Rys Sommefeldt

Architecture

There’s been plenty of press recently about Jim Keller’s departure from AMD. Ordinarily an architect leaving a producer of consumer semiconductors like AMD would deserve some press, given the architect’s role in processor design, but the discussion around Jim Keller has taken a different track because of AMD’s current embattled position in the market.

For almost a decade now, Intel have had the overall lead in x86-compatible processor design, which has mapped reasonably well to a corresponding decline in AMD’s fortunes. Once reasonably dominant in terms of performance, although granted in a fledgling PC market that doesn’t see anything like today’s volumes, AMD have been on a slow slide since Intel debuted the Conroe microarchitecture in 2006.

Jim Keller’s return to AMD in 2012, following a spell at Apple where he worked on their first in-house SoCs, as Apple transitioned away from partnership with Samsung for their chip design and manufacture, was supposed to herald a return to fortune for AMD. That never really materialised. Now he’s left, over a year before AMD expect to introduce Zen, their next-generation x86 processor architecture, industry colour commentators have a lot to chew on.

Sadly most of the commentary hangs all of the success of AMD’s future x86 processor architectures on Keller alone, as if one man is solely capable of engineering the success or failure of a processor design. It just doesn’t work like that. An architect, no matter how good, doesn’t single-handedly take the architecture all the way through to manufacture, also doing hardware design, layout, synthesis, test, manufacture and everything else.

Sure, without a great architecture it makes it harder for great end products to come out as a result, but there are so many people involved in the flow that no one person can be responsible for the success or failure of what happens. Zen’s success also hangs on Keller’s co-architects — there’s absolutely no way he designed the entire microarchitecture himself — the hardware team that took the architecture into RTL, the back-end team, their foundry partner, and plenty of other soft teams like marketing, competitive analysis, and sales.

So while it’s nice to have a talisman to look up to at semiconductor companies, to attribute success or failure to, the reality is that it takes hundreds of people connected in a web of obvious and non-obvious ways to ship a successful chip today. Keller is important to AMD, but nowhere near the only person that can endow the company with a great end product.