GDDR memory, which is basically double data rate memory for graphics, has generally been doubling its speed and throughput every generation and we currently stand at GDDR6 and GDDR6X levels while we await next-gen GDDR7.
In the meantime, AMD, alongside a few other industry giants, introduced the world to High Bandwidth Memory (HBM) back in 2015 with its Radeon R9 Fury (codenamed "Fiji") lineup. The benefit of stacked HBM over GDDR memory is the ability to provide much more bandwidth and throughput compared to the overall power drawn from the total memory footprint. In essence, HBM is much more efficient than GDDR while also taking up much less space.
However, Samsung is set to bridge that gap as the tech giant claims its next-gen GDDR6W memory technology can be nearly as fast as HBM while being smaller than traditional GDDR6. Using Fan-out wafer-level packaging (FOWLP), the packaging moves from the PCB onto the silicon wafer itself, similar to HBM. This helps in reducing the thickness of the packaging design by around 36% while accommodating twice the memory capacity as you see in the image below:
Aside from doubling the capacity, the bandwidth also increases by two times due to the increase in memory interface size. Samsung says that using a 512-bit bus and 22Gbps GDDR6W memory, a total throughput of 1.4TB/s is achieved, which is very close to 1.6TB/s bandwidth with a 3.2Gbps HBM2E memory.