You saying mask cost is 5-10M$? How many masks needs to be done before you even got a final chip? And what about IC design cost that goes up with a factor of 2 every (half)node? R&D? See the point? I would bet you AMD for example will never make anything but loss on their Fiji dies due to volume. And thats even when they reuse GCN 1.2.
Have you verified, what the IC design costs consist of? Is it the physical layout or functional stuff too? And how much was mitigated by improved design automation? If design costs go up due to mask costs (and w/o EUV they take bigger and bigger chunks of the total costs), that still doesn't say anything about the costs of improving the logic on architectural, RTL, or lower levels.
Please link your 600mm Pascal GPU.
That's a number I picked up in some discussion. When I was younger I'd measured the die myself. Now I let others prove, that this die shown in your linked photo isn't big:
http://forums.anandtech.com/showpost.php?p=37459726&postcount=16
Rumoured numbers like 17B transistors still have to be confirmed. But I think it's clear, that this is no cheap design and it won't be sold to the masses either.
The problem isnt whats possible. The problem is whats financially viable. And suddenly you sit with IGPs and no dGPU for the same reason.
You are not one of those "That [CPU|GPU] performance is enough for my surfing." proponents, are you? And an IGP still causes a lot of design costs. And those bigger dGPUs usually don't have many different units as those used in IGPs. Shader blocks are the same, HW codecs, display gens, caches, busses.. most of that is the same or very similar in one family.
So we are discussing the
cost delta of making bigger dies, scaling up the memory subsystem, adding some special units maybe, and the full feature shaders capable of 1:2 DP:SP throughput, PCB design, etc. Like $300M for the iGPU and $350M for the dGPU design. Questioning the full costs of only one of two costly options is a typical human behaviour. And margins are probably higher for dGPUs.
People don't stop buying >150 HP cars to save the climate, so why should they step back from their ever growing expectations in visual experiences?
So you believe it failed because the resolution wasn't high enough?
Maybe you're right, but I kind of doubt it. I think it failed because only a fraction of the population is willing to wear something like that on their face while they wave their hands around and kick their feet in the living room. A problem which hasn't gone away btw.
Probably yes. In some vehicle test system we had older and expensive VR/AR glasses. Even with fast head tracking and a real car under your seat this looked like an 80's, early 90's game. No true immersion then. But the brains' 3D recognition accepted that world after a few minutes.
The new VR headsets are improved and the Rift didn't feel bad. As long as it's not too heavy the human body quickly adapts to any new, but static situation, as you know. Only my IPD didn't match that well with the fixed lens positions.
For walking and interacting, some new concepts are on their way, like the walking caves/boards and special controlers.