AMD's Newest Processors Power More than Just Your Desktop Athlon MP and Mobile Athlon 4
AMD's Newest Processors Power More than Just Your Desktop
Athlon MP and Mobile Athlon 4
AMD has been making great inroads to Intel's processor market share for desktop PCs, but has been mostly, if not entirely, absent from the mobile and multi-processor workstation/server markets. That is because until just recently, AMD simply didn't have much of anything to offer in those categories. That is finally changing. AMD's latest processors, the Mobile Athlon 4 and the Athlon MP, are designed to compete with Intel in markets where Intel has formerly had no real competition. The Mobile Athlon 4 goes up against the Mobile Pentium III in the laptop and portable system market, while the Athlon MP goes up against the Intel Xeon and Itanium processors in the multi-processor workstation and server market. These new processors are based on the new AMD Palomino core, which will also be out in a desktop Athlon 4 chip later this summer. . AMD has also introduced a new motherboard chipset to enable the Athlon MP to support multi-processor motherboard designs. . So, let's first learn about how these processors, based on the new Palomino core, differ from the processors based on the existing Thunderbird core.
New Features
All CPUs based on the Palomino core use the same Socket A (Socket 462) interface that AMD originally developed for the Duron and Athlon Thunderbird processors. These processors have 128KB of L1 and 256KB of on-die L2 cache running at full processor speed. However, compared to the AMD Athlon Thunderbird series, Athlons based on the Palomino core feature improvements in the areas of power and thermal management and performance.
Improvements to the Athlon's Power and Thermal Management include:
- 20% less power usage at the same clock speed, thanks in large part to the use of optimized transistors.
- Support for AMD PowerNow!, which slows down the CPU in response to the performance needed for the current operation. This feature supports up to 32 speed increments between 500MHz and the processor's top speed, compared to just two speeds in Intel's SpeedStep technology for its Mobile Pentium III processors. Because reducing speed reduces power consumption (and thus increases battery life), this feature is aimed squarely at notebook computer users, although all Palomino-core Athlon processors have this feature.
- Thermal diode to allow monitoring of the CPU core temperature by a suitably-equipped motherboard and BIOS; this works along with PowerNow! to trigger CPU slowdowns as needed and to monitor the system's health. This can also warn you of a failed heatsink fan.
In terms of improving performance, the Palomino features:
- Full support for the same SIMD (single instruction multiple data) commands, also called SSE (streaming SIMD extentions), that Intel introduced with the Pentium III. These commands, an extension of MMX, speed up multimedia operations by allowing a single command to work with multiple sets of data. AMD's term for its SSE support in the Palomino core processors is 3DNow! Professional, indicating that these processors also support AMD's own 3D Now! multimedia commands.
- Improved data prefetch (gathering information before being specifically instructed to do so), which uses CPU Bus bandwidth that is not in use for other processes. Data prefetch is supported not only by the SSE instructions now included in the Palomino core, but is also automatically initiated by the processor itself, even if the software being operated doesn't use SSE instructions. This is similar to the way that the Windows 32-bit disk cache software "guesses" about which disk areas to read next. System performance is much faster than if the processor (or the operating system) had to wait specifically to be told which information was needed next.
- Greater number of L1 Data Translation Lookaside Buffers (TLB) (now 40, was 32 on older Athlon Thunderbird) to improve L1 cache hit rate. TLBs work by storing translated main memory addresses for quicker access to data, and more buffers make the L1 data buffer more efficient.
- Exclusive architecture in L1 and L2 cache TLBs enable each cache to store different memory addresses. This improves memory performance, especially when using faster memory technologies such as DDR SDRAM.
- Speculative reloading of TLB entries; processor doesn't wait for reload instruction to finish before reloading.
As a result of these design improvements, the performance of a Palomino-core Athlon is up to 15% faster than a Thunderbird-core Athlon at the same clock speed.
