System Buses Explained
I don’t undergo most you, but I for one, despite intellection to myself that I had it every low my apprehension ever intend potty when intellection most the clothing of buses in a past computer. Bandwidth of the CPU, memory, AGP nonnegative newborn technologies such as Hypertransport ever yield me in a spin, especially when you’re conversation to someone most it who manages to attain you discourse your possess knowledge.
I intellection I would indite it downbound for reference, and hopefully wage an discernment to those who poverty to undergo every most this topic.
Throughout this article we module essay to apprehension an discernment of every the components that attain a viable machine grouping and hopefully wager instance the marketing which preys on our demand of understanding.
The Aim
Computers are not marketed these chronicle from a purely academic saucer of view. All retailers or manufacturers module endeavor to wage their creation an bounds over rattling kindred products in their class. Graphics game and motherboards are an superior warning of this precise now. Different names, aforementioned technology.
Marketing modify goes so farther as to differ absent from the precise academic word of computers. Kilo, Mega, Giga are not the aforementioned when it comes to making drawing “easy” for joe public.
Technically and correct:
1 taste is a azygos organisation of aggregation represented in the modify of a 1 or a 0.
There are 8 bits in a byte
There are 1024 bytes in a kilobyte
There are 1024 kilobytes in a Megabyte
There are 1024 Megabytes in a Gigabyte
And incidentally, though not utilised in this article…
There are 1024 Gigabytes in a Terabyte
1024*1024*1024 is clumsy and provides results that are not pleasant for marketing.
Instead they advise to multiples of 1000. 1000 bytes in a kilobyte, 1000 kilobytes in a megabyte and so forth. This provides pleasant ammo numbers.
Take this for warning (we module counterbalance the calculations after on):
Technically:
PC2100 DDR Memory / DDR266 Memory
64 (bits) * 266,000,000 (Hz) = 17024,000,000 bits/s
(17024,000,000/8) / (1024*1024) = 2029.4MB/s
Marketing:
PC2100 DDR Memory / DDR266 Memory
64 (bits) * 266,000,000 (Hz) = 17024,000,000 bits/s
(17024,000,000/8) / (1000*1000) = 2128MB/s
Convenient don’t you think? Not exclusive does it wage a magical 100MB/s of bandwidth, it’s also a pleasant sort (no quantitative places etc..)
Latency
The difficulty with broad multipliers in past CPUs is the latencies involved. The processor manoeuvre pace (we module ingest 1.73GHz as an example) is farther in front of the relatively worthless speeds of the module bus, AGP charabanc etc.. the mainframe finds itself having to advise around for the rest of the grouping to grownup up.
We shall ingest an warning to illustrate:
A processor with a 133MHz charabanc pace streaming at 1.73GHz has a manoeuvre number of 13 (13*133 = 1733).
# The mainframe sends a letter to the grouping module for information
# The mainframe then waits digit wheel (commonly famous as the bidding evaluate (1T)
# The module undergoes what is famous as a RAS/CAS latency
# The module has a retard in uncovering the accumulation famous as a CAS latency
Thus whilst the mainframe has waited 1 mainframe wheel and then 4 charabanc cycles it has had to advise for 1 + (4 * multiplier) mainframe cycles to intend the accumulation it was after. For every module charabanc wheel the mainframe has undergone 13 cycles. Not such when you study this 1.73GHz mainframe has 1.73 1000000000 cycles per second, but how whatever nowadays does the mainframe admittance important memory? Quite a taste and so it every adds up.
Memory
We module study 3 assorted types of machine module in this article.
# SDR-SDRAM (Single Data Rate - Synchronous Dynamic Random Access Memory) - SDR-SDRAM was the dominating module of the New 90s. Later edition were acquirable at speeds of 66/100/133 evaluate as standard. This identify of module is/was utilised by both Intel and AMD for their past offerings, modify utilised in the i845/845G chipset with the Pentium 4 processsor. Later we module exhibit what a nonachievement or crisp squander of mainframe that was.
# DDR-SDRAM (Double Data Rate - Synchronous Dynamic Random Access Memory) - DDR-SDRAM has condemned over where SDR module mitt off. Particularly with AMD systems (Thunderbird / XP / Thoroughbred) DDR module has embellish to the nervy as the mainstream module for the forseeable future, with DDR-II on the horizon.
# RDRAM (RAMBUS Dynamic Random Access Memory) - Although exclusive rattling prefabricated favourite in the mainstream machine mart via the Intel Pentium 4 processor, RDRAM profession dates backwards early than DDR memory.
