Athlon64 family to grow

Advanced Micro Devices will use cache memory in somewhat novel ways to broaden out its desktop chip line, including its upcoming Athlon64 processor, according to sources.

The Sunnyvale, Calif.-based chipmaker will launch the Athlon64 with performance ratings of 3400+ and 3100+ in September and then crank the top performance rating up to 3700+ in the fourth quarter, according to sources. By the end of next year, the performance rating is expected to rise to 4300+. Ratings roughly correspond to the clock speeds of Intel chips, so that a 3100+ chip is more or less equivalent to a 3.2GHz Pentium 4.

In addition, AMD is apparently looking at changes to its existing Athlon XP line of processors. It is considering a plan where it would not raise the performance rating on these chips, but instead would change their underlying architecture--especially cache--to reduce manufacturing costs.

Details of the company's plans were first posted on a French Web site, X86-secret.com but were subsequently removed. Independent sources confirmed the information. AMD declined to comment.

The planned changes to the Athlon64 and the Athlon XP processor families should help AMD reduce manufacturing costs and use its limited factory space more efficiently.

But the development of the chip families will substantially depend on AMD manipulating the amount of cache on the processors. Cache is a small pool of memory located on the chip that enhances processor performance by providing fast access to data.

In the Athlon64 line, for instance, the 3700+, 3400+ and 3100+ chips will initially come with 1MB of cache. In the fourth quarter, however, the underlying structure of the 3100+ will change: Its clock speed will substantially increase, but its cache will be reduced to 256KB, or one-quarter the original size.

By increasing the clock speed but decreasing the amount of cache, companies can shrink the size of the chip yet keep performance constant. As these chips are smaller, they are cheaper to make--saving money for the manufacturer. But the balancing act may prove difficult to pull off.

In the Athlon XP line, the cache size will be manipulated in a different manner to the Athlon64.

The current top-of-the-line Athlon XP, the 3200+, launched in May. The chip, also known by its former code-name of Barton, comes with 512KB of cache. All the other earlier Athlon XPs have 256KB of cache. Because of the difference in cache size, the two types of chips are different sizes--Barton is larger.

Enter Thorton
AMD's plans for the Thorton chip, an upcoming member of the Athlon XP family, suggest it will be the same size as the Barton chip, but half of the cache on the processor will be disabled. (It will also be paired with a slower bus, which connects the processor to the memory.) That way it can be made on the same wafer as Barton, meaning AMD could streamline its manufacturing processes.

Thorton also potentially boosts yield--or the number of good chips produced from a wafer--because Barton chips that come out of the factory with imperfections in one segment of the 512KB cache could be sold legitimately as Thortons.

Both AMD and rival Intel have used cache size as a way of differentiating chips in the past, but AMD's current use of cache is somewhat unusual. Typically, manufacturers craft their budget processors by cutting the cache size and slowing down the chip. Lowering both creates a performance gap between processors, allowing the company to market them as different families. Intel largely separates its Pentium and Celeron chip lines in this manner, and AMD took a similar tack with the original Athlon and Duron chips.

"Can AMD increase the clock frequency on the 1/4MB (256KB) part high enough above the 1MB part to convince customers that the parts really, really do perform the same?" wrote Kevin Krewell, senior editor of the Microprocessor Report. "Not all applications respond in the same way. Some applications are more sensitive to cache size and some to clock frequency."

Invite Michael Kanellos into your in-box Senior department editor Michael Kanellos scrutinizes the hardware industry in a weekly column that ranges from chips to servers and other critical business systems. Enterprise Hardware every Wednesday.

Earlier this month, Hewlett-Packard inadvertently posted specifications of a PC that revealed that one version of the Athlon64 would come out with 1MB of cache and a performance rating of 3100+. Other sources have said that a 256KB cache version of the Athlon64 is coming as well.

HP later stated that the PC, which had been destined for the Korean market, had been cancelled, but confirmed that it planned to use the Athlon64 in future PCs.

The cache and bus reductions coming in Thorton fit the more traditional pattern of adjusting chips, but with at least one distinction: Typically, the stumped chips are given new brand names, but the plan is for Thorton to replace existing chips in the Athlon XP family.

AMD executives have said in the past that the company would not likely come out with any further versions of the Athlon XP line after the 3200+.

So, whether Thorton is "new" or not is open to question. It's never been seen before, but it doesn't represent a performance increase, and it will bear the same product name, bus speed and performance rating of chips that have been on the market for a while. And technically, Thorton is almost identical to the older Athlon XP; the main difference is the extra bit of silicon holding the disabled cache.

Sources, though, indicated that AMD may still be deciding whether to move forward on its plans for Thorton.

Kai Schmerer of ZDNet Germany reported from Munich.