⋮    ⋮  

Samsung Enters The HBM Market In 1H 2016 – HPC and GPU Ready HBM With Up to 1.5 TB/s Bandwidth and 48 GB VRAM


At IDF15, Samsung announced that they will commence mass production of HBM stacked memory starting early 2016. High Bandwidth Memory or HBM in short, is the name of the latest memory chips that will be integrated in a wide array of products and devices. Co-developed by SK Hynix and AMD, the HBM chips were first fused on the Radeon R9 Fury X graphics card from AMD that launched in June 2015.

Image Credits: Computerbase

Samsung To Commence Mass Production of HBM Memory in Early 2016

The Radeon R9 Fury X debuted with 4 GB of HBM memory which was arranged in layers of four die stacks, each of which was 4-stacks high and each die had 1 GB of VRAM. Each HBM stack had a 1024-bit wide bus while the whole thing had a large 4096-bit wide bus that was clocked at 500 MHz (1.0 GHz effective) clock to deliver 512 GB/s memory bandwidth. The bandwidth improvement for HBM is quite huge even though the difference in performance between GDDR5 and HBM (at the moment) is quite low since high-end graphics cards have large pools of bandwidth already available to them that isn't being utilized to full extent. We saw some benchmarks of the Radeon R9 Fury X with its HBM overclocked to pump out 1 TB/s bandwidth but the overclocker reported not much benefit from the HBM overclock, the graphics chip on the other hand when overclocked featured a greater impact on performance. HPC applications that use HBM will show the actual advantage when several Terabytes of bandwidth gets effectively utilized to power intense workloads and graphics demand.

With Samsung entering the HBM field, we can see a large influx of new HBM powered processors and devices. Samsung is going all out to bring several tiers of HBM stacked memory to the graphics can HPC market. Starting in 2016, the first markets that Samsung will focus towards will include the HPC and graphics department. Samsung has a wide array of HBM configurations pre-planned. Each HBM stack will be made from a single 8Gb component and range down to several tiers of HBM SKUs. The entry level models include the 2-Hi DRAM model that will be integrated on mainstream 2 GB HBM graphics cards (256 GB/s), performance based graphics with 4 GB HBM (512 GB/s). The Enthusiast graphics cards will ship with 4-Hi DRAM with 2 HBM stacks that will allow 8 GB VRAM (512 GB/s) and finally, 4 HBM Stacks with 16 GB VRAM models (1 TB/s).

On the HPC front, there are a wide array of high bandwidth and dense memory designs that include 4-Hi DRAMs with 4 HBM stacks that feature 32 GB VRAM (1 TB/s) and the bulky, 8-Hi DRAMs configured in 6 HBM stacks with 24 GB and 48 GB VRAM, both models featuring 1.5 TB/s bandwidth. There are also some network oriented HBM SKUs which are planned for launch in 2017 with 8-Hi DRAM Stacks configured in 1-2 HBM chips. In 2018, Samsung wants to focus on increase market growth by entering new applications to incorporate their HBM designs.

For all good, HBM seems to be the way forward and the mainstream applications of HBM memory sound great for mainstream and performance oriented graphics cards that are going to arrive in 2016. We know that NVIDIA has a full fledged FinFET based Pascal lineup coming out in 2016 which is rumored to feature 32 GB of HBM and 17 Billion transistors while AMD has their Arctic Islands planned for launch the same year that will finally bring a completely new GCN architecture based graphics lineup. It was previously rumored that SK Hynix might give AMD priority on access to HBM for their next graphics lineup so Samsung entering the HBM market will bode quite well for NVIDIA. Overall, this sounds great and we can't wait to see Samsung kick start their production of HBM memory.