This shows how nVIDIA purposefully designs their architectures to be viable for a 2-year period, then marks them down with inadequacies or abandons driver support for them.
That's
really not how chip design works. Nobody wants to build an architecture that doesn't last, because that means having to reinvent the wheel down the line. Building a great base and then wringing more performance by improving it over time is a much smarter strategy than throwing everything out every product cycle or two.
Kepler is a shining example of an architecture touted as the next best things since sliced bread, it’s power efficiency was marketed as a feature versus the more capable GCN HD7000 cards back in the day, but nVidia hid the fact that it couldn't do compute to for shit!
How exactly did NVIDIA hide this? The architecture details were pretty public.
When you can twist the tech media into buying into your hype and have an advantage at the given time in a certain APi (DX11) the masses will flock to your product, the 700$ tag for a 3GB card is/was a rip off of massive proportions, especially when the 290x with 4GB was around 10-15% slower in nVidia favored titles at worst!
At the time, the 780/780 Ti did well in the games that people played. What makes you think that hype was bought?
A deficit it has reclaimed over the years and is still a viable gaming card, all while the 780Ti now has to compromise on the most basic of things like texture resolution for it not to choke.
This fate will befall every graphics card eventually. VRAM requirements rise over time.
What a perfect example of hype over substance. The Maxwell cards are next, they have no muscles to Flex in DX12 and see relatively little gain in Vulkan compared to modern GCN cards.
I see good boosts for 1080 here in Vulkan over OpenGL 4.5. Pascal looks like it's doing OK here to me, which is even more impressive considering how good NVIDIA's OpenGL/DX11 drivers are.
And Pascal will suffer the same fate, it still doesn't see the same gains in DX12 compared to GCN, high core clock can only carry you so far.
Work done per cycle * core clock is the performance equation in a basic sense. There is no shame in getting the performance that you're aiming for by improving the second variable in this equation as long as you can do it efficiently.
And what do you think nVidia’s pitch is going to be for Volta? Full DX12 capabilities, hardware async? The more I look at it the more I see nVidia being the Apple of the GPU world, it looks good when it’s new and hot, but man does it lose value over time, it’s shiny but it ain’t worth the cost relative to the competition. This is why I will never buy another x80 card from nVIDIA ever again.
Apple products tend to hold their value very well (great resale value), and Apple offers OS updates on its older phones, in very stark contrast to practically every other phone maker. I actually think your analogy is correct, but it doesn't negatively reflect on NVIDIA, IMO.
Anyway, I see from your sig that you have a GTX 1070. If you feel so strongly that NVIDIA isn't building products to last, why didn't you choose a product from an alternative vendor?