Originally Posted by ChevChelios
in 1 game, on specific AA settings, on specific (non max, bc Fury X cant handle max) texture settings only
just keep on playing that Doom, but make sure never to touch the settings lest Fury X falls behind or cant run the game anymore
That's what a game looks like when it make good use of ACEs which disproves what you said about "tanking efficiency" when talking about ACEs.
We're at a point where Async Shaders are barely being scratched, Doom on Vulkan offload the entirety of the best Temporal Super Sampling AA there is into the ACEs, making literally no performance difference, thanks to ACEs a Super Sampling effect drops the performance by a whooping 0 FPS, TSSAA is 100% offloaded so the GPU can use this freed up performance in boosting FPS or other graphical presets.
ACEs can be used to offload much more, it's a matter of time before developers begin to use this more extensively, Nvidia's implementation of Async Compute is mediocre at best, how much efficiency are they wasting by using a serial GPU method with fast context switching to process TSSAA when they could have used ACEs instead for a mere 5W more power usage?
And the fury can't handle nightmare settings because it's got 4GBs of VRAM while nightmare needs 6GB, but I'm not talking about the card here but the technology that can be implemented into future products which will obviously have more VRAM than we need making your point moot.
Nvidia is probably losing efficiency for not using ACEs, Maxwell was incredibly efficient and that's what's carrying Pascall through their lack of ACEs, although a RX 480 close up to a GTX 1070 when using Vulkan on Doom thanks to Async Shaders, this is a 232mm2 card coming close to a 312mm2 card.
Saying that ACEs tank efficiency is ridiculous at best, when they're used properly they're a huge boost for performance, 5W extra power consumption for 20% extra performance (20 FPS) seems extremely efficient to me, and this is with Doom, a game that is as neutral as they get, used in Nvidia's own presentations, but not on AMD's, a game that still doesn't exploit Async Compute anywhere near its max.
Originally Posted by Kpjoslee
Poor comparison since RX480 is full die while 1070 is cut die. 1080 is also 312mm2 card lol. In terms of usable transistors, 480 and 1070 should be pretty close. Not to mention 14nm GF chips are slightly smaller than 16nm TSMC chips to begin with.
Probably, but bear in mind the RX 480 is a card stretched out of its "comfort zone" regarding its performance bracket.
Just see how the RX 470 have 12% less cores and 5% less boost frequency compared to a RX 480 and yet the RX 470 have a WHOOPING 40% less TDP, where the hell did that extra efficiency came from? It even looks like an entire different chip!
It's obvious that Polaris gets incredibly efficient at the lower ends and proves my theory about the RX 480 being stretched out of its intended performance bracket, and goes in line with what Raja said in that PCPer interview, that Polaris focus was on P11.Edited by Dargonplay - 7/29/16 at 6:26pm