About a decade ago I was building GPU based grid computing for data centers. At the time NVIDIA had very little of the data center market. It was owned by Intel.
My group was using NVIDIA GPUs to speed up risk calculations for a large bank. It was about 5x cheaper than doing the same job with Intel CPUs. There was a risk, going with a small company like NVIDIA, but the savings were too good to ignore.
Intel made some moves in this direction, notably the Xeon Phi, but it was a “me too” solution, not much better or worse than NVIDIA. And since it was displacing 5x worth of Intel hardware, Intel had very little enthusiasm for the Phi.
Intel also had its own graphics accelerations (GPUs). Depending on how you counted it, Intel owned something like 99% of the GPU market. But these were on-chip GPUs, mostly in laptops.
There was an Intel GPU in a single datacenter part (this is all from memory) but it wasn’t exposed to programmers. It was only for Microsoft VDI remote windows desktops. I tried several times to get access from Intel but was rebuffed.
Long story short, Intel protected its lucrative data center CPUs, particularly from internal competition, and eventually lost to the external competition, NVIDIA. The old lesson: if you don’t cannibalize your own market, someone else will.
Steven A. Guccione 