The big interview: APUs, HSA and where next for AMD

Tom Morgan
23 May 2013

We speak to AMD’s Sasa Marinkovic about the company's role in progressing computer performance beyond the CPU, and what role HSA has to play

AMD is on the up. As the sole provider to the next generation of games consoles, AMD’s Accelerated Processing Units – otherwise known as APUs – are soon to be some of the most popular pieces of silicon across the globe, and that’s before the launch of its new Temash, Kabini and Richland chips, which the company hopes will win a major portion of the entry-level and mid-range laptop market.

But that's just the start – AMD is working on a new approach to computer architecture which it says will dramatically improve performance, create powerful new programming models for developers and open up more natural interfaces like speech and gesture recognition. It's called Heterogeneous System Architecture (HSA) and is built to merge the CPU and GPU workloads together, rather than treat them as separate processes. This shared design would be far more efficient at tasks that traditionally tax even the most powerful systems. We spoke to Sasa Marinkovic, AMD's head of Technology Marketing, to get the lowdown on HSA and what it could mean for the technology industry.

First off, can you give us a run-down of what makes HSA so important?

In a PC, the CPU is good for serial workloads and the GPU is good for parallel workloads. You need a good balance and have to put the right work load on the right piece of hardware. When we talk about HSA, you see that the memories are shared, so the argument won't be whether you're using the CPU or the GPU part of a chip, but what's the end experience. Applications written for HSA will use both, speeding them up compared to traditional CPU-driven workloads. The difference can be huge; what we're saying is that the CPU isn't the most important part of a system any more. If you want massive performance gains, you need to enable the GPU and use both.


What makes you so confident that HSA is the future?

You hear about (Intel's upcoming CPU) Haswell getting 10% CPU improvements over the previous generation. If you ask some of the developers enabling HSA in their applications, they are seeing five times the performance – that's 500%.

What is AMD doing for developers to help spread the HSA message?

A lot of people that don't understand coding won't appreciate how important HSA is, but it makes developers' lives so much easier. They are already signing up to the HSA Foundation (of which AMD is a founding member) because they want to work with HSA architecture. We're going to be the first to market with HSA hardware, so obviously a lot of developers are looking to get their hands on Kaveri (AMD's upcoming CPUs) in order to speed up development time. Some developers were working from Trinity APUs but have been given early access to Kaveri to show what HSA is capable of.


Is there a danger that most of the Compute applications currently available are focused on industry applications like 3D rendering and video encoding, rather than everyday programs we would use at home?

Internet explorer or Chrome are very much mainstream applications. Flash 11.1 is another. You see huge gains enabling GPU acceleration, even in web browsers like these. I think the reason there are more professional applications is because at this point you can only have guys who know how to program for the GPU. But that's the point of HSA, to bring it to the mainstream programmers so they can take advantage of that in a simple and easy way, optimising and programming with it for everything.

Do you think that customers are still too tied up with Gigahertz?

Just like when you review a car and see horsepower, or 0-60 times, on a chip you have gigahertz, number of cores, GPU cores and now Gigaflops with HSA hardware. People are still living in a world where Gigahertz is everything, as it's a number they can quantify, but I think the relevance of Gigahertz is much less now – it's not one-dimensional any more. Especially when you get to more compact form factors, like tablets, Gigahertz can become a burden as higher clock speeds will drain battery much more than multiple cores.

Historically, AMD has lagged behind Intel in terms of raw CPU performance. You're now saying benchmarks need to change to reflect how people are using their systems – is that a realistic expectation if the number of GPU-accelerated applications is still relatively small?

I think there are three kinds of benchmark, or ways to evaluate new products – the past, present and future. When you look at the past it was only about CPU or GPU performance. You drive the CPU to the max and make sure the GPU isn't the bottleneck, or you drive the GPU to the max and check whether the CPU is the bottleneck. When you look at single-threaded applications, that was how we tested things in the past.

The present is about being more visual – photo editing, video editing, gaming. Looking at popular form factors like ultra-thin laptops and tablets, what do people do with them? They watch YouTube, they chat to their friends. The CPU is doing a part of that work, but it's not the full story. Whether you look at desktops or laptops, 70% of the market doesn't have a discreet GPU – whether that's Intel or AMD.

The APU is two parts – the CPU and GPU. Both of them need to work together to give you the full picture. Obviously the CPU is still part of the story, but a lot of the stuff it's responsible for - watching TV online, playing games, and editing photos – can be unlocked when paired with the GPU. CPU-only applications aren't the model any more.

You can say Intel was good in the past, but I'm confident that now it's not going to make any difference to you as the user.

Read more