AMD AthlonT Processor Performance-Enhancing
Enabling an Innovative Full-Speed On-die
for High-Performance x86 Processors
ADVANCED MICRO DEVICES, INC.
One AMD Place
Sunnyvale, CA 94088
Introduction: The AMD AthlonT Processor with
Performance-Enhancing Cache Memory
At its introduction, the AMD AthlonT processor
marked the arrival of the world's first seventh-generation microarchitecture
and in doing so set the performance standard for x86 processors. Among the
processor's now award-winning architectural features are a revolutionary
200MHz, 1.6Gbytes/sec system bus, a fully pipelined, superscalar floating point
engine, and an enhanced version of AMD's 3DNow!T technology. Additionally,
the processor offers 128K of L1 cache-four times the L1 cache of competing x86
processors-along with a 512K external backside L2 cache running at up to half
the speed of the processor core.
With new versions of the AMD Athlon processor,
however, AMD has sought to improve overall system performance by integrating the
processor's L2 cache directly onto the processor die. This white paper
explains the benefits of utilizing an AMD Athlon processor with on-chip, performance-enhancing
The AMD Athlon processor with
performance-enhancing cache memory boasts three times the full-speed on-die
cache of previous AMD Athlon processors. It
features 128K of L1 cache plus 256K of full-speed, on-chip cache for a total
internal system cache of 384K.
Integration of the L2 cache onto the die
eliminates the need for expensive, high-performance L2 cache SRAM. Moreover,
on-die L2 cache enables the L2 cache speed to scale with the processor speed,
thereby enabling a more substantial level of performance across a broad range of
standard and memory-intensive applications.
With the AMD Athlon processor's migration to
0.18-micron process technology completed, the integrated cache design does not
significantly change the size of the die. The advanced 0.18-micron manufacturing
process enables smaller transistor sizes in comparison to the 0.25-micron
process. Therefore, the space penalty for the integration of the L2 cache is
mild, as the L2 cache accounts for only 20% of the entire processor die.
The AMD Athlon processor's 37-million-transistor-die is now ~120 mm2.
Improved L2 Cache Efficiency through Lower Latency, Increased Bandwidth and
Integrating the L2 cache onto the processor die
significantly lowers the hit latency, as it takes much less time to move data
across the die than to read it from an external SRAM. The lower the latency, the
shorter the response time required for receipt of requested data. The AMD Athlon
processor with performance-enhancing cache memory delivers more than 45% lower
latency compared to previous AMD Athlon processors, dropping from 21 cycles to
The increase in cache speed enabled by integration
of the L2 cache onto the processor die increases L2 cache bandwidth by 300% over
previous AMD Athlon processors. Higher bandwidths allow the processor to work on
more data over time.
The new AMD Athlon processor's 16-way set
associative cache is eight times more associative than previous AMD Athlon
processors featuring a 2-way set associative cache. Increasing the set associativity significantly increases the
hit rate. A hit occurs when the processor successfully finds the data that it is
looking for in cache memory. If the
data is not found in cache memory, the processor must search for the data in
main memory. Because main memory is
limited by its lower speeds-currently no faster than 133MHz-and is serviced
by the front-side system bus, cache hits significantly decrease latencies.
The Benefits of Exclusive Cache Architecture
The new AMD Athlon processor with
performance-enhancing cache memory features an exclusive cache architecture as
opposed to an inclusive cache architecture. An inclusive cache architecture
requires the L2 cache to duplicate every cache block held by the processor's
L1 cache. In other words, for every
cache block in an inclusive cache architecture's L1 cache, the L2 cache must
contain the same redundant data, thereby decreasing the amount of L2 cache
available for new information. An exclusive cache architecture, on the other
hand, contains only victim or copy-back cache blocks to be written back to the
memory sub-system. This provides a
full 256KB of dedicated L2 cache and 128KB of L1 cache for a total dedicated
data space of 384KB.
To improve yield, the cache includes redundant
columns. Redundant columns improve
yield by allowing defective memory cells to be excluded from the cache without
sacrificing the overall integrity of the cache. The entire array can be quickly validated during
manufacturing using a built-in self test (BIST).
Fill Buffers, Bus Queue Entries and Write Back Buffers
The new AMD Athlon processor, like previous AMD
Athlon processors, includes eight fill buffers, eight bus queue entries, and
eight write back buffers. Intel's Pentiumr III processor only has
six fill buffers, eight bus queue entries, and four write back buffers.
The more buffers that are dedicated to holding data for the
microprocessor to process, the less likely the processor will stall while
waiting for data to be delivered to the processor.
More buffers also make it more likely that data can be placed into the
buffers for execution. These design elements allow the new AMD Athlon processor
to have a sustainable bus bandwidth of 1.6 GB/s-or 85% of peak capacity.
Enabling Improved Performance Across Standard
The AMD Athlon processor's cache architecture is
the first to incorporate a system-based MOESI (Modify, Owner, Exclusive, Shared, Invalid) cache control protocol
for x86 multiprocessing platforms. Since the system logic manages memory
coherency throughout the system by specifying all cache state transitions,
either using a MESI or MOESI cache coherency protocol, and by filtering out
unnecessary processor snoops, AMD Athlon processors are designed to deliver
exceptional performance in both uniprocessor and multiprocessor system
configurations. The AMD Athlon processor cache architecture also supports error
correction code (ECC) protection, which is a required feature for high
reliability of business desktop systems, workstations, and servers. Thus, the
AMD Athlon processor's cache architecture provides the features required for
high-performance computing from desktop to server configurations.
AMD) is a global supplier of integrated circuits for the personal and networked
computer and communications markets. AMD produces processors, flash memories,
and products for communications and networking applications. The world's
second-leading supplier of Windowsr compatible processors, AMD has
shipped more than 120 million x86 microprocessors, including more than 90
million Windows compatible CPUs. Founded in 1969 and based in Sunnyvale,
California, AMD has sales and marketing offices worldwide and manufacturing
facilities in Sunnyvale; Austin, Texas; Dresden, Germany; Bangkok, Thailand;
Penang, Malaysia; Singapore; and Aizu-Wakamatsu, Japan. AMD had revenues of $2.8
billion in 1999.
release contains forward-looking statements, which are made pursuant to the safe
harbor provisions of the Private Securities Litigation Reform Act of 1995.
Forward-looking statements are generally preceded by words such as
"expects," "plans," "believes,"
"anticipates," or "intends." Investors are cautioned that
all forward-looking statements in this document involve risks and uncertainty
that could cause actual results to differ materially from current expectations.
Forward-looking statements in this document about the AMD Athlon processor
involve the risk that the AMD Athlon system bus will not support the
requirements of next-generation system platforms; that AMD may not be successful
in developing an infrastructure to support the processor; that third parties may
not provide peripherals or the infrastructure to support the processor and the
processor's system bus; and that the processor will not achieve customer and
market acceptance. We urge investors to review in detail the risks and
uncertainties in the company's Securities and Exchange Commission filings,
including the most recently filed Form-10K.
AMD, the AMD logo, AMD Athlon and combinations
thereof, and 3DNow! are trademarks of Advanced Micro Devices, Inc. Windows is a
registered trademark of Microsoft Corporation. Pentium is a registered trademark
of Intel Corporation. Other product names used in this publication are for
identification purposes only and may be trademarks of their respective