SK Hynix has just released its Q2 2017 product databook for current and next generation graphics memory products and it includes two very interesting entries. Hynix now officially lists 12Gbps and 14Gbps GDDR6 memory as well as 1.6Gbps HBM2 memory in its products log.
GDDR6 memory will power NVIDIA's next generation GeForce graphics cards based on the Volta micro-architecture. Whilst HBM2 will power AMD's upcoming Radeon graphics cards based on the Vega micro-architecture.
Vega Due Within The Next 6 Weeks - New Radeons To Brandish second Generation High Bandwidth Memory
Vega will be the first graphics architecture to make use of second generation High Bandwidth Memory from SK Hynix. Both of NVIDIA's Telsa P100 and Volta V100 leverage Samsung HBM2 memory, rather than Hynix. So the new HBM2 product listing in Hynix's datebook bears no relation whatsoever to NVIDIA's upcoming Volta V100 which is due in Q3.
The product spec for Hynix's HBM2 also makes this abundantly clear. It should be noted that HBM2 was originally intended to run at 2.0Gbps but all versions of the technology we've seen to date are rated at significantly lower speeds than that. HBM2 in the Tesla P100 run at 1.4Gbps and HBM2 in the upcoming V100 are rated at 1.76Gbps.
Funnily enough, HBM2 was also originally announced by SK Hynix in Q3 2016 and a 2Gbps HBM2 product was listed in the company's official product log. 2Gbps HBM2 has since been removed from Hynix's product databook in 2017 and replaced with 1.6Gbps HBM2.
After three full quarters of delays Hynix has yet to designate HBM2 as an "available" product and expects to finally launch it some time this quarter. This could explain why Vega has yet to launch. It's possible that following such a long delay AMD decided to go ahead with the Q2 2017 launch regardless and just fight through the early HBM2 shortages to deliver Vega its fans. The Vega 10 GPU that AMD showcased several months back had two HBM2 stacks. In this configuration paired with 1.6Gbps HBM2 memory it would have 409.6GB/s of bandwidth, 60% more than the RX 480.
Volta Due In Q1 2018 - New GeForces To Be The First With GDDR6 Memory
Last month we saw a very peculiar press release from SK Hynix announcing that it will begin shipping GDDR6 memory for an upcoming graphics product in Q 2018. The company then mysteriously showed up at NVIDIA's annual GTC 2017 event to showcase early production wafers of its new GDDR6 memory chips. The company lists GDDR6 memory as a Q4 2017 product. Which means that it should begin mass production in Q4 in preparation for NVIDIA's Volta launch in Q1 2018.
The company will offer two flavors of GDDR6, a fast version and an even faster version. The fastest 14Gbps memory chips will enable graphics cards with a 256-bit memory interface, i.e. a Volta based "GTX 2080/1180" for example to deliver 448 GB/s of memory bandwidth, which is just 36GB/s shy of NVIDIA's GTX 1080 Ti.
| Wccftech | GeForce 10 Series | Tesla V100 | GeForce 20/11 Series | AMD 400 Series | AMD 500 Series | AMD 600 Series |
|---|---|---|---|---|---|---|
| Architecture | Pascal | Volta | Turing | Polaris | Polaris / Vega | Navi |
| Process Node | TSMC 16nm FF | TSMC 12nm FF | TSMC 12nm FF | Samsung 14nm LPP | Samsung 14nm LPP | TSMC 7nm FinFET |
| Memory | GDDR5/X | HBM2 | GDDR6 | GDDR5 | GDDR5 / HBM2 | GDDR6/HBM2 |
| Year | 2016 | 2017 | 2018 | 2016 | 2017 | 2019 |
| GPUs | GP102, GP104, GP106, GP107 | GV110 | TU102, TU104, TU106, TU107 | Polaris 10, Polaris 11 | Polaris 20, Polaris 21, Vega 10, Vega 11 | Navi 10, Navi 11 (Navi Chiplet?) |
