Created by Intel Width in bits 32 Speed Up to 2133 MB/s | Superseded by PCI Express (2004) No. of devices One device per slot | |
![]() | ||
Year created 1996; 21 years ago (1996) |
The Accelerated Graphics Port (AGP) is a high-speed point-to-point channel for attaching a video card to a computer system, primarily to assist in the acceleration of 3D computer graphics. It was originally designed as a successor to PCI-type connections for video cards. Since 2004, AGP has been progressively phased out in favor of PCI Express (PCIe); by mid-2008, PCI Express cards dominated the market and only a few AGP models were available.
Contents
Advantages over PCI
As computers increasingly became graphically oriented, successive generations of graphics adapters began to push the limits of PCI, a bus with shared bandwidth. This led to the development of AGP, a "bus" dedicated to graphics adapters.
AGP is heavily based on PCI, and in fact the AGP bus is a superset of the conventional PCI bus, and AGP cards must act as PCI cards.
The primary advantage of AGP over PCI is that it provides a dedicated pathway between the slot and the processor rather than sharing the PCI bus. In addition to a lack of contention for the bus, the direct connection allows for higher clock speeds.
The second major change is that AGP uses split transactions, where the address and data phases of a PCI transaction are separated. The card may send many address phases, and the host processes them in order. This avoids long delays, with the bus idle, during read operations.
Third, PCI bus handshaking is simplified. Unlike PCI bus transactions whose length is negotiated on a cycle-by-cycle basis using the FRAME# and STOP# signals, AGP transfers are always a multiple of 8 bytes long, and the total length is included in the request. Further, rather than using the IRDY# and TRDY# signals for each word, data is transferred in blocks of four clock cycles (32 words at AGP 8× speed), and pauses are allowed only between blocks.
Finally, AGP allows (optional in AGP 1.0 and 2.0, mandatory in AGP 3.0) sideband addressing, meaning that the address and data buses are separated so the address phase does not use the main address/data (AD) lines at all. This is done by adding an extra 8-bit "SideBand Address" bus over which the graphics controller can issue new AGP requests while other AGP data is flowing over the main 32 address/data (AD) lines. This results in improved overall AGP data throughput.
This great improvement in memory read performance makes it practical for an AGP card to read textures directly from system RAM, while a PCI graphics card must copy it from system RAM to the card's video memory. System memory is made available using the graphics address remapping table (GART), which apportions main memory as needed for texture storage. The maximum amount of system memory available to AGP is defined as the AGP aperture.
History
The AGP slot first appeared on x86-compatible system boards based on Socket 7 Intel P5 Pentium and Slot 1 P6 Pentium II processors. Intel introduced AGP support with the i440LX Slot 1 chipset on August 26, 1997, and a flood of products followed from all the major system board vendors.
The first Socket 7 chipsets to support AGP were the VIA Apollo VP3, SiS 5591/5592, and the ALI Aladdin V. Intel never released an AGP-equipped Socket 7 chipset. FIC demonstrated the first Socket 7 AGP system board in November 1997 as the FIC PA-2012 based on the VIA Apollo VP3 chipset, followed very quickly by the EPoX P55-VP3 also based on the VIA VP3 chipset which was first to market.
Early video chipsets featuring AGP support included the Rendition Vérité V2200, 3dfx Voodoo Banshee, Nvidia RIVA 128, 3Dlabs PERMEDIA 2, Intel i740, ATI Rage series, Matrox Millennium II, and S3 ViRGE GX/2. Some early AGP boards used graphics processors built around PCI and were simply bridged to AGP. This resulted in the cards benefiting little from the new bus, with the only improvement used being the 66 MHz bus clock, with its resulting doubled bandwidth over PCI, and bus exclusivity. Examples of such cards were the Voodoo Banshee, Vérité V2200, Millennium II, and S3 ViRGE GX/2. Intel's i740 was explicitly designed to exploit the new AGP feature set. In fact it was designed to texture only from AGP memory, making PCI versions of the board difficult to implement (local board RAM had to emulate AGP memory.)
Microsoft first introduced AGP support into Windows 95 OEM Service Release 2 (OSR2 version 1111 or 950B) via the USB SUPPLEMENT to OSR2 patch. After applying the patch the Windows 95 system became Windows 95 version 4.00.950 B. The first Windows NT-based operating system to receive AGP support was Windows NT 4.0 with Service Pack 3, introduced in 1997. Linux support for AGP enhanced fast data transfers was first added in 1999 with the implementation of the AGPgart kernel module.
