The AMD Opteron: Previewing 64-Bit Desktop Processing An introduction into the technology behind AMD's first foray into 64-bit computing
The AMD Opteron: Previewing 64-Bit Desktop Processing
An introduction into the technology behind AMD's first foray into 64-bit computing
Introducing the Opteron
You might not ever sit down in front of a computer using AMDs new 64-bit Opteron processor; its aimed at the technical workstation and server market rather than at desktop users. However, the odds are fairly good that at some point you might connect to an Opteron-based server or view technical documents generated stored on an Opteron-based workstation. For most of us, what may be even more important about Opteron than its performance in these tasks is what it indicates about the future of desktop computing.
64-bit computing offers access to larger amounts of RAM, faster and more precise floating-point calculations, and the ability to process larger databases more quickly. Consequently, business users are very interested in 64-bit processing. Opteron provides a second way to reach that goal.
64-bit Processing That Leaves the Past Behind
Strictly speaking, the Intel-compatible world has had 64-bit computing for a few years now, thanks to the Intel Itanium and Itanium 2. However, the Itanium series hasnt been the overwhelming success Intel hoped for. According to IDC estimates, only 3,500 servers running Itanium processors were shipped in 2003.
The reason is software. Itanium uses a new 64-bit processor architecture called EPIC (Explicitly Parallel Instruction Computing). EPIC (also known as IA-64 architecture) uses parallel processing and intelligent compilers to perform six operations per clock cycle, boosting processing speed when native 64-bit applications are used. However, because EPIC is a completely different architecture than the x86 architecture used by the Intel Pentium 4, AMD Athlon XP and previous Intel and Intel-compatible processors, Itanium and Itanium2 processors need specially-compiled 64-bit operating systems and applications to run at top speed. While Windows XP and some Linux distributions are available in 64-bit versions, application support is limited. When these processors are switched to an x86 emulation mode to run existing 32-bit software, their performance is very poor. Intels next version of the Itanium 2, code-named Madison, will have improved 32-bit emulation. Madison will run 32-bit software at speeds comparable to Intels 32-bit Xeon 1.5GHz MP processor.
Moving to 64-Bit Processing with Backward-Compatibility
AMDs Opteron takes a completely different path to 64-bit processing. Opteron is the first processor to use AMDs AMD64 architecture (which is also used by the forthcoming Athlon 64; AMD64 was previously known as x86-64). As the original x86-64 name implied, AMD64 is an extension of the current 32-bit x86 architecture. According to AMD, AMD64 is designed to be a mixture of classic CISC (complex instruction set commands) and RISC (reduced instruction set commands) designs. Opteron runs 32-bit applications as fast as 64-bit applications, and, unlike current Itanium processors, can run both 32-bit and 64-bit applications at the same time. With AMD64, AMD is betting that businesses would prefer to move to 64-bit computing at their own pace.
The Opteron family contains three series of processors, the 100 (single-processor), 200 (dual-processor) and 800 (four or eight-processor configurations). The first Opteron processors to ship are the 240, 242, and 244 models, which run at 1.4GHz, 1.6GHz, and 1.8GHz respectively.
Major Features of the Opteron Processor
Besides 64-bit processing, the Opteron has several other features which separate it from previous AMD processors. These include:
- Integrated 128-bit DDR memory controller supports up to eight registered DIMMs per processor; the North Bridge is used for AGP support
- 1GB L2 cache
- ECC support for all memory, including L1 and L2 cache and tags
- Three separate high-speed serial HyperTransport links to the chipset
- Support for up to eight processors (server versions) or up to four processors (workstation versions)
- Socket 940
Benefits of the Integrated Memory Controller
The biggest single difference between the Opteron and its predecessors is the integrated memory controller. Instead of the overhead involved in sending a memory request to the North Bridge chip, waiting for the North Bridge chip to access memory and waiting again for the North Bridge chip to return the results, the Opteron connects directly to DDR memory. As a result, the Opterons front-side-bus speed is the speed of the processor. Consequently, the Opterons memory throughput is significantly better than with conventional North Bridge access.
