VGA technology until 2014 is still scared or constrained because of memory in VGA.
Now no longer, because of a new discovery with HBM or High Bandwidth Memory.
Memory changes never happen before where VGA uses slow DDR memory and is replaced with GDDR.
Now GDDR is obsolete alias is not sufficient memory bandwidth requirements and display technology such as 4K monitor resolution. Why GDDR5 is an obstacle in VGA technology.
GDDR has been used for almost a decade. But it makes a problem for the modern GPU. The size of a large alias can not be minimized, some devices require higher memory transfers and can not be supported by GDDR, for the size of the board come into large because of the need for memory space, use of GDDR power requires a voltage regulator that adds VGA board size.
read also AMD innovation in CPU
HBM technology or High Bandwidth Memory
AMD introduced the first HBM. The difference between HBM is lower power consumption with Ultra-wide bus transfer capability, some chips can be stacked like building high floor, interconnect using TSV and uBUMBP to connect DRAM chip to other chip, TSV and uBUMP can be used to connect GPU chip, so far this AMD and Hynix work together to develop HBM.
HBMs that appear will use 4 stacks of DRAM, stacked on the bottom up, each DRAM contains 2 gigabit memory. When installed 4 can reach 1GB. The first generation HBM uses a 4GB DRAM stack system. Each stack uses a 1025 bit channel interface and works at 500Mhz. Using 4 Way system, 4GB HBM speed will be equal 2-3x faster than GDDR5.
GDDR5 is only able to mentrafer up to 10.66GB / s while the speed of HBM reaches 35GB / s more
Power consumption on HBM memory
HBM memory usage on power will be lower, when calculated per watt can save 15-20% power requirement. For example one VGA requires power in the range 250-300W TDP, then the need for VGA power can be suppressed between 37-60W.
HBM design as shown below. The GDDR5 VGA dimension requires an area of 672 mm square, with only 35mm square HBM.
The total area of the GDDR5 VGA PCB board requires a 9900mm square field like the Radeon R9 290X GPU + RAM design. For VGA with HBM memory requires only area of half or below 4900mm square. This means the size of VGA with HBM memory can be made half the size of VGA generation GDDR
HBM3 August 2016
Hynix and Samsung will offer HBM memory up to 64GB VRAM. Faster than HBM2, but cheaper.
Hynix calls HBMx while Samsung uses the name xHBM or Extreme HBM.
Features speed from 2Gbps HBM2 to 3Gbps in HBM3. The speed of each chip reaches 256GB / s stacked at a top speed of about 200GB / s per stack.
This means that HBM memory can be mass produced and used for graphics devices. Suppose notebook gracik and VGA PC.
HBM2 technology is already on the Nvidia Tesla P100 chart, but not for the general consumer like gamers.
HBM3 + and HBM4 March 2018
HBM memory technology still continues. Between 2022 and 24. Memory will be stacked more and more capacity.
Increase bandwidth at the same time.
HBM3 + is estimated to reach 3TB / s speed and 384GB persocket capacity.
When compared to GDDR, the speed of HBM is not a match for the old technology.