[G3]-TechNews : Home| RSS | Atom | MSN | WAP

Article: Step-by-Step Guide: How to set up a VPN
Article: Download: Microsoft Monad (Beta)
Article: Building a 64-Bit Multimedia Workstation
Article: Coming Soon to Windows: The Microsoft Shell (MSH)
Article: How to Reset Win NT / 2000 / XP Administrator Password
Article: The Technology Behind Dual Core CPUs
Article: How-To: Wireless Network Security
Article: Article: PCI Express - technology backgrounder
Article: Tutorial: Access Hidden Files on Your iPod
Article: Troubleshooting drivers with XP's hidden Driver Verifier Manager
Article: How to Make a 5 in 1 Network Cable
Article: Comparison - Blu-ray & HD DVD
Article: Beginners Guides Linux : Part 1 | Part 2 | Part 3
Article: How To Crack WEP (Wired Equivalent Privacy)
Article: Email Addresses Spoofing.
Link: Free PHP ebook
Link: FREE ASP.NET books and eLearning course
Link: Free registration code for Opera 8.
Invitations: Gazzag (Here) | Yahoo! 360 (Here)  | Orkut (Here)

Article: PCI Express - technology backgrounder
Contributed by: G3nu1n3, at 5/30/2005 07:40:00 PM.

Article By: James Morris
Source: Link

Increasing numbers of motherboards carry PCI Express slots. James Morris explains why this new serial bus standard is replacing the older PCI parallel bus and how the move benefits video editors

When the PCI bus first took over from ISA - about the time that Intel’s original Pentium processor arrived - it seemed to offer enough bandwidth for a good while to come.

Well, although that did turn out to be the case, today, PCI is no longer able to handle the throughput of the latest high-performance cards and programs. Processing sound and video place considerable strains on PCI and are among the tasks that its successor, PCI Express, are intended to handle better.

PCI offered far greater bandwidth than ISA. The most prevalent form of PCI has a 32-bit/33MHz bus and offers a theoretical maximum throughput of 133MByte/sec. Standard ISA, in contrast, is a 16-bit/8MHz bus capable of a theoretical 8MByte/sec of bandwidth, but much less in practice.

When processor speeds were measured only in a few hundred megahertz, PCI offered plenty to be getting on with. But once processors became able to handle data more quickly, the PCI bus started holding back some potential new capabilities.

With 3D acceleration beginning to take off and stoke the fires of the PC gaming market, greater graphics bandwidth to CPU and to memory was required. So the AGP bus was born.

This offered a direct 66MHz/32-bit connection to the system bus via the GART (graphics address remapping table), and graphics cards were the first mainstream adapters to vacate the PCI bus to avoid congestion.

The basic AGP specification doubles the bandwidth of PCI to 266MByte/sec, and successive levels of clock-doubling technology eventually increased this to 2,133MByte/sec for AGP 8x.

A comparison of the relative bandwidths available to the most popular peripheral adapter bus standards

But graphics cards aren't the only adaptors needing more bandwidth than is available over PCI. Gigabit Ethernet has a maximum throughput of 125MByte/sec in each direction, so can max out PCI on its own.

That's why Intel’s integrated Gigabit Ethernet controllers already use Communications Streaming Architecture - a non-PCI bus connection found on numerous workstation and server motherboard chipsets, including 865 and 875.

Some storage add-ons also require more bandwidth than PCI can provide. Even the previous generation of SCSI (Ultra160) had a theoretical maximum of 160MByte/sec, while the latest, Ultra320, has doubled this – far in excess of what PCI can deliver.

The Serial ATA disk-drive channel is potentially too fast for PCI as well. A single-drive Sata channel on its own is rated at 150MByte/sec, and most Sata PCI adapter cards have two or more channels.

However, no current hard disk requires anywhere near this much bandwidth. Even Western Digital’s Raptor, a massively fast 10,000rpm enterprise Sata drive, offers average throughput of "only" 65MByte/sec - less than half of Sata's theoretical maximum.

