AMD has recently announced a technology called High Bandwidth Memory(HBM), and makes bold claims about how this new technology will address the major problems with existing computer memory.
AMD lays out the issues with GDDR5, which is the current industry standard for graphics memory. As GPUs increase in speed, they require more bandwidth to move data to and from memory, otherwise that extra speed is wasted. However, increasing the bandwidth of GDDR5 requires more power. According to estimates by AMD, GDDR5 will soon reach a point where it will no longer be feasible to increase the bandwidth, because the energy required will be too high. The other main weakness with GDDR5 is that each increase in bandwidth will require an additional GDDR5 chip, which will use up more space.
AMD’s HBM technology aims to take care of both of these problems. HBM vertically stacks memory chips on top of each other, which increases the bandwidth, but uses up less surface area. These stacks of memory as well as the GPU or CPU will be stacked on top of a layer known as the interposer, which will allow high bandwidth communication between the processor and memory. Although the memory stacks are not integrated into the GPU/CPU, AMD claims that speed provided by the interposer will make this system “nearly indistinguishable from on-chip integrated RAM.”
With this new technology, AMD promises to take up 94% less surface area than GDDR5, for the same bandwidth. It also claims to use 3 times the bandwidth per watt of power. If these claims are true, this technology could be a game changer. HBM might become the new standard for memory. While this looks promising, it remains to be seen if AMD’s claims will actually hold up.
Do you think HBM will be a game changer for graphics memory? Leave your comments below.