Articles: Graphics
 

Bookmark and Share

(68) 
Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 ]

PCI Express Is Coming! (To Your System, Too)

This review would be incomplete without any mentioning the products that support the new data-transfer standard, actively promoted by Intel. PCI Express is a next-generation bus, born to replace PCI and AGP interfaces.

The PCI Express bus does offer numerous advantages over its predecessors. Particularly, these are the point-to-point topology, bi-directional data transfers, and highest bandwidth. Even the slowest version, PCI Express x1, provides twice the bandwidth of the PCI (250MB/s against 133MB/s), and that in each direction, to the total of 500MB/s! As for the PCI Express x16 slot, intended for installation of graphics cards, its peak bandwidth is 4GB/s in each direction, while AGP 8x provides 2.1GB/s from the chipset to the GPU and about 200MB/s in the opposite direction.

Intel has already announced platforms with PCI Express x1 and x16, but they started developing graphics cards capable of working with this bus much earlier. As you know, PCI Express x16 will replace AGP 8x in computer systems of the future, so let’s consider this transition closely.

ATI Technologies and NVIDIA Corporation took diametrically opposite approaches to developing PCI Express-compatible products. The former taught the GPU itself to talk with this bus, while the latter devised a special HSI chip that functions as an AGP-to-PCI Express bridge. This chip also allowed NVIDIA to transfer the existing products to the new interface without any redesign. The result was the announcement of the GeForce PCX series that includes:

They also issued some GeForce PCX 5900 cards, which are based on the GeForce 5900 and have a HSI bridge.

Besides the obvious advantage of easy transition, the use of the bridge is deficient in its very nature: it’s impossible to use the potential of the PCI Express bus to the full this way. Besides that, this bridge generates quite an amount of heat and requires a passive heatsink. This design looks rather clumsy and the company admitted this, too, moving the HSI bridge to the chip’s substrate in the NV45, which is going to be used in PCI Express GeForce 6800 cards. This is a temporary measure, too, but the dual-die chip looks more secure and reliably than two separate chips. So, the NV45, against everyone’s expectations, turned to be nothing more than a combination of the NV40 and the HSI bridge.

ATI Technologies looks more advanced technologically as its GPUs have an inborn support of PCI Express. They are the R420 (RADEON X800), RV380 (RADEON 9600 XT) and RV370 (RADEON 9600 made by 0.11-micron tech process).

Recently, NVIDIA questioned the implementation of the PCI Express support in GPUs from the Canadian developer. You may have seen the snapshots on the Internet that compare RV380 and RV360 cores – they suppose that ATI just integrated an AGP-to-PCI Express bridge into the die, without using external chips. Well, this supposition may be true, but it is more likely that the part of the die that was responsible for communicating with the AGP was replaced with another circuitry, responsible for PCI Express x16. As a proof to that point we have the results of our tests of the bandwidth of PCI Express graphics cards. Although the results of the RADEON X600 are far from the theoretical maximum, they are anyway higher than the numbers the GeForce PCX 5900 got, meaning a better realization of PCI Express support in ATI’s products.

The whole transition affair is reminiscent of the move to SerialATA-150: of all the hard disk drive makers only Seagate equipped its devices with “native” support of the new interface, while others were content using bridge chips. It is only today that we see hard disk drives emerge that fully utilize the capabilities of the SerialATA interface. That’s probably the scenario the PCI Express x16 bus will be following: products with this interface won’t take the market by storm, in a couple of days. Moreover, appropriate drivers are necessary, so PCI Express will most likely uncover its full potential only with the release of the new version of Windows, codenamed Longhorn.

So, the first half of the current year brought the following into the realm of desktop 3D graphics:

  • Announcement of the next-generation GPU from NVIDIA – NV40;
  • Announcement of the next-generation GPU from ATI – R420;
  • Graphics cards started transitioning to the PCI Express x16 bus;
  • First solutions with “native” support of the new bus appeared.
 
Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 ]

Discussion

Comments currently: 68
Discussion started: 08/01/04 08:24:29 PM
Latest comment: 08/30/06 11:16:51 PM

View comments

Add your Comment