Last year, Micron collaborated with NVIDIA and brought forward GDDR5X memory, with 2.5 GHz QDR clock i.e. “10Gbps per pin” achieving record bandwidth per pin. With GTX 1080 Ti and TITAN Xp, Micron and NVIDIA went 10% up and reached 11Gbps, with some lucky owners able to reach 3GHz QDR i.e. “12Gbps per pin”. That all is set to change with SK Hynix launching the GDDR6 memory.
Debuting on AMD (and NVIDIA) graphics cards in about six-to-eight months from now, GDDR6 replaces GDDR5 and GDDR5X, bringing great times for improvments for GPUs and FPGAs.
SK Hynix introduced the world’s fastest 8Gb (i.e. 1GB) Graphics DDR6 DRAM chip. Its operating speed (I/O data rate) is set at 16Gbps per pin, or a physical clock of 4GHz, which is almost twice as the GPUs that will drive the memory. GDDR6 is an upcoming JEDEC standard that runs “twice as fast” as GDDR5, while offering a 10% lower operation voltage. Doubling the bandwidth while reducing power consumption by 10% should keep improving the performance of graphics cards that don’t have the pricing structure to support the use of expensive and rarely available HBM2 memory. Best proof of how difficult is to get HBM2 memory is the fact that NVIDIA only ships HBM2 memory on its Tesla P100 and Quadro GP100, while AMD Vega suffered through delays and delays, even though it only uses two HBM2 chips instead of 4 (as seen on AMD Radeon R9 Fury & NVIDIA Tesla P100 / Quadro GP100).
Just so much that there is no confusion, GDDR is specialized DRAM for processing graphics data, also used for networking switches and digital currency mining. “With the introduction of this industry’s fastest GDDR6, SK Hynix will actively respond to high quality, high performance graphics memory solutions market.” said Senior Vice President Jonghoon Oh, the Head of SK Hynix DRAM Product Development Division.
“SK Hynix Inc. (or ‘the Company’, www.skhynix.com) today introduced the world’s fastest “2Znm” 8Gb(Gigabit) GDDR6(Graphics DDR6) DRAM. The product operates with an I/O data rate of 16Gbps(Gigabits per second) per pin, which is the industry’s fastest. With a forthcoming high-end graphics card of 384-bit I/Os, this DRAM processes up to 768GB(Gigabytes) of graphics data per second. SK Hynix has been planning to mass produce the product for a client to release high-end graphics card by early 2018 equipped with high performance GDDR6 DRAMs.”
Wait here. An official press releases mentions the upcoming next generation graphics card… from AMD? The release states that the products using the GDDR6 memory should come in early 2018 and their bandwidth is up to 768GB per second. Does this mean a successor to Polaris, i.e. Radeon RX 600 Series will feature a 384-bit memory interface? Oh yes it does.
Radeon RX 600 should feature 6GB of GDDR6 memory, with the professional Radeon Pro hitting 12GB of GDDR6 memory. Apparently one of design changes in the memory itself should enable easier (read: cheaper) integration of double the memory chips per memory interface.
We expect successor of Polaris to slot above Vega (8GB 2048-bit HBM2 memory interface, 512GB/s) and under Navi (Navi’s memory controller combines HBM2 and GDDR6 interface and is more CPU than a GPU, but more on that later). NVIDIA will also have its skin in the game with Volta using the same combo (HBM2 + GDDR6), dropping the HBM2 for more mainstream models. Overall, we can expect that the future will bring similar configurations to what Intel debuted with their Xeon Phi made of unobtanium (Knights Landing), which combines 16GB of
HBM HMC memory with a full blown six-channel, 384-bit memory interface supporting up to twelve DDR4 DIMMs.
For us mere mortals, we can expect GDDR6 memory to rule the roost, with HBM2 being offered only on “top-dog” products.