Another benefit of the integrated memory controller is most apparent in multiple-processor configurations: the crossbar memory controller. I have discussed crossbar memory controllers before (NVIDIA's nForce Invades the Motherboard Chipset Market and Understanding the GeForce4 Family,) as a way of speeding up graphics card memory access. In the Opteron, the crossbar memory controller enables multiple processors to access each others memory. Each processor can access up to 8GB of RAM. The memory bandwidth with a single processor runs at 5.3GB/second. The memory bandwidth increases by the same amount for each additional processor used in a multi-processor configuration instead of being shared as with other multi-processor designs.
In the future, as different memory designs such as DDR-II are introduced, the integrated memory controller can be disabled and a separate North Bridge chip with onboard memory controller can be used instead.
Hyper-Fast Performance from HyperTransport
AMD co-developed HyperTransport, a high-speed serial interface between the processor and North and South Bridge chips, so its not surprising that HyperTransport is part of the Opterons design. The Opterons version of HyperTransport is much more sophisticated than previous implementations because it supports both coherent and non-coherent data transfers. Coherent data transfers are used in multi-processor configurations for handling memory transfers between processors, enabling operating systems to see the memory connected to each processor as a single memory space. Non-coherent data transfers are used between the processor and I/O devices.
The speed of the Opterons HyperTransport implementation is also impressive. Three separate HyperTransport connections provide 3.2GB/second connections in each direction; since HyperTransport supports full-duplex operations, the bidirectional bandwidth adds up to 6.4GB/second per HyperTransport connection.
Performance, Not Raw Clock Speed
Although the naming scheme used by AMD for its Opteron processors doesnt make any performance rating claims (as do the names used by its Athlon XP processors), AMDs long-time design philosophy of achieving more per clock cycle instead of simply increasing clock speed is again apparent in the Opteron processors.
In early tests performed by PC Magazine, a Polywell server using dual Opteron 242 (1.6GHz) processors outperformed a dual-processor Polywell server running Xeon 2.8GHz processors in most tests. In a different mix of benchmarks performed by Toms Hardware, dual 1.8GHz Opterons pitted against dual 3.06GHz Intel Xeon processors on a motherboard with the new Placer (E7505) chipset showed better performance for Opteron in server tasks, but the Xeon-based system did better in workstation-oriented tasks.
What the Athlon 64 Will Inherit from the Opteron
Although the Athlon 64 is still months away from release, and will differ from the Opteron in many ways (64-bit memory bus, no ECC support, and Socket 754), the Opteron gives us several clues about how the Athlon 64 will perform once it is ready to ship. These include:
- Efficient design will enable Athlon 64 to perform very well despite having a lower clock speed than Pentium 4 processors.
- Although Athlon 64 uses only one HyperTransport link, its bi-directional speed of 6.4GB/second will enable it to handle 3D graphics and other data-intensive operations comfortably.
- Since Athlon 64 is the first desktop processor to incorporate an integrated memory controller for DDR memory, its memory performance will surpass other desktop processors.
- The first-generation Opterons look like excellent choices for servers and workstations, and also suggest that the Athlon 64 will be worth the wait as a desktop processor.
For Further Research
See the official AMD website for the Opteron at
http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_8825,00.html
ExtremeTechs Special Report Opteron Arrives has links to early tests, architectural details, and chipset information. See it at http://www.extremetech.com/article2/0,3973,1036405,00.asp
Toms Hardware Guide tests the Opteron against the Xeon and high-end Pentium 4 and Athlon XP processors at http://www.tomshardware.com/cpu/20030422/index.html
AnandTechs review of the Opteron includes testing the processor in a single-processor setup using normal desktop benchmarks as a predictor of Athlon 64 performance. See the results at http://www.anandtech.com/cpu/showdoc.html?i=1818&p=1 and find links to other parts of their coverage.
Copyright©2003 Pearson Education. All rights reserved.