|
G e e k G l o s s a r y
970
In October of 2002, IBM introduced the 970, one-half the dual-core "Power" series 64-bit CPU (touted the world's most powerful
processor). The 970 combines the 64-bit processing capability of the Power series with the 32-bit AltiVec unit found on Motorola's
G4s.
AltiVec
A processing engine (data pipe) found onboard the Motorola G4 CPUs. AltiVec enables the ability to process "vector"
code, a streamlined set of calculations that fit neatly into the AltiVec's registers. AltiVec can process 8 x 32-bit instructions in one clock cycle, offering a potential performance of 20
GFLops:
(1.25 x 1 billion cycles/s) * (8 floating point ops/cycle) * (2 processors) = 20GFLops
... with a realworld application of roughly 19 GFLops depending upon how well the code 'vectorizes'. More
information, programming guidance, and an intelligent threaded discussion group exist at www.altivec.org.
cache
Temporary storage of applications and data while a computer is powered on. Cache is closest in physical proximity and speed to the CPU.
Data that is moved from CPU to cache and back to CPU again offers the highest performance, therefore, higher performance computers often
offer larger memory caches. Modern computers offer level 1, 2, and 3 cache, varying in size and performance.
Cell Broadband Engine
The Cell Architecture grew from a challenge posed by Sony and Toshiba to provide power-efficient and cost-effective
high-performance processing for a wide range of applications, including the most demanding consumer appliance: game consoles. Cell - also known as the Cell Broadband
Engine Architecture (CBEA) - is an innovative solution whose design was based on the analysis of a broad range of workloads in areas such as cryptography, graphics
transform and lighting, physics, fast-Fourier transforms (FFT), matrix operations, and scientific workloads. Cell offers a "Primary Processing Unit" (PPU) which is
binary compatable with the IBM 970 CPU (AltiVec unit included) and 8 Synergetic Processing Engines (SPEs) which offer 64-bit native processing and a 32-bit AltiVec
unit.
cluster
Multiple computers connected for the purpose of providing greater performance than single or individual computers.
COTS, Commodity Off The Shelf
Used in the context of clusters built from nearly standard computers that can be purchased from a local
reseller.
CPCI, Compact Peripheral Component Interconnect
A more compact PCI form-factor that readily allows for 'hot-swap' (the removal/replacement) of 'blades' while power is applied. The
'blade' configuration is similar in form-factor to VME.
CPU, Central Processing Unit
The unit on a computer that conducts the primary processing. The CPU is traditionally x86, PowerPC, MIPS, Strongarm, or Spark.
32-bit vs 64-bit
Some computer programs generate numbers that are 32 bits in detail (length), others 64, and fewer 128. The more bits the
calculation carries, the more accurate the result as there are more places behind the decimal point. 64-bit is used for advanced weather
modeling, nuclear simulations, and some bio/medical research. 32-bit is most common and used for genomics, bioinformatics,
interplanetary image processing, synthetic aperture radar, seismic studies, realtime target analysis, some mechanical and structural
engineering, PET/MRI brain imaging, drug research, and some weather modeling.
It is important to note that 64-bit CPUs are not necessarily better than 32-bit CPUs just as 64-bit dependent code is not
better than 32-bit code. 32-bit CPUs with vectorized data pipes can process 32-bit code very quickly, at a better price, and typically
in a higher density environment than the 64-bit CPUs.
cross compiling
When the foundation code for a program is compiled on one architecture (x86) to run on another (PowerPC).
distributed computing
More than one computer (node) interconnected via a communications medium that enables them to work independently, but toward a common
goal. Distributed clusters typically incorporate loosely coupled (various speeds) nodes in either close or distant proximity to one
another. Distributed systems may include similar-performance computers in a single room or dissimilar-performance computers spread
across an entire continent, as in the famous "grid" computing project.
embedded systems
Computers that boot from flash RAM or ROM. This means that they do not require a hard drive from which to load their operating system.
Used in small devices such as PDAs, cell phones, car computers, and many military applications. Embedded systems out-number desktop
systems 10 to 1.
form factor
The physical size and shape of a computer housing.
G3
IBM and Motorola jointly manufacture the 603, 604, 750 (marketed by Apple as the 'G3'). The G3 series has two 'data pipes'
(processing engines) onboard, one to handle integer and the other floating point calculations.
G4
Motorola alone develops and manufactures the G4 (74xx)
family which introduces a 3rd 'data pipe', the AltiVec (Apple's "Velocity Engine")
unit.
