Change or die. For AMD, 2016 marks a year filled with changes — the largest changes to hit the company since it launched 46 years ago. And if these changes aren’t successful, it will die.
The beginning of AMD’s problems started in 2006 when its Phenom II processors failed to keep pace with Intel’s latest and greatest Core 2 architecture. It continued sliding in market share until it released the first Accelerated Processing Unit (APU), which fused a graphics processor with a Central Processing Unit (CPU).
The APU overturned the video game console market in 2013, but after struggling with process node technology, AMD tumbled into near insolvency. Then in 2015, Intel caught up with AMD’s APU technology by delivering Skylake with robust integrated graphics. With Intel controlling markets, AMD hovered near bankruptcy.
Then Intel ran into the 14nm barrier. With Intel stalled and AMD rapidly introducing new technologies, the stage is now set for a serious reversal. AMD’s strategy into and beyond 2016 revolves around the convergence of four synergistic technologies:
- Polaris GPUs
- High Bandwidth Memory
- Zen APUs
- Dual Graphics
Polaris is AMD’s 4th generation Graphics Core Next (GCN) GPU architecture that brings a much-needed shrinkage to their process node technology.
Smaller transistors allow for increased sophistication, reduced power consumption, and improved performance — but instead of this being an incremental improvement, Polaris is a massive leap from 28nm to 14nm dies, which represents about two generations of growth in a single bound.
According to the latest claims, Polaris focuses on power-efficiency and memory bandwidth. This focus on memory bandwidth extends from the largest bottleneck in performance at higher screen resolutions: GDDR5 memory.
GDDR5 memory is starting to reach its limits, particularly in application to immersive virtual reality (VR). Truly immersive VR requires a tremendous amount of memory bandwidth — at least enough to render dual 4K displays and, ideally, 8K dual screens.
That’s why AMD is starting to release GPUs with a feature called High Bandwidth Memory.
High Bandwidth Memory
High Bandwidth Memory (HBM) is often referred to as “memory stacking”, or placing RAM on top of the CPU (or in AMD’s case, the APU).
The arrangement superficially resembles the way smartphone manufacturers stack RAM on top of CPU dies. HBM allows for decreased GPU size, improved heat distribution, decreased power consumption, lower latency, and — as the name suggests — substantially increased memory bandwidth.
Memory bandwidth right now is the single area in which graphics processors fall short. If HBM proves viable, it could make some emergent tech fields, such as immersive virtual reality, accessible to the average consumer (and VR is going to blow your mind).
But that’s not all. As mentioned before, HBM ties into another one of AMD’s core technologies — the APU — and there’s some good news on that front as well.
For those who aren’t familiar with the differences between CPUs, GPUs, and APUs, all you have to know is that an APU is what you get when you combine a CPU and a GPU together. And as a refresher, AMD was the leader of integrated graphics until Intel released Skylake in 2015.
AMD’s new Zen architecture — which is based on the previously-mentioned 14nm process node technology — purportedly offers a 40% improvement in one critical but difficult-to-measure metric: Instructions Per Clock (IPC).
Simply put, IPC is a measurement of processor performance, but it varies depending on the kind of tasks tested. Up until now, AMD processors have scored well for integer-intensive operations but fared poorly at floating point calculations, which most of our everyday computing tasks fall under.
So while AMD has traditionally offered higher CPU frequencies than Intel, its per-core performance has been around 40% worse than Intel. If AMD’s numbers prove accurate, however, the new Zen APUs should level the playing field.
Long story short, IPC will remain AMD’s competitive disadvantage going into 2016, but Zen’s integrated GPU is where they’ll find their competitive advantage. There’s been no official confirmation yet, but Zen’s APU design should also include the GPU architecture from the latest generation of AMD graphics processors, Polaris.
AMD released several presentation slides which show a development of Polaris, Zen, and HBM on a single die. A series of leaked slides from Fudzilla.com shows a monstrous 16-core Zen APU, containing 16 GB of HBM, and a Polaris integrated GPU. But what possible application could such a mammoth APU have?
The Race for Console VR
Both the Playstation 4 and Xbox One have one thing in common: AMD’s APU technology. While APUs don’t deliver the same raw power as discrete (“standalone”) GPUs, they’re cheaper and require less space — which makes them ideal for use in smaller video game consoles priced around $400.
Immersive Virtual Reality will be the most resource-intensive computing activity for consumers over the next few years. The recommended specifications for an Oculus Rift call for a system that costs at least $900, including a top-of-the-line GPU and mid-range CPU.
To compete with regular consoles, manufacturers need to somehow create a VR system within the $400 range. AMD seems to be converging several lines of technology in order to fill this void, and in the coming years, AMD will likely bring all three of these technologies into one product.
While AMD officially stated that its Zen architecture won’t initially use High Bandwidth Memory, it seems almost certain that video game consoles will be the first products to combine at least three (and hopefully a fourth) of AMD’s technologies into a single package for VR.
Sony and Microsoft are both developing VR technologies aimed at consumers. We don’t know when the next generation of consoles will launch, but there have been some hints that the technology could arrive within the next couple of years. In the race to release a VR system, these two companies may cut the current generation of consoles short.
The real advantage of APU, HBM, and Polaris isn’t just to make AMD competitive. It’s to make virtual reality technology affordable — and the company to do it first will win the next console war.
I can’t mention virtual reality without also giving lip service to AMD’s Dual Graphics technology.
Dual Graphics arrived in 2012, allowing integrated graphics processors to operate in tandem with discrete GPUs. While dual GPU systems go back as far as 2005, the combination of an integrated GPU with a discrete GPU remained difficult to implement because of the asymmetry between them.
In theory, Dual Graphics allows for smaller systems to utilize two different graphics processors, which makes it ideal for use on consoles. But because of various teething problems, the technology won’t make it past laptops or budget gaming systems in the immediate future.
Is AMD Poised For a Comeback?
After years of failure and decline, AMD finally stands a chance of a complete turn-around, owing mostly to its GPU and CPU architecture refinements and first-mover advantage in two key technologies: High Bandwidth Memory and Dual Graphics.
But more importantly, in the rush to push out consumer-ready VR systems, only AMD offers suitable hardware. In some sense, AMD may be our only hope for powerful-yet-affordable virtual reality.
Is AMD is heading in the right direction? Do you think this is the right path to take given the current landscape? Are you going to buy into virtual reality? Let us know in the comments!