AMD basically wants an ALL compute future, a la Intel Larrabee style ...
It makes sense why.
1.
If compute tasks take off, Zen APU would appear to have a key competitive advantage.
2. If compute takes off, AMD may have an upper hand again with Polaris/Vega. Extra free performance is already baked inside GCN architecture.
3. Try to shift the pro market away from CUDA to OpenCL. To do that, having a strong compute foundation is a good start. Also, it allows AMD to make a single chip that serves compute and graphics markets.
4. If AMD wins Nintendo NX, for next gen consoles in 2019-2020, there will be a lot of pressure on MS/Sony/Nintendo to continue with an evolved GCN architecture for backwards compatibility. That means realistically by 2019-2020, there will be AAA games being made for PS4-PS5/XB1-2/Nintendo NX, or 5 consoles! All major developers will be optimizing for GCN. It's why Nvidia is starting to mimic a lot of the GCN architecture with Pascal.
Brand loyalty matters and surely Nvidia is the stronger brand. But AMD was actually gaining market share in Q1 and Q2 2014 with Hawaii and reached 38% market share before Maxwell launched and destroyed AMD's market share. AMD had older, slower and much more power inefficient cards. AMD did not turn up to the fight and they were decimated.
That's probably because NV was clearing out old products ready for GTX750/750Ti launch. If you look at NV's shipments, there was a huge dip during this time which means AMD "gained" market share when they really didn't. Look at the unit volume sales instead. AMD has been tanking consistently, BEFORE 970/980 launched.
https://www.techpowerup.com/forums/threads/nvidia-geforce-titan-x-12-gb.210662/page-7#post-3257195
You are not telling the truth. AMD made serious market share gains and for the first time after AMD bought ATI they had overall GPU market share lead. AMD had > 40% desktop and close to 60% notebook GPU share.
You are only looking at the first start of that HD5800 generation. Generations roughly last 2 years or so. What happened during HD5800/6900 vs. GTX400/500 generation? You know well. AMD got decimated. Most of the key metrics that NV fans have touted with Kepler and Maxwell as key selling features, AMD won during those generations.
RS I sometimes get the feeling you are just too impulsive to jump to conclusions. Lets see how Polaris 10/11 and GP104/GP106 turn out in performance and pricing.
I am connecting the dots. There are repeating rumors of 970->980Ti line-up being discontinued and GP104 taking over the price points of those cards. There are no rumors of Fiji cards being discontinued. There are rumors that Polaris 10 is a small chip (232mm2) with only 2304-2560 shaders. How in the world is that enough to beat 980Ti by 20-30%? AMD themselves said their goal was to bring 290/970 spec to lower price levels. Since we were shown Polaris 11 as a 950 competitor, it stands to reason that Polaris 11 isn't going to be the card bringing 290/970 performance level. That leaves us with Polaris 10 as that chip. I don't understand how can Polaris 10 be able to go head-to-head against a 1080 that should outperform 980Ti by 25-35% and yet it's also bringing 290/290X performance < $299?
Remember the volume is at lower price points so if Polaris 10 and Polaris 11 beat the competition's cards at comparable price points and provide comparable or better efficiency then they will do very well.
Again, if I take perf/watt of Maxwell and Tonga, and apply 2.5X to AMD architecture, and 2X to Maxwell, Pascal wins. That's because AMD is massively behind AMD in perf/watt unless it's the Nano. On AMD's road-map, the 2.5X perf/watt isn't over the Nano. Again, I am just connecting the dots here. Doesn't appear to me that AMD's Polaris will somehow beat Pascal in perf/watt. At least there isn't sufficient data to support this assertion.
AMD need a card like HD 4870 which is priced to sell and tremendously competitive. Remember HD 4870 never beat the GTX 280 but it brought AMD back in the game after the HD 2900XT fiasco. The times now are more difficult for AMD as they have gone to historical lows of market share in more than 2 decades (including ATI). AMD need good products with great pricing to get back to traditional 35-40% market share.
- Having lots of market share but making little $ (HD4000->6000 generation) isn't the answer. It's why Lisa Su finally abandoned this strategy.
- HD4870 also beat $100 more expensive GTX260. When I got my 4890, it cost barely more $ than a much slower GTX260 216 and less than the GTX275. Back then AMD delivered more performance on the mid-range and their cards cost less, at the same time. This would be like having Vega 11 priced at $299 and beating $399 NV Pascal. Why do you think that's going to happen? AMD is content selling slower performance for $600+ with Fury X.
Also, if Polaris 10 brings 390X performance for $249 at 120W TDP, how is that ground-breaking?
970 cost $329 > 1.5 years ago. Today, this card sells for
$280 with a $30 game.
To me ground-breaking would be bringing 980Ti/Fury X performance to $299-349 price level in June. Do you honestly believe Polaris 10 can do that with 256-bit bus and 8Gbps GDDR5?