When AMD finally gets around to making 45nm CPUs, there will be two kinds, those with L3 cache, and those without. The cacheless ones go under the code name “Propus.”
We now hear, unofficially, that L3 cache doesn’t matter too much.
Really. Well, if that’s so, who is going to get fired at AMD for making the boneheaded decision to include all that L3 cache in the other K10.5 designs? After all, all that L3 means AMD will effectively reduce its yearly capacity by millions, maybe by as much as 10% of total capacity.
Seriously, folks, of course L2 matters; it just doesn’t matter as much with the Hammer (or Nehalem) design as it does for the C2D/Penryn design.
However, given the competitive and financial environment, while L3 doesn’t make a gross difference in performance, the marginal difference it does make matters.
Adding L3 to the K10.5 was just a “toss the kitchen sink” attempt to get K10.5 within fighting range of Nehalems and show some significant improvement for K10.5 clock-for-clock over the K10.
Was Intel desperate, too, when it started piling up cache in C2D and Son? Sure, that design needs the extra cache. The key difference is that Intel isn’t capacity-restrained like AMD is now.
Keep in mind that AMD sliced back cache on its Hammers from 1MB to 512K a few years back just to get a few more chips out, so putting in an extra 6MB, even given process shrinks, is a real capacity sacrifice.
The real reason why AMD is willing to do that is the hope that any reduction in CPU production will be more than offset by increased CPU pricing. If K10.5 with cache trails Nehalem by 10% rather than 20%, it’s less likely that Intel will be able to put out a complete mainstream product line that outperforms AMD’s best. That ought to improve AMD’s overall pricing structure.
We’ll see about that, but getting back to Propus, the PR may be nonsense, but the CPU isn’t. The purpose behind Propus is obvious. This is going to be the entry-level OEM chip line, the successor to the X2s, the chip line you sell when the OEMs won’t pay squat for it.
Why should overclockers care about this? This will also probably be the first $100 quadcore (with any dual or tricore versions costing even less).
Need I say any more?
Now that I’ve piqued your interest, I’m afraid I’ll have to let you down a bit. We won’t see Propus until sometime in 2009, let’s say June-August, assuming all goes reasonably well. When might we see a $100 quad? I’d guess around Christmas or a bit after.
Like K10 duallies, the launch date of Propus really tells us when AMD is ready to seriously start making CPUs, rather than looking like they’re making them by making just a relative handful.
Again, assuming no major screwups, we’ll see Propus’ big brother with all that unnecessary cache for sale at prices a lot higher than $100.
But those who are patient will be rewarded.