[EDIT]hmmm.....whats going on at AMD? (and history of 3Dfx Rampage)

[SteelCrysis]

https://www.extremetech.com/gaming/24961...later-year

For the past few months, I’ve been wondering if HBM2 has run into more problems than initially expected. When AMD introduced HBM, it seemed to be taking a leadership position on memory technology, much as it did with its adoption of GDDR4 and GDDR5 (Nvidia skipped GDDR4 and didn’t adopt GDDR5 until after AMD had done so). At the time, it seemed that AMD would be the only company to field HBM, but that Nvidia and AMD would both use HBM2 for their next generation of 14nm graphics cards. Instead, we saw Nvidia opt for GDDR5X, while AMD had no high-end solution at all.

Fast forward to 2017, and we’ve got AMD still stuck on GDDR5, GDDR6 waiting in the wings, and no sign of HBM2 on anyone’s consumer graphics cards. Nvidia has rolled HBM2 out on its highest-end Quadro line, but those GPUs still cost thousands of dollars more than your typical consumer parts. Nobody is pushing HBM2 to consumer hardware yet, and Raja’s specific willingness to call out HBM2 as the tech that’s been difficult to get into consumer parts as “not easy,” seems to highlight what the hold-up is.

Several years ago, it looked as if HBM and HBM2 would become the de facto replacement for high-end graphics cards. Lower-end cards would still use GDDR5, but even this would eventually become less common, as RAM loadouts rose above 4GB for lower-end cards. That forecast looks much less certain now. If AMD and Nvidia can’t get HBM2’s costs under control, it’ll remain a fringe technology, limited to the highest-end SKUs. Once GDDR6 is widely available, both firms might move away from it altogether.

Of course, it’s still possible that these are growing pains and that HBM2 will sort itself out, but the longer it takes to get the hardware into market, the less likely that looks. AMD is already working on follow-up designs to Vega and Ryzen, and acknowledge during its FAD this week that it’s simultaneously building the follow-up parts to those follow-up parts at the same time. HBM2 simply hasn’t emerged as the affordable next-generation technology we were hoping it would be, and I suspect — though again, this is a suspicion, not a claim — that HBM2 is the reason why Vega’s consumer launch is running late.*

* — AMD would undoubtedly quibble with whether Vega is late, since the chip will ship in a professional card within the first half of 2017, but let’s be realistic here. Consumer Vega hardware was expected in-market by Q2 2017, and AMD still isn’t saying when the consumer cards will launch.