For video, the amount of bandwidth required is a problem, too. Although compressed DV only requires 3.6MByte/sec, the video doesn’t remain compressed during editing.

Most editing systems have to uncompress the video to single frames before applying any kind of effects processing. This brings the data rate up to around 20MByte/sec per video stream – and editing systems intended to handle uncompressed standard-definition footage start off with this throughput requirement.

And, of course, it's normal to be editing multiple streams of video, not just one. A real-time editing system using a PCI adapter with hardware effects will be passing numerous streams of uncompressed video back and forth across the PCI bus, quickly saturating it.

The arrival of HDV camcorders might mean greatly improved picture resolution but it also compounds the bandwidth problem. Uncompressed, the video in each stream of 25-frames-per-second 720p HDV requires a bandwidth of about 35MByte/sec – and that ignores audio, timecode data, flags and other stuff that also needs to be punted around. With 25fps/50-field 1,080i HDV, the figure rises to 58MByte/sec – and, remember, this is for a single stream.

PCI has already been superseded for some time in professional servers where the need arose earlier and return on investment was most immediate. The solution was a version of PCI using a 64-bit slot that’s otherwise the same as PCI, and there’s also PCI-X – not to be confused with PCI Express.

Although the interface for PCI-X is also 64-bit, rather than 32-bit, it operates at a higher frequency. The current version of PCI-X supports slots running at 66MHz, 100MHz or 133MHz - so bandwidth ranges from 533MByte/sec to 1,066MByte/sec.

Most PCI-X equipped motherboards will only have one or two slots running at the top frequencies, perhaps paired with a couple that are lower-rated. Know, too, that most PCI-X slots also support standard PCI cards, and many PCI-X cards will also work in standard PCI slots.

The added expense of supporting the PCI-X interface means that only adapters that need the bandwidth are offered as PCI-X. This includes Gigabit and 10 Gigabit Ethernet controllers, plus SCSI and other high-performance storage adapters, notably multi-port Sata.

In addition, some HDV-editing cards are available in PCI-X, the most talked about being the card that comes with Canopus's Edius NX for HDV combined hardware/software package.

Despite the arrival of PCI Express, PCI-X is continuing to develop. Version 2.0 is already set, offering further increased frequencies up to 533MHz using clock doubling or quadrupling. This will boost throughput to nearly 4.3GByte/sec.

PCI-X is now firmly entrenched in the enterprise market, and will remain so. But, it’s expensive to implement, and the connector is bulky - areas where PCI Express scores. Despite the name, PCI Express is not strictly an evolution of PCI. Whereas PCI and PCI-X are essentially parallel data connections, PCI Express uses serial technology.

All devices on the PCI or PCI-X bus share the bandwidth, but each PCI Express device has a dedicated point-to-point connection to the controller, so it gets all the bandwidth available for its own specific use.

In its basic form, PCI Express uses a four-wire system that runs at 2.5Gbit/sec in each direction. In duplex mode, this four-wire lane allows around 250MByte/sec of data throughput each way.

And this is dedicated bandwidth, so PCI Express is not only almost twice as fast in one direction as PCI, it's unaffected by the number of PCI Express devices connected to the system. Then add PCI Express's duplex mode into the equation, and it's clear that overall performance is another order of magnitude faster than PCI.

Lanes can also be grouped in parallel to increase bandwidth. The specification currently allows for up to 32 lanes to be ganged together, but shipping implementations use a maximum of only 16 lanes. Even so, this equates to 4GByte/sec in each direction - twice as fast as AGP 8x.

Hence, PCI Express 16x is already being offered as the replacement interface for graphics cards. There is now a decent selection of PCI Express graphics card available based on Nvidia and ATI chipsets, and Matrox has come to market, too.

