The target at the time of creating this bus was to obtain a valuation of transference of information graphic micro-card superior to the one that the PCI of 32 bits offers to 33 MHz, 132 Mb / century This valuation turns out to be sufficient for applications 2D, but insufficient (at least theoretically) for the new cards 3D, that must transmit several "gentle" of textures to obtain the maximum realism.
Types of AGP
As many young technologies, AGP was thrown to the market as soon as it was prepared, although he still had not become polished completely. For it, several AGP ways exist:
- AGP 1x: way with bus of 32 bits and to 66 MHz. His theoretical valuation of maximum transference is 264 Mb / century At present, few cards of mark have only this way.
- AGP 2x: way with bus of 32 bits and to 66 real MHz, or 133 "virtual" MHz thanks to the double use of every sign of clock (a used technology, for example, in the DDR-SDRAM). His theoretical valuation of maximum transference is 528 Mb / century It is at present used by the quality cards.
- AGP 4x: new way that will be implanted in a future; approximately 1 GB / century
Undoubtedly, the way 1x is an "experimental" way extracted to the market with hurries. His yield is and it will be indistinguishable of that of PCI, so his life time has already ended (except like way for compatibility with more ancient cards). The way 2x is the authentic AGP, although as we will not see either it is the panacea...
AGP or PCI?
That is the question. For those hastily, my conclusion: it gives almost the same, depends on the chip of the card. Let's justify this risky affirmation:
528 MB/s of the AGP 2x (1x to comment on it) is theoretical. The question is that the chip has to distribute his access to the memory with the channel AGP, and with the current memoirs it turns out to be impossible that this 528 Mb / century number is Even reached with the 66 MHz SDRAM (the EDO is already a history), the breadth of bus of memory is exactly 528 MB/s, with what it turns out to be clear that if one gives him everything to the AGP, the mike remains without reading the memory, what cannot spend since the computer would not work, skylight.
So the yield of the AGP 2x lowers even less than the half, with what the difference with the PCI becomes very scarce, up to the point of which at present almost all the cards of the market do not offer difference with bus AGP or PCI, or differences of the order of 2 %. Also, neither Windows 95 nor NT make use of some important characteristics of the AGP.
Conclusion: the AGP will be an advance at the moment when there happens someone (if possible several) of these questions: that the memoirs RAM are even more rapid (memoirs type PC100, which they allow a memory bus to 100 MHz of 800 MB/s), that the AGP overcomes 2x (what will depend on the chipset) or that Windows 98 and NT 5 include progress in the support software.
Meanwhile, the most reasonable thing is to buy the best card that they offer him, with independence if it is AGP or PCI; if to chip equality they cost the same, choose undoubtedly AGP. Of course, if he chooses PCI make sure that the motherboard has his slot AGP 2x for when into the future it changes the situation.