undefined
AGP flies under Linux....well it works well as fast PCI
1.0 Is there anything wrong with PCI ?
PCI was a great open solution to the bandwidth bottlenecks in the ISA bus architecture that addressed the licensing issues inherent in IBM's MCA solution. AGP has no killer application to speak of other than a few spot and slick presentations showing projections and extrapoloations of current graphics card memory usage models (x% for display, y% for Z-buffer, and z% for texturres) from todays lowish resolution 3D games (i.e. MechWarrior at 640x480 ). The carrot on the stick was that concurrent engineering on the operating system side, the PC chipset side and the graphics chipset could have you running Quake at 1280x1024 at 30fps with no blocky 3D objects or backgrounds. Local existing memory, say 4MB, could be used for display and Z-buf and the an unlimited amount, say 16 MB, could be used to give rich textures.
Oh yeah, imaginative software engineers could find other uses for this technology as well. So if you are sticking with business apps, you probably do not need it....that is until Microsoft starts to actually use the capability and bloating the backgrounds with textures. Then we are would have to upgrade our Pentium 100 machines...Intel and the collective hardware industry would again put on shades (cuz the futures so bright...you gotta...). For most business apps today, a large drive and a 486 processor would probably suffice if people opted away from integrated office packages and used a set of seamlessly integrated "functions"....(I think Linux has real potential in this area if a few key business appls could be written (spreadsheet, word processing, drawings) that could import and export files back and forth from MS native formats.) I'm getting off track....Anyways for 80% of us AGP is not required. Problem is that all current Intel chipsets will not be like Volkswagon's the VX, HX and TX will go the way of the FX (remember Triton) chipset. The only silver lining that will keep the industry churning out today's practical designs are Chipset competitors like VIA, Acer, and SIS and the large third world market. Tommorows chipset that you can buy today either :
that will eanble motherboards to run software/hardware you have today and tomorrows AGP ...so it is insurance if that Killer AGP app gets invented or if it finds itself into bloated software that you might want to zip through. Just be sure you can live only two ISA slots :)
A common trap is to equate AGP with better 2D performance in the same way that VESA and PCI local buses offered a boost to MS Windows PCI increased the bandwidth 8-fold over the ISA bus by offering 32 bit wide interfaces(doubing) and a quadrupuling of the bus clock (from 8Mhz to 33 Mhz). This increased in bandwidth took place just when accelerated graphics cards began to decreased CPU bus traffic requirements. PCI offered a lot of bandwidth. For higher 3D resolution at 800x600, older 3D designs without a setup engine will have significant PCI traffic. The combination of a separate dedicated bus (really single slot called AGP) running PCI protocol at twice the current 33MHz rate should improve congestion caused by other PCI bandwidth contending peripherals ,such as high speed PCI SCSI cards, and make PCI-66 a non-issue. Hardware performance freaks out there...raise your hands....How many of you guys run business apps ?
The speed of a typical PCI interface might be fast enough for texturres... problem is both bandwidth and memory size requirements go up quickly as texture detail and realism increase.......AGP ...when properly implemented ...can provide a computer with:
The implementation of AGP requires co-operation between the PC chipset, graphiics card/chipset (, and the operating system. The AGP slot without the AGP speciifics coded into the operating system will behave like a PCI slot running at 66 MHz as the AGP protocol is built on top of PCI. Various AGP cards from ATI have been found to run on Linux operating systems with no special coding by virture of lineage and inheritance of PCI. Great design....Intel!
The Intel AGP spec was written partly as a white paper to persuade the world that they needed another bus. Thus, they wanted to "kickstart the add-on market" into adopting yet-another bus interface, but allowed the market to implement the various subsystems defined in the specifications as they saw fit and compete on a market driven basis. Things are clearer on the motherboard side as both the current offereings (VP-3 and LX) support one all encompassing full standard. Things are not so clear on the add-on AGP side.
PC World did a real good job on demystifying AGP from the point of add-on graphics card. In summary, various graphics cards labelled with AGP will have one or more of the following component parts in their AGP design:
Most of the legitimate AGP add-on cards introduced lately come in one of 3 combinations:
The AGP versions of the Xpert@Work (Xpert@Play) have been tested on recent motherboard designs for Linux and Xfree86 X11 compatability. The cards do not utilizize any of AGP's advance features but rely on AGP base support of PCI to run. The boards were tested on older release of Linux (Version 1.2.8) and found to work fine once the card was set up to recognize as at RageII chipset by setting appropriate options. Futher work by Xfree86 should resolve any remaining glitches. The cards as the were are limited to the same programmimg limitation with respect to the maximum DOT CLOCK of 170 MHZ keyed to the RageII. The boards were tested on two different systems using the LX chipset and one system using the VIA VP-3 chipset:
All systems were able to run demos involving several windows of DOOM, Quake and Mesa OpenGL running concurrently.
5.0 Fake AGP....replace Accelerated with Advanced
If I replace the "Kentucky" in "KFC", could I still call it KFC in bold. Believe it or not ...yes...adopting all or even none of the above features have allowed some boards to be given the status of AGP: Advanced Graphics Port is another scam: from the same makers of "FAKE CACHE" who sold unsuspecting legions of 486 board owners non working cache chips. These guys shamelessly exploit and redefine AGPs use of "Accelerated" with "Advanced" and use the same AGP acronym....shame! Is Intel doing anything ?
See following links:
The above boards use "shared" memory chipsets in which even "display memory" has to be accessed gotten from main memory....very slow..... The memory cannot be (dynamically) released back to the operating system and other applications. I have seen these chips advertised AGP 133 and advertised for $99...buyer beware...it is too good to be true.
...as
of Nov 97
