In baseball, amassing a single, double, triple and home run in the same game is known as hitting for the cycle. AMD will try for the chip industry equivalent next year.
The company announced plans Monday to introduce a desktop PC processor with three cores in the first quarter of 2008. The three-core chip will carry the same Phenom brand name that AMD plans to attach to its quad-core desktop chips due to ship to PC companies by the end of this year.
Bob Brewer, corporate vice president of marketing and strategy, said that AMD's move was made in recognition that demand for quad-core chips has been tepid. There's simply not a lot of desktop software that can take advantage of four cores, and most of the growth in the PC market right now is in the notebook section, where quad-cores haven't yet made an impact.
But PC owners can still take advantage of multiple cores if they're running several applications at once, Brewer said. Therefore, AMD wants to offer the choice of buying a three-core processor for perhaps a little less than a quad-core chip.
Now, for the real reason AMD's doing this.
The three-core Phenom chip is basically the same as the quad-core one, it just has one less working core, Brewer confirmed. One disadvantage of the monolithic quad-core design that AMD chose for its quad-core chips is that just one manufacturing defect on part of the chip can knock out an entire quad-core processor. But if you invent a category for three-core chips, suddenly you can make money off those chips that would otherwise have to be discarded because of a defect that disabled one core.
This can lead to all sorts of speculation about the yields AMD is getting on its first quad-core parts. Barcelona, the server version of its quad-core design, was six months later than expected due to unspecified "technical glitches", which may or may not have to do with yields. It's really hard to know for sure, since no chip company wants to discuss yields.
But it's conceivable that AMD needs to allocate all of its working quad-core models toward the much more lucrative server market to boost its average selling prices. Then, to serve the desktop market, the company can trot out tri-core (triple-core? three-core?) chips.
If yield pressures aren't forcing its hand, then this is actually a good move for AMD. After all, you weren't going to get anything out of quad-core chips with a single busted core. Chip companies have been doing this for years; Intel's Celeron and AMD's Duron chips were the same as the Pentium 4 and Athlon XP chips, just with some cache memory disabled. The disabled transistors don't work, obviously, but they don't have any adverse effect on the working transistors.
This is also a product that Intel won't be able to easily match just yet. Intel's road map for the next six months or so calls for packaged quad-core chips, two dual-core chips put together into the same package. Intel's unlikely to switch to a design that could accommodate three cores until later in 2008 when it releases the Nehalem generation of chips.
But I tend to think this is going to be confusing for PC buyers already faced with four-digit model numbers when trying to make a decision on a processor. Because the three cores will be able to take advantage of the amount of cache memory that's usually allocated for four cores, a fast triple-core chip could outperform a low-end quad-core chips in certain situations. AMD will have to figure out a coherent way to explain that if triple-core chips are to take off.
Perhaps it's also worth noting that the triple is the hardest part of the cycle to pull off. AMD declined to specify its pricing plans for the triple-core Phenom processors, but the pricing will probably play a large role in how the chips are received by PC companies and buyers.
This article was originally a blog post on CNET News.com.
Play video
Three cores is ideal for low-cost multiprocessor development platforms. Why? There are many concurrent algorithms that work for two threads of execution, but not more.
Although this is true for any number greater than two as well, the difference is that any reasonably simple algorithm that has a limit at one of these higher values rather obviously has such a limit. Not so for two threads, however.
As a result, developers building and testing concurrent algorithms on a machine with at least three processors are far less likely to fail when executed on machines with four or more processors. This will result in better code and easier migration to higher numbers of CPUs.
While I've been waiting for four CPU machines to come down in price, I would gladly purchase one with three if there is a significant price difference (which AMD should be able to offer).
Posted by James Parker on Wednesday, September 19 2007 09:47 AM