View Full Version : Define 4X AGP

04-03-2002, 07:30 AM
I am trying to understand exactly what 2X and 4X AGP means and why you want an AGP versus PCI. Since it seems PCI is a standard form, why did manufacturers move to this AGP standard a few years ago just for the video card?

If the first (i.e., graphics) PCI slot on my old Yosemite is a 66 mhz PCI, how is that different from a 66 mhz AGP (I mean, performance-wise, I can see that they are physically different)? Is the 4X AGP in my QS '01 somehow "accelerated" to be 264 mhz? If I put the same video card in my QS and in a G4 AGP with a 2x, is it two times faster in the QS?

Like most things with computers, it seems there are four or five different variables (video RAM, card speed, AGP 2X or 4X, PCI, etc) for video cards and being faster in one variable doesn't make one card necessarily better than another if that card is faster in another variable.



04-03-2002, 11:50 AM

Theoretical PCI/AGP slot transfer rates:

33MHz, 32-bit slot maximum data transfer rate of 132MBytes/sec.
33MHz, 64-bit slot maximum data transfer rate of 264MBytes/sec.
2x AGP slot maximum data transfer rate of 533MB/sec. (about 2x 33MHz, 64-bit slot)
4x AGP slot maximum data transfer rate of 1066MB/sec. (about 4x)

These are approximate and theoretical. I gleaned the above info from several different sources and just looking at it, quite possible that the exact numbers are really simple multiples. Or some of the above could just be wrong. The card in the slot would have its own limitations. And the controller chip may have other limiting factors. Apple gives a figure for the QuickSilver 33MHz, 64-bit PCI bus at about 215MB/sec which I do know is conservative. k

04-03-2002, 11:52 AM
The AGP port is a different and special port "Advanced Graphics Port" - I think are the terms. The slots is specfic for graphic video cards. The PCI slot can also contain a graphics card. They have different configurations a PCI card will not fit into an AGP Port.

The Sawtooth around the first G4 450/500 time was the first with an AGP port it was a 2X - any older machines only have PCI slots. The newer G4's with the 133MHz bus have 4X AGP ports.

Two - three good links ... http://www.macgurus.com/ubb/Forum1/HTML/000115.html ... http://www.macgurus.com/ubb/Forum21/HTML/000139.html ... http://www.macgurus.com/ubb/Forum4/HTML/000406.html

Good luck

04-10-2002, 01:24 AM
Imagine an Intel motherboard block diagram as four items.


ignore the periods. The BBS doesn't preserve spaces.

The Northbridge is a fast traffic cop, which controls all data transfers for the CPU and memory. CPU never talks to memory and memory never talks to CPU. Northbridge controls all transactions.

The Southbridge is a more relaxed traffic cop which handles all other IO for an Intel motherboard: disc controllers, graphics, NIC, human peripherals, and add-in cards. The Southbridge sends the Northbridge only the data that is truly necessary for the CPU and memory, allowing the Northbridge to run unfettered by slow peripheral access times.

Set the WayBack machine to 1997. There were still 12-14 players in the PC graphics industry. VL-bus and ISA are already too slow for 3D video. PCI graphics were the True Way. A year before, Doom was THE demo one showed on one's 2D graphics card, and Quake was now THE demo for one's graphics card with a 3D processor card added between your VGA card and the monitor (for example, the amazing accelerator by enfant terrible 3Dfx, for which I designed the daughtercard connector).

As all-in-one 3D+VGA graphics PCI cards were being designed, chip architects were discovering the limitations of PCI bandwidth as we tried to bring both VGA and 3D acceleration onto the same add-in card.

I started my sabbatical from the graphics industry in August 1997, and AGP was already announced. Intel invented Advanced Graphics Port as their prelude to bringing graphics on-board the CPU. Remember MMX technology? That was Intel's failed first move. All graphics companies were quaking in our boots, so we ignored MMX. Intel still is the BORG of hardware; but resistance wasn't entirely futile back then. Anyway, back to AGP.

Instead of being limited by the Southbridge, 3D graphics now had direct access to memory and CPU signals via the AGP Northbridge interface. Each time we need a speed bump, the Northbridge allows the AGP clock a 2x multiplier to get the attention of CPU and memory.


I have specs on my desk at work. I'll post the exact AGP 1x-8x numbers presently.

[This message has been edited by ricercar (edited 10 April 2002).]

04-23-2002, 07:43 PM
AGP Data transmit clock speeds on p21 of the AGP design guide available on http://www.agpforum.org/
1X 66 MHz
2X 133 MHz
4X 266 MHz
8X 533 MHz

[Note Kaye's numbers below are data transfer rate, not clocks. I missed this the first time and had to go back and double check my work...]

To get back to the original post, compare AGP these speeds with PCI at 33 MHz and you'll see why Intel was so keen to use AGP.

[This message has been edited by ricercar (edited 23 April 2002).]