AMD’s Processor Lineup: A Yield Party
All modern processors consist of three main elements that must be carefully balanced: core count, cache capacities, and clock speed. This balance must take into account the manufacturing process, possible voltages, clock rates, thermal and electrical limits, yields, and eventually total cost.
Shrinking the manufacturing technology—for example, from 65nm to 45nm—allows chip makers to optimize one or several of these parameters. Smaller and more efficient transistors can usually operate at higher clock speeds. But it’s also possible to add more cores or larger caches to increase performance. Lastly, manufacturers may leave processor designs largely unchanged and simply bring power consumption down. This approach can also buy chipmakers more time to collect experience on a new process before making modifications.
Since AMD does not have Intel’s vast manufacturing capacities (the company recently spun off its facilities into GlobalFoundries), it has to focus on maximizing output. As a result, the vast majority of AMD’s products at any given time are based on a single processor design that can be modified (usually simplified) to address different price points and market segments, all the while maximizing yield rates. The issue here is simple: one size doesn’t fit it all on the market anymore, but one size has to fit it all in manufacturing.
Intel, by the way, has been doing very much the same thing. All 45nm Core 2 processors are technically based on the dual-core Wolfdale design, and the firm utilizes two of them to create the Yorkfield quad-core CPUs (Core 2 Quad, Extreme). Intel only modifies the dies by limiting L2 cache capacities. AMD, however, has been much more aggressive in creating different products out of the 45nm quad-core Deneb design. The company dives deeper into the dies, switching more individual units off (or on) to master the yield challenge. The result is sometimes different dies that all share the same origin. Here’s a quick overview of multiple AMD products all based on the same underlying design:
Deneb, quad-core, 6 MB or 4 MB L3 cache (2.4 to 3.4 GHz)Heka
, triple-core, 6 MB L3 cache (2.4 to 3.0 GHz)Callisto
, dual-core, 6 MB L3 cache (3.0 to 3.1 GHz)
Propus, quad-core, no L3 cache (2.6+ GHz)
Rana, triple-core, no L3 cache (2.7+ GHz)
Regor, dual-core, no L3 cache (2.8 to 3.0 GHz)
Editor's Note: Getting Turned On
Incidentally, AMD has confirmed that early Athlon II X4s are being sold on Propus and Deneb core designs, the former without any L3 cache by design and the latter with 6MB of L3 disabled.
We dusted off our faithful ASRock M3A790GXH/128M motherboard, which we've used in the past to unlock Phenom II X3s and X4s, and then unlock Phenom II X2s. Unfortunately, while we've seen the screenshots of Athlon II X4s with 6MB L3 cache, our 620 ran with ACC enabled, but didn't unlock the L3 cache, while our 630 simply wouldn't boot.
As before, don't buy one of these less-expensive chips counting on an easy upgrade with the right SB750-equipped motherboard. Though a handful of processors might surprise you, chances are good that you won't get the equivalent of a Phenom II X4 out of one of these new Athlon IIs.