I think anybody trying to downplay Navi needs to see what's going on here.........Turing was just released a few months back......Jensen Huang blew a gasket over Vega VII because nobody is buying 2080 ti's...In reality, AMD does not need to compete with a card few people buy, yet they will aim for 2080ti perf with NAVI......Maybe not this year but eventually at a lower price, with most likely more VRAM and some smoking Raytracing support...
Folk should ask themselves, why are there rumors of a RTX/GTX 2070ti and a refreshed 2080ti? You think Nvidia would just do so? Refreshing the 20 series so soon....From 14Gbps - 16Gbps, will allow much better bandwidth and they also have plans to boost clocks as well.....Why would Nvidia do that if all that's on the horizon are RX 580 and 590 replacements? Anyone remember when Vega 56 came on stream and waxed the GTX 1070? Nvidia went on to develop the 2070ti to combat the Vega 56, they boosted the 1060's memory setup to combat the RX 580 etc...
If Nvidia was only worried about 75 watt Rx 580 perf on 7nm, they would never bother with boosting their current lineup...
Speculation Part on Navi Tech
I think we've already established that Navi CU's are more potent than Fiji and Vega by a good margin....Less CU's will give you much more performance over Vega....,It's very likely that 1 Navi CU = 4 Vega CU's at optimum........Yet, how can AMD break the CU limit for even more breakthrough performance and RTX processing power? They could put 2-4 full 64CU chips on a single wafer and introduce an Infinity Fabric multi-gpu setup.......I think that's the future.......Even if they can't get 64 CU's per chip right now, they can do 4 clusters of 20CU's, 24, 32, 40 and scale up to 56 and 64 clusters in time, maybe not even right away, but that could be a roadmap.......With a two chiplet GPU design we're looking at 40, 48, 64, 80 CU's and 112 and 128CU's on the extreme end.......With a 4 chiplet GPU system, we're looking at 80, 96, 128, 160, 224, 256 CU systems scaled up.......
First gen/release Navi may very well be a monolothic setup, especially for the low end to mid, but I think the ground work has already been established to push GPU technology for the longterm.......And I dare say, we're are not going to go anywhere for much longer if all GPU manufacturers are looking to do is simply wait for another die shrink to increase clocks and lower thermals.......Someone must be thinking ahead to push GPU technology forward.....
Im thinking about two things that Lisa said makes me think, they're going that route.......Sometime last year, Lisa Su said that they were doing heavy Research and Development into compute......and more recently she said...
------------
This builds on Dr. Su’s comments at an interview with Barron's a month prior where she asserted that AMD will be “competitive in high-end graphics,” and that AMD is “making high-performing quality products and building
a solid long-term foundation.”
And TSMC’s 7nm process is key to all of it.
AMD CEO Lisa Su – Q3 2018 Earnings Call
“We see significant opportunities to build on this momentum as we transition to our next generations of high performance products and launch the industry’s first 7-nanometer x86 CPUs and discrete GPUs over the coming quarters.”
------------
The bolded is what stuck with me the most, nobody saw threadripper coming, but it was a genius move relative to CPU design....Just Like Ryzen, AMD knows it has to do something revolutionary in the GPU space and Navi has been cooking for a while.........There are many small hints that they are going the multicore route, not just for CPU's but for GPU's as well....I hope they are able to nail it.....
