HyperTransport, Lightning Data Transport
Implemented on AMD K8 series processors, this is, for all intents and purposes a bidirectional FSB but is clocked significantly higher. The base clock for the HyperTransport is 200MHz and the multiplier's go up to five. Factor in the principle of DDR and you get a maximum net
effective hypertransport clock of 1000MHz (2000DDR). Naturally, marketers will often write this as FSB1600 or FSB2000 however this is incorrect as the actual clock speed is still half of that (marketers forget that DDR only means "effective") and that HyperTransport and Front-Side-Bus are mutually exclusive. Intel platforms do not have support for HyperTransport.
PCI-X, PCIe, PCIx1, PCIx16, PCI, AGP etc- PCI. This is the most primative of the bunch. It's a 32bit, 33MHz interface with a throughput capacity of 133MB/s. (33Mhz x 32bits / 8bits-per-byte = 133MB/s)
- 64bit PCI. This is tha 64bit implementation of PCI and has a throughput capacity of 266MB/s and 533MB/s. (33MHz or MHz x 64bits / 8bits-per-byte = 266MB/s or 533MB/s)
- PCI-X. PCI-X is a backwards compatible (compatible to PCI) interface which just essentially offers an increased throughput capacity. Currently found on server boards, its throughput capacity is 1.06GB/s, 2.1GB/s and 4.3GB/s (for PCI-X, PCI-X 266 and PCI-X 533 respectively)
- PCI Express. Formerly known as 3GIO, also denoted as PCIe, this is an extremely highspeed serialized interface. The individual serial lanes can be grouped together and when done as such, they are denoted as PCIxN where N is the number of lanes that are grouped together. Each lane is capable of 250MB/s
- AGP. A 66Mhz 64bit variant of the PCI bus, AGP was designed purely to facilitate the bandwidth requirements of newer and newer videocards. Several revisions of AGP are available with throughput capacities of 266MB/s, 533MB/s, 1066MB/s, and 2133MB/s respectively
Hyperthreading
Hyperthreading is an innovation made by Intel designed to facilitate and assist the execution of multiple threads. For most consumer applications this translates to an
improved multitasking experience however marketers tend to like to "suggest" that it is like having twice the processing power -- which it is not. It should also be noted that Hyperthreading really only shines through when multiple CPU-intensive tasks are executing simultaneously.
Dual/Multi Core
A recent craze, Dual and multi core processors are, for all intents and purposes, "two processors inside one convenient package". Each processor will have independent access to cache and stackspace; by having multiple cores the processor as a whole can deal with multiple threads of execution (i.e., multiple applications, multiple instructions etc) without any [significant/measureable] overhead (i.e., Dual Core processors are what marketers hype-up HyperThreading to be)
OPN
Order Product Number, this is the funky code number written on processors that identify its make and model unqiuely. For AMD processors, you can read this right off the chip while for Intel chips, identification software is required to read this. While this kind of information isn't entirely useful to the majority of consumers out there,
- For those buying AMD chips, it helps to be able to verify that you are indeed getting the processor you want
- For overclockers and such you can hunt down information about your specific core etc
Performance Rating (PR)
In recent times, originally introduced by AMD (with their AthlonXP processors) as a means of expressing, as the name suggests, the performance of a given processor. Naturally consumers
incorrectly drew a parallel between the
PR of the AthlonXP processor and some Pentium4 processors (granted this was one of the reasons AMD utilized such a
PR schema and it worked brilliantly).
Of course AMD isn't the only culprit of this "confusion" (granted it's only confusing because people insist on trying to form a direct correlation between AMD's
PR and Intel's Pentium-Rating, also denoted
PR) but Intel has also shifted to using an indexed scale for processor designation starting with their S775 processors.
Cache
Cache is essentially, a chunk of very high speed memory available exclusively to the processor for storing very commonly accessed data. When the CPU is looking for an instruction/data and it is found in the cache, that is known as a
cache hit. There are several types of cache:
- L1 Data and L1 Trace (collectively known as L1), this type of cache is usually quite small (small on AMD processors and tiny on Intel) and is the first place the CPU looks for information. Due to its small size and the variance of instructions/data available, cache hits in L1 are not a performance bottleneck
- L2 cache is the most marketed cache and is big enough to actually influence the performance of the processor with budget processors cheaping out on the amount of L2 cache available. Although "the more L2 cache available, the better", also note that the more cache they cram onto a processor, the slower and less efficient it becomes and the more the cache becomes dependent on optimizations designed to improve cache hits
- L3 cache is essentially, another type of L2 (but slower still) and is reserved for storing a (relative) whack of data somewhat close to the CPU. L3 cache was available only on select models of the Intel Pentium4 Extreme Edition and has since been discontinued in favor of a larger L2 cache.
Pipeline
Much like an assembly line, processors breakdown the execution of an instructions into stages:
- The more stages available, the simpler each stage becomes and the faster each stage executes. The downside is however, if nearing the end of the 'assembly line' something "wrong" happens, the entire thing has to start over (thus the processor becomes heavily dependent on predictors and optimizations to prevent this from happening). Also, the more stages available, the faster the overall clockspeed can be.
- For a processor with few[er] stages in the pipeline, the processor is far more efficient and is able to complete the same amount of work with fewer clock cycles and also the other side of the worst-case pipeline-scenario: if something "wrong" happens near the end of the pipeline, fewer stages need to be re-traversed. The only downside is that net clockspeed is limited (because with fewer individual stages, each stage is more complex than the longer-pipeline variant and you can only rush it so fast)
Cooln'Quiet, SpeedStep/EIST
Originally developed for mobile platforms, Intel and AMD realized that running a processor at full blast 24x7 meant that (a) the processor tended to overheat and that power-consumption was maxed out. Thus several approaches were developed to deal with this (and then transitioned to the desktop market). All these approaches do essentially the same thing: when CPU power isn't required, the processor throttles itself down to reduce power consumption and heat production (it automatically throttles itself back to full speed when that speed is required)
64bit, AMD64, EM64T
Generically this describes processors capable of operating in 64bit mode (using a 64bit operating system etc). As the names suggest, AMD64 refers to the 64bit extensions (extensions of the x86 set of assembly level instructions) and EM64T is the Intel equivalent. Naturally, one can only make use of 64bit mode with 64bit hardware and a 64bit operating system
NX/XD Support
AMD's No-eXecute and Intel's eXecute-Disable refer to equivalent features that are designed to, with support from an operating system like WindowsXP (with SP2), limit the effectiveness of buffer overflow errors.
Centrino
Commonly mistaken as an actual processor or something, Centrino is just a wireless chipset
associated with two mobile processors: the Intel Pentium M and the Celeron M.