AMD next-generation flagship graphics card will use SK Hynix HBM (High bandwidth Memory) technology, AMD Radeon R9 390X will be equipped with 8GB HBM memory which will be paired with the equivalent bit wide 4096bit, bandwidth can reach TB / s level. NVIDIA on yesterday also released, the next generation of GPU roadmap for 2016 and they announced that the NVIDIA Pascal architecture GPU will use 3D memory which is the HBM 3D memory technology, so we can say that AMD will be the first to use first Generation of HBM1 and NVIDA will directly jump to the HBM2. You can estimate that the bandwidth capacity of HBM2 will be doubled, and this is one of the reasons that NVIDIA Pascal architecture can achieve 32GB of video memory.
SK Hynix showcase HBM1 and HBM2
HBM is a joint development of AMD and SK Hynix, the full name is high bandwidth memory, which is one of the new standard memory, according to analyst it would be 9X faster than current GDDR5. Now it seems NVIDIA also participates, Pascal architecture uses a 3D memory manufactured by Micron, IBM led the HMC camp, but it is actually based on Hynix’s HBM technology.
The current HBM can stack four-layer chip (4-Hi stacks), each bit wide 1024bit, single chip capacity 1-2GB, AMD’s R9 390X had said that four layers 1GB HBM chip stack, so the capacity would be 8GB, it should be 2GB chip stack of four layers. In addition, the real 8-layer stack (8-Hi) is also in development, after the implementation the capacity can be doubled again.
HBM first generation memory for now will be used by AMD, SK Hynix also prepared HBM 2 second-generation technology, Width or 1024bit, but I / O prefetch width double bandwidth upgrade from 128GB / s to 256GB / s capacity doubled both bits wide.
If you remember yesterday’s news of Pascal architecture, which mentions Pascal architecture memory capacity is 2.7 times faster than current Maxwell architecture, current the GTX Titan X based on GM200 is featuring 12Gb of VRAM, so you can expect 32Gb of memory limit in NVIDIA pascal which rely on the memory of the HBM 2. Next-generation architecture would be the Volta in which memory capacity is even doubled to 64GB, it would have to rely on a stack of 8 layers (Cool!!). NVIDIA Volta architecture will make its entry in 2018.
Like with the previous GDDR5 technology, we have to admit it that AMD once again leads when you compare both technically. Prior to AMD in FreeSync dispute with the G-Sycn technology also dedicated to the author spoke of their openness to technology, suggesting that NVIDIA likes to eat food alone, the AMD earliest development of GDDR5 memory (HD 4870 was first to use it), but we have not seen it as their own standards, but they open it to the industry. This is the first time that AMD will implement HBM, I do not still we see the war of words or not.