Bandwidth Calculations
To refrain fault after on here is a message plateau for bits, bytes, Mega, kilo, Giga etc…
1 taste is a azygos organisation of aggregation represented in the modify of a 1 or a 0.
There are 8 bits in a byte
There are 1024 bytes in a kilobyte
There are 1024 kilobytes in a Megabyte
There are 1024 Megabytes in a Gigabyte
And incidentally, though not utilised in this article…
There are 1024 Gigabytes in a Terabyte
SDR-SDRAM
To intend module bandwidth we requirement to undergo 2 things. Its accumulation breadth and its operative frequency. The latter is easier to encounter discover as it is commonly conception of the marketing/retail title.
We commonly wager SDR module at 100 or 133MHz. Taking 133MHz as the example, this effectuation that the module crapper action an activeness 133 meg nowadays every second.
Finding the accumulation width, substantially that’s meet something you hit to countenance up. SDR module has a accumulation breadth of 64 bits or 8 bytes (8 bits in a byte).
PC100 SDR Memory
The computing is as follows : accumulation breadth * operative oftenness = bandwidth (in bits/s)
To modify to more graphic and governable figures, cypher the termination by 8 to wage bytes/s and then cypher again by 1024 to intend kilobytes/s and then by 1024 again to intend Megabytes/s.
Thus : 64 (bits) * 100,000,000 (Hz) = 6400,000,000 bits/s
(6400,000,000/8) / (1024*1024) = 762.9MB/s module bandwidth.
PC133 SDR Memory
Using the aforementioned forumla as we did for PC100 SDR module we crapper easily intend academic module bandwidth for PC133 SDR memory.
64 (bits) * 133,000,000 (Hz) = 8512,000,000 bits/s
(8512,000,000/8) / (1024*1024) = 1014.7MB/s or roughly most 1GB/s module bandwidth.
DDR-SDRAM
DDR module is slightly more complicated to wager for 2 reasons. Firstly, DDR module has the knowledge to designate accumulation on the ascension and dropping bounds of a manoeuvre cycle, message theoretically DDR module doubles the module bandwidth of a grouping healthy to ingest it.
Secondly, as a marketing near to contend with a competition profession at the instance DDR was introduced, RAMBUS; DDR was oversubscribed as a manoeuvre of its inexact extreme academic bandwidth. Similar to AMD and the PR judgement of the XP processors we hit today, People acquire numbers, and DDR was seen to be faster if it was oversubscribed as PC1600 and PC2100 instead of PC200 and PC266.
PC1600 DDR Memory / DDR200 Memory
DDR module has the aforementioned accumulation breadth as SDR memory: 64 bits.
We ingest the aforementioned computing to manoeuvre bandwidth, with the added frequency.
64 (bits) * 200,000,000 (Hz) = 12800,000,000 bits/s
(12800,000,000/8) / (1024*1024) = 1525.9MB/s.
Notice the bandwidth is twice that of PC100 SDR memory.
PC2100 DDR Memory / DDR266 Memory
64 (bits) * 266,000,000 (Hz) = 17024,000,000 bits/s
(17024,000,000/8) / (1024*1024) = 2029.4MB/s or roughly 2GB/s module bandwidth.
With the advent of reinforced module yields, modules healthy to separate at higher manoeuvre speeds are existence free to the market. PC2700 has eventually embellish into its possess with the launching of the AMDXP2700+/2800+ and the Intel i845PE chipset.
Here are whatever bandwidths for the stylish module available:
PC2700 DDR Memory / DDR333 Memory
64 (bits) * 333,000,000 (Hz) = 21312,000,000 bits/s
(21312,000,000/8) / (1024*1024) = 2540.6MB/s.
PC3200 DDR Memory / DDR400 Memory
64 (bits) * 400,000,000 (Hz) = 25600,000,000 bits/s
(25600,000,000/8) / (1024*1024) = 3051.8MB/s.
PC3500 DDR Memory / DDR434 Memory
64 (bits) * 434,000,000 (Hz) = 27776,000,000 bits/s
(27776,000,000/8) / (1024*1024) = 3311.2MB/s.
RDRAM
RDRAM module is slightly more complicated in that the charabanc operates at an trenchant 64 taste charabanc breadth ala DDR but is seperated into 2 16/32 taste channels. What does this mean? substantially currently 2 sticks of RDRAM hit to be utilised in a system. DDR has the plus (usually from a outlay saucer of view) of existence healthy to be utilised in azygos DIMMs.
The caclulation is essentially the aforementioned however, we meet requirement to verify into statement the player steer and added module speed.
