The question is what their price will be, since current Navi cards are not priced aggressively against nVidia...
https://www.tomshardware.com/news/lisa-su-high-end-navi-gpus-are-on-track,40068.html
Wash, rinse and repeat. Doesn't matter how good the cards are if the drivers are crappy. And vice versa.
Then when they release the new cards, half the existing GPUs will be broken by the new changes. Thank goodness this VII experience taught me to stay away from everything AMD before jumping to a Ryzen 3000 series CPU. Doubly so, since Tom's Hardware is reporting users are lucky to hit max boost on one core even for some CPUs.
The paid-for-hire CPU genius that AMD used for the ryzen series is clearly gone.
qwixt Ryzen 3000 series CPUs - Mix of 1 "fast" and X "slow" cores (TomsHardware with AMD responses) aye.
ajlueke Not to mention Navi is less efficient than Turing, and AMD has a history with the RX 590 and Vega 64, as most recent examples, of not caring about power consumption as long as the card is faster, no matter that it results in a card that's louder and more expensive to own.
bearcat22 It's not all roses and rainbows on the green side either, lest we forget the couple of driver issues that resulted in bricked cards and denied warranty coverage.
That does depend on how aggressive they are with clocks. The vanilla 5700 is just as efficient as Turing, while the 5700 XT is worse, but nowhere near Vega levels. So odds are, it will be pretty close. I'm not expecting much from pricing at this point, they'll just slot them in based on NVidia's pricing structure.
This is a bit interesting. The custom XFX 5700XT Black Wolf is listed on a Chinese website for $475, which puts it -really- close to the RTX 2070 Super...
https://www.tomshardware.com/news/xfx-radeon-rx-5700-xt-black-wolf,40075.html
ASUS just announced the two STRIX models are coming this month as well.
It brings into sharp relief the downsides of additional overhead. In only one or two cores on a chip can actually hit the "boost clock", then the microcode, driver and OS all have to be "aware" of which core that is in order for single threaded loading to hit rated speeds. That's just more places for things to go wrong. If all the cores could hit the "boost clock" under single threaded conditions, there would be nothing to worry about.
On the other hand, it would be harder to manufacture chips that are as performant.
Not too hard to figure out based on AMD's current pricing strategy.
The 5700XT has a 2560:160:40 (Shader:TMU:CU) layout, which is identical to the RTX 2070 Super. However, the 5700XT is slower than the RTX 2070 Super, so we know that Turing is still faster than RDNA when all things are identical (by about 10% or so).
The RTX 2080 Super is a 3072:192:48 card. So we'll probably get a 3584:224:56 variant of Navi 12 that will perform similarly at the $699 price point.
Meanwhile, it sounds like there will be a Navi 12 (4096:256:64) version as well. That card should be faster than a RTX 2080 Super, but still well behind the RTX 2080 Ti (4352:272:68). Not sure how they would price that. $899?