AMD’s Vega GPUs Inch Closer To Launch As SK Hynix Gets Ready To Ship HBM2
SK Hynix, the Korean memory and NAND flash company, is expected to start shipping its HBM2 memory during Q3 2016 (July-Sept). This move will finally usher in the era of AMD’s high end Vega GPUs as right now Nvidia’s Tesla P100 graphics accelerators are the only ones confirmed to use HBM2 although it is expected to release during Q1 2016.
The first generation of HBM had people excited but the end result didn’t live up to the expectations of most of them. It supported 2GB per DRAM die in a 4 Hi stack configuration with 1GB memory. HBM2 on the other hand looks to be changing the game in a drastic fashion.
The upcoming memory standard not only features 8GB per DRAM die but also double the speed per pin and bandwidth per stack compared to first generation HBM. SK Hynix will be shipping 4GB HBM2 stacks in 4 Hi-stacks (4 die per stack) form factor. The memory will also be available in two speeds, 2.0Gbps and 1.6Gbps. This should allow manufacturing of GPUs with up to 4096 bit HBM2 interface and a video memory of 16GB.
There have already been reports of AMD working on two new High end GPUs, codenamed Vega 10 and Vega 11. Both these graphics cards are expected to use the HBM2 memory in order to counter Nvidia’s GTX 1080’s GDDR5X.
Through Vega and HBM2, AMD seeks to deliver amazing next generation performance for people looking for 4K gaming and a massive leap over the recently released RX 480 which was targeted towards mainstream market.
Based on a fine tuned 14nm FinFET process, AMD Vega would deliver a significant Perf/Watt improvement over Polaris which itself offers 2.8x increase over older 28nm products. So we’re looking at up to 70% performance boost with Vega 10 over the R9 Fury X.
This however doesn’t mean that Vega will launch alongside HBM2 and still has a long way to go before it is ready for consumer use, as confirmed by AMD Graphics boss Raja Koduri. It will probably see the light of the day during Holidays 2016.