Understanding the Athlon 64 Family Learn how Athlon 64 processors bring the promise of 64-bit computing to the desktop as well as high performance with today's 32-bit software
Understanding the Athlon 64 Family
Learn how Athlon 64 processors bring the promise of 64-bit computing to the desktop
as well as high performance with today's 32-bit software
On September 23, 2003, Advanced Micro Devices (AMD) introduced its long-awaited Athlon 64 family of processors, the first 64-bit desktop processors based on x86 technology. In this article, youll learn about the common features the three members of the Athlon 64 family share, how each is optimized for its market niche, and how the family compares to its arch-rival, the Intel Pentium 4.
Meet the Athlon 64 Family
The Athlon 64 family includes three different processor versions:
- Athlon 64 for desktops
- Athlon 64 for notebooks
- Athlon 64 FX for extreme performance desktops
All three members of the Athlon 64 family share a common processor architecture: AMD64 (see my article The AMD Opteron: Previewing 64-Bit Desktop Processing for details).
AMD64 enables the Athlon 64 family to move as desired from todays 32-bit x86 software to the developing wave of 64-bit AMD64 software (some Linux distributions already support AMD64, and more are coming; a 64-bit version of Microsoft Windows XP is also being developed). Fortunately, given the current dearth of 64-bit software, AMD64 alone doesnt account for the performance of the Athlon 64 family.
Larger Memory Caches for High Performance
All members of the Athlon 64 family feature very large L1 and L2 memory caches (which run at the same speed as the processor). Memory caches contain a copy of the most-recently read information from main memory so that the processor can use the caches copy rather than accessing slower main memory for information. In my book Upgrading and Repairing PCs, I compare the relationship of memory caches and the processor to a restaurant waiter whose ability to anticipate what I want to order next keeps me well-fed and hardly ever waiting for the next course.
The cache size and design used by the initial members of the Athlon 64 family are the same as those pioneered by the AMD Opteron:
- The Athlon 64 familys L1 and L2 caches keep the processor core well-fed with information in part due to their large size: each Athlon 64 processor contains a 64KB L1 cache for instructions, a 64KB L1 cache for data, and a 1024KB (1MB) L2 cache. By comparison, the Intel Pentium 4 has a 20KB L1 cache and 512KB L2 cache.
- The L1 instruction cache supports parity checking, and the L1 data cache and L2 cache supports ECC error checking. These features help ensure that the contents of L1 and L2 cache are reliable.
- The Athlon 64 familys L1 and L2 cache also feature improved branch prediction, which enables the processor to more accurately anticipate which instructions will be used next and enhanced TLB (translation lookaside buffers) to help the processor recover faster from cache misses.
Although the initial versions of the Athlon 64 desktop and mobile processors have the same 1MB L2 cache as the high-performance Athlon 64 FX and the Opteron, future versions might use smaller 256KB or 512KB L2 caches.
Breaking the Chipset Paradigm An Integrated Memory Controller
As I have discussed over the years in my book, Upgrading and Repairing PCs, traditional processor designs communicate directly with only one type of memory: integrated cache memory. Since the days of the Intel Pentium, all processors have contained L1 memory cache, and since the days of the Intel Pentium II and Pentium Pro, almost all processors have also contained L2 memory cache. Recent processor designs have integrated both L1 and L2 caches into the processor core for full-speed memory access. Some workstation and server processors such as the Intel Xeon MP, Intel Itanium family, and the new Intel Pentium 4 Extreme Edition add a third level of cache known as L3.
Because the memory controller for main memory is typically located in the North Bridge or Memory Controller Hub chip on the motherboard and these chips use a relatively slow bus to connect to the processor, overall memory performance isnt as fast as it could be.
The AMD Opteron was the first x86-based server and workstation processor to break this paradigm by moving the memory controller into the processor itself, and the Athlon 64 family brings this new design feature to the desktop. All Athlon 64-family processors include an integrated DDR memory controller with support for up to PC3200 (400MHz) memory. To improve reliability, the Athlon 64 familys integrated memory controller supports ECC memory; ECC memory corrects single-bit memory errors.
Placing the memory controller inside the processor helps improve performance because the connection between the processor and the North Bridge chip no longer needs to carry memory traffic, freeing up bandwidth for AGP graphics, ATA/IDE drive, and other data.
HyperTransport for Faster Chipset Connections
AMD has long used the high-speed HyperTransport bus to connect its processors to the North and South Bridge chips on the motherboard. The Athlon 64 family continues this tradition, using a 16-channel by 16-bit wide HyperTransport connection with a maximum speed of 6.4GB/second (3.2GB/second in each direction).
The Athlon 64 also add support for SSE2, the second generation of the SSE instruction set for enhancing graphics, sound, and multimedia processing introduced by Intel in the Pentium 4. Support for SSE2 enables the Athlon 64 family to benefit from the same SSE2 software optimizations that boost the performance of the Pentium 4.
The initial versions of the Athlon 64 family use a .13 micron process, featuring copper interconnects and silicon-on-insulator technology. This design should allow the Athlon 64 to scale to faster clock speeds quickly.
By mid-2004, AMD expects to have transitioned the Athlon 64 family over to the smaller and more efficient .09 micron manufacturing process, which should make it possible to ramp up clock speeds even further.
