either trollin' or has a wildly different understanding of the word cheap...
He's right as far as the costs on Intels side are concerned. Crystallwell is 80% of the die size of Bay Trail -> lowest Bay Trail list price $37 -> Crystalwell adds some $30 actual list price if calculated with Bay Trail margins. OEM costs will be lower.
However, Intel can't currently price Crystalwell like that, because they have limited 22nm capacity and Crystalwell competes with production of all leading-edge CPU products. So they have to keep demand low by keeping prices high. Once 14nm productions gets going Intel should have more capacity for Crystalwell production, so Broadwell may have more Iris Pro SKUs. List prices will probably remain high, so that Intel can control demand by offering attractive pricing only to selected OEMs.
Once Intel reaches the massive capacity expansion of 450mm wafers a solution like Crystalwell could go mainstream, but that won't happen prior to 2016. I'd expect that AMD will have some form of embedded RAM for their APUs by then, so Intels advantage will be less or may be negated completely. Comes down to price/perf/W for CPU/GPU, too many variables to judge the outcome for that far into the future.
the only brilliant thing about iris is that it took them no time at all to develop a counter to amd's gpu prowess...
As Crystalwell is a full CPU/GPU L4 cache, they must have planned uncore integration pretty much from the beginning. Crystalwell seems to use part of the L3 cache as tag array for the L4: even the highest-end Iris Pro SKUs have only 6MiB, in contrast to the 8MiB of the standard i7s. The cache and uncore need to be designed to support that. And they must have had the technology under development and testing for some time before feeling ready to integrate that into shipping product and be certain that the performance increases will be worth the effort. So Crystalwell must have been under development for at least four years or so. This isn't a stop-gap measure, but long-term product strategy.
If you want to paint the technology in a bad light, I'd recommend the occasional performance degradation in CPU workloads. In the
AnandTech review, x264 2nd pass showed a performance regression of 4.5%. But be sure not to mention that CB 11.5 was up 9.5%, 7Zip +5.1% and x264 1st pass +5.9%.