The link you provided shows that the Titan X merely holding it own against gpus $400 cheaper. It's computer performance isn't that much better than a 980 Ti and it doesn't have as good FP processing as previous titans have had.
You still don't get it.
1. The Titan X's GPGPU selling point is the 12GB memory capacity. GPGPU workloads including visualization and rendering that require large onboard memory capacity - basically any rendering at 4K and up for example and large parallelized data set workloads. Name a single other card that offers at least 12GB of vRAM that retails for $1K.
2. When the Titan X launched there were no $400 cheaper cards offering the same performance. Everyone knew they would come, but the months when they ruled the roost were obviously worth it to a large percentage of buyers. If this were not the case, why would enthusiasts continue to do the same thing generation of cards after generation?
You know why people don't put 8-16 980Ti in a server rack for GPGPU for the same workloads? Because the vRAM cannot hold the data set and you would have an instance of 8-16 cards sending huge quantities of overflow data across the PCI-E bus and rely on system memory segmenting the individual workloads and constantly moving them backwards and forwards across the bus. It doesn't work. It introduces stalls and latency and broken animation and TDR issues.
I could post PDF's on the subject but I suspect you'd just ignore them, just as you seem to be ignoring the fact that vRAM capacity is in an arms race for the exact same reasons I've outlined. It is the reason that the M6000 is now available as a 24GB card as well as a 12GB card (to be able to undertake 5K-8K rendering), and why the AMD increased the
W9100 to 32GB
Good for them but purely looking at fact the Titan X doesn't provide a good performance / cost ratio except for a very very small segment of the already small ultra-enthusiast class.
Newsflash! No $1000 card has ever provided good performance/cost ratio, but some people could give a damn about perf/$ - these are the buyers. Some people need the vRAM capacity for GPGPU workloads - these are the buyers.
What I find laughable is that you are trying to prove that no market exists for these cards
DESPITE the fact that they have sold well, are popular with enthusiasts/benchmarkers and are prevalent in GPGPU systems - according to you, everyone who doesn't share your view lacks common sense - despite the fact that you obviously know very little of the mindset of the actual users or the usage the cards are put to . You devote thousands of words to this subject, but when a prime example of a card from a rival vendor that has an even worse perf/$ ratio and will sell in minuscule numbers
you raise not a single word of criticism...now why would that be? At first I would have thought that it might be because its usage characteristics make it suitable for niche market workloads which mitigate its pricing.....but that is clearly wrong, because you believe no such argument is valid.