What other architectures are AMD planning in the near term then?
Near term? If you mean 2019, nothing. The smart money is on "small" Navi (Navi10, Navi12) this year and a larger die with more shaders, possibly TSMC 7nm+, and other upgrades next year to try to take on 2080Ti. AMD has limited resources at RTG, and people need to remember that a lot of those resources are aimed at enterprise. This assumption is also based on Navi's development being heavily guided by MS and Sony (for Xbox 2/PS5); that could be wrong, but we've already got reports that both 2020 consoles use Navi, so in a way it seems correct.
My money is on this year's Navi dGPU being similar in power to the best of the console Navi products we should expect next year. Allegedly, the better of the two Xbox 2 consoles (uh . . . Ananconda?) is going to have a 14 Tflop GPU, which would be about twice as fast as an RX590. RX Vega is "only" 12.58 Tflops in FP32 but 25.17 Tflops in FP16, so that still wouldn't necessarily put Anaconda's GPU in a superior position, and ditto for this year's Navi10. In my opinion. I could be wrong.
It should be noted that a few people outside of the Adored crowd (such as
@Yotsugi ) insist that Navi's development was not guided by MS/Sony completely (or at all), and that AMD is actually swinging for the fences here. Personally, I just don't see it as being likely.
I could be confused, but my impression was that Navi was the next generation of AMD GPUs, and that IF they wanted to compete at the high end that would be the GPU to do it.
Navi is, by all accounts, still related to GCN. The expectation is that the next major run of GPU designs they roll out, probably in 2021, will be the "generation next" or whatever it is they're been putting on their slides. No information exists that I know of to describe that design. Most people expect it to be clean sheet.
Based on what? What makes it highly unlikely? Rumors suggest there is at least two chips, possibly three.
Based on the high probability that Navi's development has been heavily influenced by its place in future consoles. AMD likes to repurpose designs where they can, so if they can sell something close to an Xbox 2/PS5 GPU as a dGPU this year, all the better. As I posted above, a 14 Tflop dGPU would replace Polaris easily and wouldn't necessarily suck up much power on 7nm. It would be a solid midrange product that could take on 1660Ti and 2060 while not costing too much to produce. There is no way that particular product winds up being a $499 Vega64 replacement. And AMD doesn't have the resources to do that many different SKUs based on Navi, hence the rumours that there are two variants coming out this year and other, more-powerful designs coming out next year.
I am cutting this discussion off because this is the graphics subforum and we have been off topic for far too long.
Fair enough.
Lisa Su has made it pretty clear that Navi (at least initially) is going to slot below the Radeon VII
It's also implied on this image:
https://images.anandtech.com/doci/14352/amd-shareholders-slide-2.png
Well yeah, it's definitely not going to unseat Radeon VII this year. I'm not even sure it's going to replace Vega64 this year though. See above.
No. Vega is the prime candidate for replacement.
Radeon VII already did that . . .
I won't be surprise at all if AMD leave Polaris lingering around for a while longer.
Navi would be about as cheap. The only thing more expensive about it is its reliance on a newer, more-expensive node. Polaris really needs to be put out to pasture, and Navi is a perfect replacement.
Almost tit-for-tat: ~GeForce RTX 2070 ["2075"?] level for 500 USD
Ambitious, but overpriced. If that is true - and I kind of hope that it isn't, because I'd
much rather see AMD aim its resources at underserved parts of the dGPU market - then it confirms that AMD is just as interested in running up dGPU prices as nVidia. Overall, it's a bad sign.
Facebook
Twitter