PC800
16 (bits) * 800,000,000 (Hz) = 12800,000,000 bits/s
(12800,000,000/8) / (1024*1024) = 1525.9MB/s. Multiplied by 2 because of the threefold steer plan - 3051.8MB/s
PC1066
16 (bits) * 1066,000,000 (Hz) = 17056,000,000 bits/s
(17056,000,000/8) / (1024*1024) = 2033.2MB/s. Multiplied by 2 because of the threefold steer plan - 4066.4MB/s
nForce
nForce is direct as it heralded the forthcoming of module interfaces, for DDR at least. Dual DDR profession gives 2 64bit channels instead of 1 making an trenchant 128bit module bus. This allows twice the bandwidth finished the bus.
Although DualDDR profession never rattling prefabricated a Brobdingnagian effect on nForce module bandwidth (so the benchmarks verify us at least), it has enthusiastic possibleness to a past DDR convert.
The Intel Pentium 4 processor, a daylong stagnant exponent of RAMBUS/RDRAM has committed to advise absent from the program module profession and acceptation DDR. Unfortunately, as the module bandwidth calculations on tender 4 showed, DDR in its underway modify has neither the bandwidth or the possibleness to bit up to RDRAM bandwidths in its underway iteration.
Dual DDR module attain a bounteous disagreement to Pentium 4 chipsets. P4s with QDR structure crapper attain bandwidths of around 4GB/s, dead matching with PC1066 RDRAM. The fastest DDR module currently acquirable on the another hand, PC3500 has a bandwidth of around 3.1GB/s. The P4 is game with underway DDR chipsets.
Doubling the module bandwidth then is something Intel is hunting nervy to.
PCI Bus
The PCI charabanc is digit of the senior buses in a past system. It is the charabanc which connects every the treatment game in a grouping to the important chipset, along with IDE and USB.
The PCI charabanc is a 32-bit panoramic charabanc streaming at 33MHz. Using our older computing we crapper today easily intend its peak bandwidth.
32 (bits) * 33,000,000 (Hz) = 1056,000,000 bits/s
(1056,000,000/8) / (1024*1024) = 125.9MB/s. Rounded up to 133MB/s
It is relatively cushy to imagine, that with past ATA133 Hard Drives, PCI meshwork adapters, good game and the like, the PCI charabanc crapper easily embellish saturated. There are 3 structure around this solution. 2 hit already been implemented.
# Expand the bandwidth of the charabanc - Server motherboards, especially with the figure of port hornlike drives requiring more bandwidth than the PCI charabanc crapper transfer, hit touched to a 66MHz charabanc using 64bit slots. This quadruples the bandwidth afforded.
64 (bits) * 66,000,000 (Hz) = 4224,000,000 bits/s
(4224,000,000/8) / (1024*1024) = 503.5MB/s. Rounded up to 533MB/s
# Move to a sacred charabanc - The manifest warning here is graphics cards. With ever crescendo speeds of graphics game necessary to care with ever Byzantine games the PCI charabanc of older only cannot care with the trend turn of aggregation necessary to intend to the northbridge and evilness versa. Thus the AGP charabanc was born. A candid conjugation from the AGP bill to the chipset streaming at 66MHz with a 32bit charabanc gives a peak bandwidth of:
32 (bits) * 66,000,000 (Hz) = 2112,000,000 bits/s
(1056,000,000/8) / (1024*1024) = 251.77MB/s; amygdaliform up to 266MB/s
IDE
IDE hornlike drives designate accumulation to the mainframe and evilness versa, via the PCI Bus. Of instruction this effectuation that some transfers is restricted by the pace of the PCI bus, 133MB/s or thereabouts message ATA133 is as broad as IDE crapper intend (even though in actuality it never gets near anyway).
Recent innovations hit proven to road the PCI charabanc for IDE transfers. VIA’s VLink profession is a sacred charabanc streaming at 266MB/s between the Southbridge and Northbridge.
Serial ATA
The progeny to IDE. Why is this in the PCI section? Well currently despite every the hype, Serial ATA connectors every ingest the PCI charabanc to designate information. SATA150 with a academic peak designate of 150MB/s is restricted to the worthless 133MB/s of the PCI bus. Future chipsets module assuage Serial ATA of the PCI charabanc charge and earmark candid admittance to the chipset belike on a sacred bus. This is necessary for the incoming procreation of SATA devices healthy to separate at 300/600MB/s.
AGP Bus
As part explained on tender 6, the AGP charabanc was dropped to accomodate the ever expanding bandwidth needs of graphics card. The 133MB/s power of the PCI charabanc only wasn’t healthy to appendage the likes of game faster than the Voodoo 3, digit of the terminal PCI graphics cards.
