of the transistor count... But the bulk was essentially spent to drive up clockspeeds in one way or another.
In the end though, depending on what kind of library you use, 7nm isn't a 3x density over 14nm anyway. - If you were to compare worst to best, maybe... Apples to apples you are probably looking at a 1.5x-2x improvement.
What makes matters worst is that TSMC, Samsung, Global Foundries etc' 14nm process isn't a true 14nm process anyway, they tend to have a 20nm BEOL... But enough about that, I could rattle on about fabrication for weeks.
--::{PC Gaming Master Race}::--
