As per the source who earlier shared a picture of GA104 (RTX 3070) chips undergoing validation, the shortage of NVIDIA’s RTX 30 series GPUs is due to the lack of GDDR6X memory, rather than limited 8nm capacity from Samsung.
The source claims that Samsung has a testing capacity of tens of thousands of units per day. The actual fabrication, keep in mind, will be notably lower. This directs the attention towards Micron’s GDDR6X memory developed in partnership with NVIDIA. It’s basically the same as GDDR6, except using PAM4 signaling and a lower prefetch.
Unlike GDDR5X and GDDR6, GD6X isn’t a JEDEC standard, rather something NVIDIA and Micron have cooked up themselves. It’s unclear whether JEDEC will go with the same specs and or come up with different ones.