G5
IBM alone develops and manufactures the G5 (970) which
marries one-half of the dual-core Power4 64-bit CPU and the 32-bit DSP AltiVec unit into a single powerhouse. IBM's
5th Generation (G5) PowerPC CPU.
GHz, Gigahertz
One billion (1,000,000,000) cycles per second. Used to determine the number of calculations a CPU can perform where each hertz is know
as a "clock cycle" and within each "clock cycle", a particular number of calculations can be performed.
HPC, High Performance Computing
Computing beyond the expected performance of an individual desktop machine. Typically associated with parallel clusters.
HPEC, High Performance Embedded Computing
Similar to HPC, but using embedded systems instead of desktop compute nodes.
interconnect fabric
A hardware means of establishing communication between nodes in a cluster.
interconnect protocol
The software that governs how the interconnect fabric is used to maintain communication between nodes in a cluster.
ISP, Internet Service Provider
The company that a home or office 'dials into' to gain access to the internet. ISPs typically provide email and web accounts as well.
Linux: "L"inus + "UNIX'
Developed by Linus Torvald, a Helsinki, Finland grad student who developed an 'open-source' OS as a school project. He gave away his
work, via the internet, asking for assistance in its development. 11 years later, it is the fastest growing OS in the world.
Mb: Megabits
One million bits of data where 8 bits equals 1 byte --see MB.
MB: Megabytes
One million bytes of data where each byte is comprised of 8 bits. Used to determine the size of a hard drive or RAM.
MHz, Megahertz
One million cycles per second --see GHz.
motherboard
Also known as the "mainboard", the motherboard is the PCB (Printed Circuit Board) on which the components of a computer are fixed with
solder. The motherboard provides the medium by which the transfer of data can occur between the CPU, RAM, drive, and any input/output
devices.
node
An individual computer in a cluster. This can be a desktop computer, a 'blade', or in some contexts, an individual processor (CPU)
OEM, Original Equipment Manufacturer
The manufacturer of a product.
parallel computing
More than one computer (node) interconnected via a communications medium that enables them to work collectively with computational
problems shared evenly over the incorporated nodes. Parallel clusters typically incorporate tightly coupled (high speed),
similar-performance nodes in close proximity to one another.
PCI, Peripheral Component Interconnect
A standard for connecting peripherals to a personal computer, designed by Intel and released around Autumn 1993. PCI is supported by
most major manufacturers including Apple Computer. PCI is the communication protocol by which the CPU 'talks' to the cache, RAM, drives,
and all input/output devices. It's an understood language and is governed by an assumed speed.
porting apps
The effort required in modification to code before the final cross compile.
PowerPC
A chip architecture manufactured by IBM and Motorola (IBM trademark licensed by Motorola). Unique in the high performance, low
power consumption, and RISC foundation. G5 represents the fifth generation.
rack, rackmount
The ability to place a computer in an industry standard 19" rack.
RAM, Random Access Memory
Temporary storage of applications and data while a computer is powered on. RAM is used when cache is filled. RAM can (with desktop
machines) be replace and increased by users.
RISC, Reduced Instruction Set Computer
Developed at IBM, reduced instruction-set computing (RISC) is based on studies showing that the simpler computer instructions are the
ones most frequently performed. Traditionally, processors have been designed to accommodate the more complex instructions as well. RISC
performs the more complex instructions using combinations of simple instructions. The timing for the processor can then be based on
simpler and faster operations, enabling the microprocessor to perform more instructions for a given clock speed.
U
Defines the vertical 'thickness' of a rack mount enclosure = 1.75".
video rendering
Using a computer to generate a movie, frame-by-frame, given a set of data and instructions as to how to generate the image from the
data.
VME, Versa Module Europa
A flexible open-ended bus system which makes use of the Eurocard standard. VME was introduced in 1981 by Motorola, Philips, Thompson,
and Mostek. It was intended to be a flexible environment supporting a variety of computing intensive tasks, and has become a rather
popular protocol in the computer industry. It is defined by the IEEE standard 1014-1987.
x86
Derived from the original Intel-based "80x86" (as in 286, 386, 486) series CPUs. A common abbreviation for any of the Intel CPUs or
Intel compatibles/competitors, such as Cyrix and AMD.
HPC Intro |
Price/Performance |
IBM 970 |
AltiVec |
Embedded Systems |
Geek Glossary
|
