Didn't want to derail the ASUS Strix 1070 ALL Using Micron Memory ICs?
Yes, there's apparently a "shortage" of GDDR5 memory, so most of the GPU card makers have (quietly) ...
The conspiracy theorists are quick to point at past memory price-fixing cartels and "antitrust" practices. The market analysts believe Samsung/Hynix silicon fabs are just focussing on higher (more profitable) priorities. The overclockers are unhappy because Micron stuff doesn't seem to perform as well overspec as Samsung/Hynix counterparts.
EVGA, MSI, Palit, and Gainward have already released revised GTX1070 VBIOS firmwares which promise to "fix/correct issues" on overclocked Micron-based GDDR5.
Micron enjoys a very special (ASUS-like) working partnership with Intel. So Micron is currently the only mass-producer of JEDEC GDDR5X
(which NVIDIA calls "G5X"), Micron has a decisive lead in Hybrid Memory Cube (HMC)
mass-production by ("experts" say) two full generations, and Micron now mass-produces extreme performance hybrid nonvolatile memories (which Micron calls "QuantX", Intel calls "Optane", and NVIDIA calls "3D Xpoint"). Strangely, Micron does not mass-produce any "conventional" or "complementary" High Bandwidth Memory (HBM)
Samsung, SK Hynix, Micron (and other partners) are all active members of the Hybrid Memory Cube Consortium
, but only Micron (and perhaps Altera) enjoys many hidden advantages from working directly and intimately with Intel.
AMD currently uses GDDR5 in their top (RX460/470/480) GPU cards, having (again) broken their promise to use GDDR5X with Polaris
. The new RX490 looks like it'll be Polaris (or dual-Polaris) but at least we can hope perhaps AMD will strive to actually use GDDR5X in their upcoming Vega/RX5xx GPUs.
Let's face some hard and ugly facts: AMD's main focus is (economical) TSMC 28nm fab and they've just barely dipped into (costly) Samsung 14nm fab; AMD's GCN4 (and GCN5) architectures can hardly use extreme-bandwidth memory efficiently anyhow; and it seems doubtful AMD is going to spend more money per-card buying faster/denser GDDR5X hardware than buying faster/better GPU ASIC hardware (AMD's focus has always been "bang for the buck", and they can get more raw fps spending on better GPU than spending on better VRAM). Plus, AMD has already shifted to Samsung ASIC fab, it makes sense for them to shift to Samsung memory fab, and Samsung doesn't make GDDR5X.
NVIDIA currently uses "G5X" in their top (GTX1080/1080Ti/TitanX) GPU cards. NVIDIA keeps "leaking" hype about upcoming GeForce 20 Series Pascal GPUs (GTX2070/2080/2080Ti)
using "G5X" in 2017 then GDDR6/HBM2 in 2018.
None of this help all you folks with "sucky" Micron-stuffed GTX1070 cards, lol. That's what the other thread is about. (My advice is to buy a GTX1080 or other newer/better card if it's released before your GTX1070 manufacturer releases a VBIOS "fix".)
Just saying that Micron, Samsung, and Hynix have each leapfrogged to the top before but in my opinion it looks like Micron is climbing fast into top position and will be hard to knock down once (NVIDIA's) extreme-tier 2017 and 2018 GPUs hit market.
"All opinions are not equal. Some are a very great deal more robust, sophisticated and well supported in logic and argument than others." - Douglas Adams