SK Hynix’s 460GB/s HBM2E memory has entered Mass Production – Ultra-fast HBM2

Last year, SK Hynix revealed high-speed HBM2, what SK Hynix calls HBM2E, memory modules, offering customers up to 16GB of memory with speeds of up to 460GB/s per memory stack. 

For context, Nvidia’s RTX 2080 ships with 8GB of GDDR6 memory and delivers 448GB/s of memory bandwidth. Yes, that means that SK Hynix can offer two times as much memory capacity and more memory performance on a single HBM2E memory chip. AMD’s RX Vega 56 also offered 8GB of HBM2 memory over two memory stacks to deliver 410GB/s of memory bandwidth. 

SK Hynix’s HBM2E memory if incredibly fast, and its density will allow manufacturers to develop some incredibly fast hardware, and while HBM2 has fallen out of favour in the mainstream consumer market, it remains highly popular within the data center market, where the latency, bandwidth and space-saving characteristics of HBM2 memory make it ideal for high-end accelerators, AI compute modules and FPGAs.  

These memory stacks are created using up to eight stacked 16Gb chips using Through Silicon Via (TSV) technology, more than doubling the capacity of SK Hynix’s older HBM2 memory modules. 

Right now, the most bandwidth rich consumer-grade graphics card to use HBM memory is AMD’s Radeon VII, which offers 1024GB/s of total memory bandwidth. Using SK Hynix’ new HBM2E memory, a Radeon VII like graphics card would offer 1840GB/s of bandwidth, delivering a performance boost of almost 80%. With 16GB HBM2E modules, AMD’s Radeon VII graphics card would also feature a 4x increase in VRAM capacity, giving the card 64GB of VRAM instead of its usual 16GB.