But PCI Express isn’t just about increased bandwidth. It also offers better power management, including native hot-plug support. And, although the physical interface is not backwards compatible, the driver architecture is. So PCI products can rapidly be moved over to PCI Express with little rewriting of software needed to make them work.
Since PCI is a parallel technology and PCI to PCI Express is serial, there's no such thing as an adaptor for PCI-to-PCI-Express. To obtain PCI Express connectivity, a new-generation motherboard will be required.

Intel-based motherboards were the first to offer PCI Express, typically using the Intel 915 and 925 chipsets and running only Intel processors.

Most Intel-based motherboards have a single 16-lane PCI Express slot for graphics, along with two or three of the single-lane variety for other peripherals, plus a couple of PCI slots for backward compatibility.

However, upgraders should bear in mind that motherboards with these Intel chipsets also have a new 775-pin processor socket and, in many cases, support a new kind of memory called DDR2, rather than the current DDR.

A couple of recent chipsets for Intel’s Xeon workstation and server processor also support PCI Express. The one most relevant to video editors is Intel’s E7525, codenamed Tumwater, which is aimed at desktops.

This supports the new 800MHz Front Side Bus Xeons and DDR2 memory. But it also has fully fledged PCI Express support and motherboards are starting to arrive from companies other than Intel. Supermicro’s X6DAE-G2 is one and has two PCI Express slots. One is 16-lane and the other is four-lane but the same size as the 16-lane to allow a second graphics card to be used .

Editors who prefer AMD processors can now also take advantage of PCI Express thanks to the arrival of Nvidia's Nforce 4 motherboard chipset for AMD processors. This is compatible with all existing Athlon 64 and Athlon 64 FX processors, plus the new Socket 939 Athlon 64s, depending on motherboard design.

Nforce 4 supports two 16-lane PCI Express slots, offering the ability to run two fast PCI Express graphics cards side-by-side – which is great for gamers and may turn out to be no less useful for video editors.

DFI was one of the first to incorporate Nforce 4 and its LanParty UT nF4 SLI-D PCI Express motherboard has been well received by reviewers. It offers four PCI Express slots (two of them x16, one x4 and one x1), plus two standard PCI slots, and can run two fast graphics cards at once, and use a low-cost AMD Sempron processor or an Athlon 64 FX or 64 CPU. The first shipping PCI Express cards were 3D graphics accelerators such as Nvidia’s nForce 6800 GT and ATI’s Radeon X800. These are mostly of interest to gamers, but Matrox gave a convincing demonstration at VideoForum 2005 that its own first PCI Express graphics card, the Parhelia APVe, would appeal to those wanting to edit HD or HDV footage.

Matrox paired the triple-head PCI Express card with Ulead's MediaStudio Pro 7 software on a PC with two monitors – and the card was also feeding out full HD to an HD broadcast monitor.

Pinnacle's Liquid Edition editing program can take advantage of virtually any PCI Express graphics card to increase the video processing bandwidth, as it uses the graphics chip for real-time rendering of some effects.

But, even lower-bandwidth PCI Express slots intended for non-graphics cards offer advantages to video professionals and enthusiasts. In its 1x implementation, PCI Express removes one of the barriers to real-time editing. Processors have become fast enough to mix multiple streams of video and, with PCI Express, there’s now a dedicated path capable of at least 250MByte/sec in either direction to any hardware editing add-on. That’s enough for more than ten streams of uncompressed TV-resolution SD video.

PCI Express slots are now fitted to a range of PCs costing less than £1,000, so the groundwork has already been laid for the introduction of a new generation of real-time video editing cards – though it remains uncertain which video editing hardware firms will take this route. In the prosumer arena, our money's on Canopus and Matrox – the arrival of Liquid Edition 6 showed that, in the prosumer arena, Pinnacle is planning to leave all the work to the graphics card and one or more CPU

Source: Link

To Read MOST UPDATED News Items browse to HOME page.


Post a Comment

<< Home

[G3]-TechNews : Home| RSS | Atom | MSN | WAP

Archives :

- Monthly Archives :

- Post Count: 1,783 before June 1, 2005. (Since: October 26, 2004)