AMD's HSA - A Threat to Intel's Dominance and Mobile Ambitions?
The Open Heterogeneous Systems Architecture (HSA)
A major feature which is just as important as processing power in this age of mobile computing, is the power management capabilities of AMD's new APUs. The company just launched the true second generation APUs codenamed Triniy and as we found, was a really interesting chip bringing up the rear-end of what used to be sub-standard graphics capabilities built within the main processor. AMD claims that with the new chips, Heterogeneous System Architecture (HSA, rebranded from Fusion System Architecture or FSA), can run much more efficiently than ever before. HSA is essentially an enabler that allows multiple processing units (or accelerators) to work in tandem with shared system resources. In the case of the APU, it the CPU and GPU work in tandem with shared system memory, and the key advantage here is lower power consumption and increased compute power when running modern applications that are designed to leverage on the advantages of both the CPU and GPU. Applications that deal with analytics, search and facial recognition are some aspects that stand to gain from this boost.
To ensure that AMD gets into a better position when dealing with Intel, AMD is pushing for HSA to become an open standard. Right now, the first steps towards this ideal is taken up by AMD to try and win over proprietary systems, enable larger markets and ecosystems, when they form a consortium to work around the HSA. The primary target is to have HSA be a model that all operating systems and applications support. Right now though, the specifications recommended by AMD are being reviewed by their technology partners, and the outcome will be made public at the AMD Fusion Developer Summit 2012 in June.
There are a couple of upfront benefits that come with using the HSA model. Starting with last year’s Llanos platform that brought unified power efficiency, this year’s Trinity platform boasts improved compute power utilization, such as enabling the GPU to access CPU memory without having to pipe out to a dedicated buffer. In the next two years, the newer platforms will introduce simplified data sharing, and quality of service where the CPU and GPU usage would be completely transparent to end users and software developers - at least that's the goal.
How Would it Pan Out?
On the Trinity platform, as the first APU with HSA-MMU equipped hardware, AMD claims that tasks like emulation, automation, diagnostics and validation will be done more efficiently than last year’s Llano platform. According to AMD’s own internal roadmap, the next platform (codenamed Kaveri) will be able to boost overall compute performance by another 15%. In order to do that, each new platform that AMD introduces will be have to use the more agile System on a Chip (SoC) methodology, while focusing on power-performance optimized cores. Along with their HSA value proposition, AMD also hopes that further down the road, their new technology will also be adopted by industry partners, as well as have a more significant presence in data centers.
But what does it mean for end-users? It's probably too early to say now, but come June, when AMD will formally announce the partners who would have signed up for the HSA consortium, we may then have a clearer idea on how this will progress. However from what we can glean from the brief mention of the HSA consortium during the Trinity event in April, AMD's sights could very well be to create an entire eco-system based on their hardware. As you can see from what patents AMD will bring to the table, the kind of members that may join the HSA party could be hardware manufacturers and software developers that will further bolster the already extensive patent portfolio.
Consumers may eventually have to choose between eco-systems, just like the currently ongoing Android, iOS battle. However, should the momentum for HSA be strong enough to convince major players, that should help steer the direction for the rest of the industry. Ultimately, joining the consortium is one step, but whether they can effectively come up with future solutions as a whole remains to be seen. AMD supports several open standards (though they may not have helped form them), but initiatives like their AMD HD3D technology supporting open standards has been rather weak in adoption and proper 3D gaming solutions though there are vendors supporting it from the hardware and software stack. It's all about who's able to drive the initiative and invest resources to get things moving. But that's a concern for the next stage after the consortium is formed.
Opening Up a New Attack Strategy
By forming a system architecture consortium filled with members like the British mobile chip maker ARM, and other hardware/software manufacturers and developers, AMD will be able to extend their reach even further, targeting more customers. Imagine a scenario whereby the new APUs down the road are able to co-work with other third-party processing units integrated within the processor, but the design itself is modular such that certain APUs can opt to have these extra processing blocks or not.
Does that mean an ARM processing block might be get integrated replacing AMD's own CPU processing core? That sounds unlikely seeing who's driving the initiative at the moment and of course, who's got the wafer manufacturing and engineering expertise. Still, it doesn't rule out the fact that an existing AMD APU might have a supplementary ARM processing block. Further to that, their GPUs can definitely stay because they're pretty awesome. In the end, how it will work out for AMD is anyone's guess at the moment. However, when you have a standard architecture where members of your consortium are big technology players, it would also mean that the sum-of-all-parts would stack up against Intel - and much more convincingly than if AMD were at it alone.
It's not just the hardware vendors who're coming in as partners because HSA would allow software developers (perhaps like Microsoft, Adobe and others) to simply tailor and target their software around a standard hardware architecture, which would then (hopefully) work on multiple devices and platforms, thus boosting the HSA ecosystem uptake and progression. If you try to analyze this further, there may be some indications that the HSA consortium might be AMD's ticket and the kicker to get back into the mobile device ecosystem and spear a new strategy with the help of its supporting partners.
After all, the handheld mobile device market is shaping up rather well with rapid development and always pushing the boundaries of its capabilities. With every major tech vendor like Facebook, Google, Microsoft, ARM and Intel fighting for a slice of the action, it's sad that AMD as a major enabler in the desktop, notebook and visualization departments, doesn't have any real involvement in the mobile market (yet). If AMD were to stay relevant in the future, it has to gain a foothold in the mobile market too and we're glad to mention that they are already working behind the scenes to penetrate the tablet market with their low-powered Fusion processors. Next stop, would be smartphones and we can't wait to see when and if they make this transition.
For now, we can only sit back and try to make as much sense as we can from AMD's rather veiled HSA consortium momentum and we'll have to wait till June's AMD Fusion Developer Summit reveals more plans for the coming months or years ahead - if it takes off as planned.