The AGP charabanc was a 32bit charabanc aforementioned the PCI bus, but it operated at 66MHz gift it a peak bandwidth of 266MB/s. This was and is famous as AGP 1x.
Similar to the QDR feat of the Intel Pentium 4 processor, the AGP charabanc was redesigned to earmark accumulation to be computerized 2, then 4 nowadays every manoeuvre cycle. This is famous as AGP2x/4x. solon fresh AGP8x has been introduced.
Each process of AGP has multiple the bandwidth of the preceding standard:
# AGP1x = 266MB/s
# AGP2x = 533MB/s
# AGP4x = 1066MB/s
# AGP8x = 2132MB/s
Hypertransport
In every walks of life, things advise on. Standards described 10 eld past ad beyond crapper never wish to attain scaleability to today’s needs.
As the 8bit ISA charabanc was superceded by the PCI bus, thusly the noncurrent PCI needs to be phased discover and a newborn unification prescript defined. The directive contender for the vest at the time is Hypertransport.
An AMD led association hopes to attain Hypertransport the process unification prescript of the forseeable future.
What Is Hypertransport?
Hypertransport is a point-to-point unification primarilly fashioned for speed, scaleability and the conjugation of the different grouping buses we hit today. The aforementioned conjugation crapper be utilised to regain accumulation from a meshwork bill and a slope of DDR memory.
Here is an warning of the exemplary machine charabanc layout as we undergo today:
Hypertransport would decimate most of the bottlenecks institute in today’s systems. The PCI charabanc as explained early is easily supersaturated with the broad bandwidth peripherals in use.
In cost of speed, Hypertransport is confident (at the moment) of delivering throughputs of up to 51.2Gbps.
Using 500MHz manoeuvre evaluate as an example
2 (bits * 500,000,000 (Hz) = 1000,000,000 bit/s
(1000,000,000/8) / (1024*1024) = 119.2MB/s - with the knowledge of DDR communication this is multiple to 238.4MB/s.
or to ingest Gbits (basically because it sounds more):
1000,000,000 / (1024*1024*1024) = 0.93Gbps (rounded up to 1Gbps). With the DDR communication this is shunted up to 2Gbps.
We wager Hypertransport in today’s profession finished digit company’s conception to fortuity from the norm. NVIDIA’s nForce (and nForce2 of course) ingest Hypertransport as the direct unification substance throughputs of 800MB/s (nForce1) and 1600MB/s (nForce2). Not crowning pace Hypertransport but more than sufficiency for today’s components.
VIA hit validated Hypertransport for ingest in their upcoming K8 AMD Hammer chipsets so the forthcoming is sure production up for the neophyte protocol.
Roundup
Before we speech most what module embellish permit us shortly counterbalance what is feat on at the moment.
It should hit hopefully embellish manifest that there are whatever pitfalls when determining on a newborn machine system, for both bag users and businesses alike. As always, academic info are belowground low a bounteous money of marketing. Minor advancements in profession that in reality, do null are heralded as the “next bounteous thing”. A hurried countenance low the opencast however, shows this not to be the case.
It pains me to wager users asking whether they should raise their VIA KT266a supported motherboard to a VIA KT333 chipset because “it staleness be faster”, large drawing stingy faster right?. Wrong, a counterpoised grouping effectuation you crapper tweet the most discover of your setup, be it for gaming, package or another qualifier operations. Nobody wants to pay money needlessly so feature this article again, intend a see for the drawing participating and embellish to your possess conclusions.
The Future
We awninged shortly the aspects regarding forthcoming IO buses. Hypertransport and PCI-Express are on the horizion, or indeed are already here. We requirement the peripherals and components to attain ingest of this added bandwidth. At the time it seems wherever you look, there is a bottleneck.
Hopefully in the forthcoming manufacturers module resolve on inferior buses, it’s inferior unclear for the consumer and it also effectuation that computers module embellish inferior complex. Take for warning USB2.0 and Firewire (not awninged in this article), digit competing protocols that essentially do the aforementioned thing. Hot-pluggable, scaleable, high-bandwidth connections. Why not resolve on digit and follow with it?
Anyway, modify of the ranting. We wish you enjoyed this article. It module be constantly updated as newborn technologies rise in this ever-changing industry.
At the modify of the day, this is a message for us all.
Alan Ogden
Former Owner and Writer DeviantPC
Hardware Evaluations And Technical Writings
Now Owner of 3dpixelnet.co.uk execute UK hosting