Specific Features of Athlon 64 for Desktop and Mobile
The Athlon 64 desktop and mobile processors are practically identical inside. Both feature
- A single DDR memory controller
- 1MB of L2 cache
- 64KB L1 instruction and 64KB L1 data caches
- A single HyperTransport channel
- Sophisticated power management
- Socket 754
- Organic micro pin grid array package
Essentially, they differ in how the top of the processor is prepared and in clock speeds offered. The Athlon 64 desktop processor is built with a metal heat spreader across the processor, similar to the packaging used by the Intel Pentium 4 processor. The first Athlon 64 desktop processors have the 3200+ model number, running at an actual clock speed of 2.0GHz.
By contrast, the first Athlon 64 processors for mobile computers use a so-called lidless design, enabling OEM vendors to create customized heat spreaders matched to the specific design requirements of various portable computers. The Mobile Athlon 64 is available in two models, 3000+ (1.8GHz actual clock speed) and 3200+ (2.0GHz actual clock speed).
Athlon 64 FX, or Opteron, Take 2
The high-end Athlon 64 FX differs in several ways from the mainstream Athlon 64, including:
- Dual-channel (128-bit plus ECC) integrated memory controller using registered DDR memory modules up to 400MHz (PC3200). Because PC3200 registered memory has not yet been approved by JEDEC, early Athlon 64 FX systems might use PC2700 (333MHz) registered memory instead.
- Ceramic packaging with a metal heat spreader
- Socket 940
- 2.2GHz actual clock speed (Model FX-51; higher speeds are expected in future versions)
A careful examination of these features suggests that the Athlon 64 FX is in reality a modified version of the AMD Opteron 100 series. Compared to the AMD Opteron 100 series, the Athlon 64 FX has only one HyperTransport link (all Opterons feature three HyperTransport links), but the Athlon 64 FX runs at 2.2GHz, compared to just 2.0GHz in the AMD Opteron model 146 (which is the top of the Opteron model 100 series at present).
Although current Opteron and the first Athlon 64 FX processors use the same socket (Socket 940), the Athlon 64 FX will eventually switch to a 939-pin socket. AMD is expected to sell processors in both 940-pin and 939-pin sockets during 2004.
Although the dual-channel memory controller feature in the Athlon 64 FX improves memory access, the Athlon 64 FX currently requires the user to install registered modules instead of the unbuffered modules required by the Athlon 64 and other high-performance processors such as the Intel Pentium 4 and AMD Athlon XP. This means that moving up to the Athlon 64 FX requires a new motherboard, new processor, and new memory an expensive proposition. Future versions of the Athlon 64 FX might use unbuffered memory.
The first Athlon 64 and 64 FX-based motherboards to hit the market use one of the following chipsets:
- nVidias nForce 3; a single chip product with support for RAID 0, 1, 0+1 with SATA/ATA-133; AGP 8x graphics interface; USB 2.0; Gigabit Ethernet; onboard six-channel audio. See http://www.nvidia.com/page/nf3.html for details
- VIA Technologys K8T800; a two-chip product with support for 533MB/second V-Link connection between North and South Bridge chips; integrated 5.1 audio with support for 7.1 audio through the PCI bus; optional Gigabit Ethernet; ATA-133/SATA RAID; USB 2.0; optional additional SATA drives. See http://www.viatech.com/en/k8-series/k8t800.jsp
Most benchmark tests list results for both products, because the differences in chipset design can make a difference to your results.
Benchmarking the Athlon 64 Family
The Athlon 64 family arrived complete with benchmarks performed by AMD in its AMD Processor Performance Evaluation Guide available in Adobe Acrobat PDF form from http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/30579_AMD_Processor_Evaluation_Guide3.1.pdf.
AMDs benchmarks use popular productivity and gaming applications such as Microsoft Office 2000, Microsoft Office XP, Adobe Photoshop versions 6 and 7, Quake III, Comanche 4 and many others to compare the 2.2GHz versions of the Athlon 64, Athlon 64 FX to two configurations of the 3.2GHz Intel Pentium 4 using the 800MHz FSB bus. In the published benchmarks, both the Athlon 64 and the Athlon 64 FX beat the Pentium 4 in overall performance as well as on most individual tests (see the evaluation guide for details). Third-party benchmarks, as always, show mixed results (see the links following this article for details).Some third-party benchmark tests compare the Athlon 64 family to the brand-new Intel Pentium 4 Extreme Edition as well as the regular Pentium 4 3.2GHz model. To learn more about this chip, see my article "The Pentium 4 Extreme Edition", running at a full 1GHz speed advantage over the Athlon 64 FX (3.2GHz versus the Athlon 64 FXs 2.2GHz), surpasses the FX on some benchmarks, but loses out on others.
The Athlon 64 family demonstrates that an optimized design running at a relatively slow clock speed (compared to top of the line Intel processors) can compete with faster processors using current 32-bit software. However, the full benefits of 64-bit desktop processing afforded by the Athlon 64 family arent available at present because of a lack of 64-bit applications and operating systems.
As faster Athlon 64 versions are released and 64-bit operating systems and applications are also released, the Athlon 64 family will become a better choice than its 32-bit rivals for both 32-bit and 64-bit applications.
The official AMD website for the Athlon 64 FX processor
The official AMD website for the Athlon 64 desktop processor
The official AMD website for the Athlon 64 mobile processor
These sites contain links to FAQs, technical documents, and architectural briefs.
Third-party reports and benchmarks (including comparisons with the Pentium 4 Extreme Edition):
Neoseeker offers links to various other reviews
Copyright©2003 Pearson Education. All rights reserved.