3DVelocity logo


Content

Latest

News
Articles

Community

Forum


Main
AMD's work with the Heterogeneous Software Architecture (HSA) Foundation may well help developers make use of GPUs but the firm has to be careful not to find its products lost among other Foundation members' products.

The last 15 years haven't been kind to AMD and the firm but the firm desperately needs the goodwill of developers now more than ever if it is to survive for another decade. I recently wrote on The INQUIRER how important the HSA Foundation is to AMD's future, but here I will touch on a potential problem.

While the HSA Foundation is critical in building up software support, support that is crucial to AMD, it puts greater pressure on the firm's silicon engineers to deliver the goods.

A decade ago AMD was busy putting the finishing touches to what still remains its opus magnum, the AMD64 instruction set and in particular the SledgeHammer architecture, that made its début with the first Opteron processor. That time AMD was also part of a consortium, the HyperTransport consortium, that pushed the boundaries of high bandwidth inter-processor interconnect.  

Software cripples the superior architecture

As I insinuated in my INQUIRER article, back in 2003 AMD didn't do enough work with its software partners. I've met people that claim Microsoft's half-hearted support for 64-bit Windows in 2003 was due to its well known relationship with Intel, but in reality even application developers such as Adobe were slow to get on the 64-bit bandwagon even when Windows Vista 64-bit arrived.

Conspiracy theories, especially ones that involve sabotage, make for an entertaining read but stepping back and looking at the wider ecosystem it is clear that AMD's 64-bit architecture was ahead of its time for consumer computing. The primary advantage of AMD's then new architecture was the ability to address more than four gigabytes of RAM, which is very useful in servers and high performance computing but not particularly useful when the majority of consumer machines in 2003 were shipping with 2GB RAM from the factory.

At this point it should be noted that AMD's SledgeHammer architecture had many other advantages, such as being the first consumer CPU to have a memory controller on-die, a technology that Intel would eventually adopt despite a few Intel old-timers telling me that it would effectively tie the CPU architecture to a specific memory architecture, rather than letting the 'Northbridge' chipset act as a intermediary.

So AMD had a good architecture but really couldn't show it off and despite its work within the HyperTransport consortium helping it to win business in the two-socket and four-socket server market it still couldn't really capitalise. This time however AMD has gone for software support through the HSA Foundation.  

Pressure from all directions

I do believe the work AMD is doing with its partners in the HSA Foundation, which include chip vendors such as Qualcomm, Samsung and Texas Instruments alongside GPU vendors such as ARM and Imagination Technologies, is absolutely vital to its future survival. However given that AMD's hard work will also help other chip vendors, it places an extraordinary amount of pressure on its silicon engineering team.

With AMD helping create a software ecosystem that will help other chip vendors, I was told by John Taylor director of AMD's Global Business Units Marketing that the firm's "silicon can stand on its own". I wouldn't expect Taylor to say anything less and it is a fair comment from an industry watcher's standpoint; the silicon should be able to win given a fair software ecosystem, just as it should have done with AMD's SledgeHammer architecture back in 2003.

However Taylor's comments shows just how much work AMD will have to do in make sure both its CPU and GPU can compete with numerous chip vendors that will also take advantage of the same fair software ecosystem. When I asked Taylor whether last year's 10 percent cut in AMD's workforce would effect its ability to manage its software and hardware obligations he claimed the firm doesn't have a resource issue.

AMD wil find it hard to pursue the software relationships it needs to build and more importantly maintain alongside silicon development for GPUs, x86 Opteron processors, x86 APUs and ARM SoCs. Taylor also told me that AMD is looking for more chip design work in the same vein as the APU used in Sony's PlayStation 4, all of which require long term engineering commitment.

AMD has shown that working with software developers works when it comes to GPUs, the Radeon HD 7970 graphics card has seen considerable performance improvements thanks entirely to AMD's work with games developers and driver optimisation. The firm has clearly realised that it needs to put more effort into developer relations, something Intel has done for decades, if it intends to make its silicon investment go further.

The problem for AMD isn't its vision because for the first time in five years it really does seem like AMD is on the right track, but whether the firm has enough resources to build out its silicon and push it ahead of others in the HSA Foundation.
comments powered by Disqus