Athlon Multi-Processor Specific Features
The Athlon MP, while it can be used on a single-processor motherboard, is the first Athlon CPU intended and certified for use in a dual-processor motherboard, a popular configuration for servers and high-end workstations. However, all Palomino-core Athlons share the following multi-processor-specific features:
- Support for the Point-to-Point CPU Bus used by the AMD 760MP chipset. This type of CPU Bus enables dual processors to communicate directly with each other when the data needed by one processor has already been stored in the cache of the other processor. Other MP-capable CPUs use a shared CPU Bus, which requires that CPU to CPU communications be routed all the way back to main memory, which is extremely inefficient by comparison.
- A better cache coherency protocol (the process of keeping both processors updated with information) known as MOESI (Modified, Owned, Exclusive, Shared, and Invalid) . The addition of the Owned state to the standard MESI protocol used by other processors, including the Intel Pentium 4 and all Xeon processors, works along with the Point-to-Point CPU Bus to improve the speed at which information is share between the processors.
Clock Speeds
Clock speeds for these news AMD chips vary based on the hardware it's being targeted for. The Mobile Athlon 4 is available at the following clock speeds:
- 850MHz
- 900MHz
- 950MHz
- 1GHz
The Athlon MP is available at the following clock speeds:
- 1GHz
- 1.2GHz
AMD's 760MP Chipset Makes Multi-Processing Possible
The trick with using a CPU capable of multi-processor operation is that it can't be used in this mode unless a suitable chipsets exist. The 760MP chipset enables the Athlon MP to live up to its name. It comprises the 762 System Bus Controller (North Bridge) chip, a large (949-ball) BGA (ball grid array) chip and the same 766 Peripheral Bus Controller (South Bridge) that was also used with AMD's 760 chipset.
Unlike previous AMD chipsets, which were primarily used by AMD to get new processors rolling, the 760MP is expected to be a long-term product for AMD.
The 762 chip provides both CPUs with their own North Bridge connection to improve memory bandwidth; as previously discussed, their caches are upgraded via the North Bridge (Point to Point CPU Bus), not via main memory (as with Xeon and other MP processors). As a result of providing each processor with its own connection to memory, the 760MB can provide 2.1GBps memory bandwidth per processor. While its rival, the Intel Xeon, has 3.2GBps bandwidth, the Xeon must divide this among the processors on-board. In a dual-Xeon configuration, for example, the Xeon can offer each CPU only 1.6GBps memory bandwidth.
The 760MP also supports 64-bit PCI slots, which can also accept normal 32-bit PCI cards, and the AGP Pro slot for high-speed video. The 760MP supports up to 4GB of PC2100 DDR SDRAM, a 266MHz CPU Bus, and AGP 4x graphics. The Tyan (www.tyan.com) Thunder K7 (S2462) is the first motherboard available based on the Athlon MP processor and 760MP chipset.
AMD Mobile Duron
In addition to the Mobile Athlon 4, AMD is releasing a new version of its lower-cost Mobile Duron processor, which is based on the Morgan core. This processor has many of the features of the Palomino core that the Mobile Athlon 4 and Athlon MP use. Both the Morgan and Palomino cores share the following major features discussed earlier in this article:
- PowerNow! power management
- 3D Now! Professional (3D Now! + SSE) multimedia enhancements
- Enhanced hardware pre-fetch features
- Thermal diode
The Mobile Duron, like other Durons, has 128KB L1 cache and on-die 64KB L2 cache running at full CPU speed. The Mobile Duron is being introduced at clock speeds of 800MHz and 850MHz. The Mobile Duron works with the same chipsets that support the Mobile Athlon 4, including Ali's MagiK 1 and VIA's KT133A.
Web Site Resources
For technical information on these processors, see the AMD web site at www.amd.com. You can find benchmark tests results for these processors at sites such as AnandTech www.anandtech.com, Tom's Hardware www.tomshardware.com, and Ace's Hardware www.aceshardware.com.
Conclusion
AMD's Mobile Athlon 4, Athlon MP, and Mobile Duron make the company a player in two areas where it's previously had either a minimal presence (notebook computers) or no presence at all (multiprocessing servers and workstations). These processors (along with the AMD 760MP chipset) should help make AMD a major player in these markets, as they continue to push Intel on the desktop market as well.
© Copyright 2002 Pearson Technology Group. All rights reserved.