Versions
Intel released "AGP specification 1.0" in 1997. It specified 3.3 V signals and 1× and 2× speeds. Specification 2.0 documented 1.5 V signaling, which could be used at 1×, 2× and the additional 4× speed and 3.0 added 0.8 V signaling, which could be operated at 4× and 8× speeds. (1× and 2× speeds are physically possible, but were not specified.)
Available versions are listed in the table on the right.
AGP version 3.5 is only publicly mentioned by Microsoft under Universal Accelerated Graphics Port (UAGP), which specifies mandatory supports of extra registers once marked optional under AGP 3.0. Upgraded registers include PCISTS, CAPPTR, NCAPID, AGPSTAT, AGPCMD, NISTAT, NICMD. New required registers include APBASELO, APBASEHI, AGPCTRL, APSIZE, NEPG, GARTLO, GARTHI.
There are various physical interfaces (connectors); see the Compatibility section.
AGP Pro
An official extension for cards that required more electrical power, with a longer slot with additional pins for that purpose. AGP Pro cards were usually workstation-class cards used to accelerate professional computer-aided design applications employed in the fields of architecture, machining, engineering, simulations, and similar fields.
64-bit AGP
A 64-bit channel was once proposed as an optional standard for AGP 3.0 in draft documents, but it was dropped in the final version of the standard.
The standard allows 64-bit transfer for AGP8× reads, writes, and fast writes; 32-bit transfer for PCI operations.
Unofficial variations
A number of non-standard variations of the AGP interface have been produced by manufacturers.
Internal AGP interface
Compatibility
AGP cards are backward and forward compatible within limits. 1.5 V-only keyed cards will not go into 3.3 V slots and vice versa, though "Universal" cards exist which will fit into either type of slot. There are also unkeyed "Universal" slots that will accept either type of card. When an AGP Universal card is plugged-into an AGP Universal slot, only the 1.5 V portion of the card is used. Some cards, like Nvidia's GeForce 6 series (except the 6200) or ATI's Radeon X800 series, only have keys for 1.5 V to prevent them from being installed in older mainboards without 1.5 V support. Some of the last modern cards with 3.3 V support were the Nvidia GeForce FX series (FX 5200, FX 5500, FX 5700, some FX 5800, FX 5900 and some FX 5950), Geforce 6 Series (6200, 6600/6600 LE/6600 GT only) and the ATI Radeon 9500/9700/9800(R350) (but not 9600/9800(R360)). Some Geforce 6200 and Geforce 6600 cards will function with AGP 1.0 (3.3v) slots.
AGP Pro cards will not fit into standard slots, but standard AGP cards will work in a Pro slot. Motherboards equipped with a Universal AGP Pro slot will accept a 1.5 V or 3.3 V card in either the AGP Pro or standard AGP configuration, a Universal AGP card, or a Universal AGP Pro card.
Some cards incorrectly have dual notches, and some motherboards incorrectly have fully open slots, allowing a card to be plugged into a slot that does not support the correct signaling voltage, which may damage card or motherboard. Some incorrectly designed older 3.3 V cards have the 1.5 V key.
There are some proprietary systems incompatible with standard AGP; for example, Apple Power Macintosh computers with the Apple Display Connector (ADC) have an extra connector which delivers power to the attached display. Some cards designed to work with a specific CPU architecture (e.g., PC, Apple) may not work with others due to firmware issues.
Mark Allen of Playtools.com made the following comments regarding Practical AGP Compatibility for AGP 3.0 and AGP 2.0:
"...nobody makes AGP 3.0 cards, and nobody makes AGP 3.0 motherboards. At least not any manufacturers I can find. Every single video card I could find which claimed to be an AGP 3.0 card was actually a universal 1.5V AGP 3.0 card. And every motherboard which claimed to be an AGP 3.0 motherboard turned out to be a universal 1.5V AGP 3.0 motherboard. It makes sense, if you think about it, because if anyone actually shipped a consumer-oriented product which supported only 0.8 volts, they would end up with lots of confused customers and a support nightmare. In the consumer market, you'd have to be crazy to ship a 0.8 volt only product."
Power consumption
Actual power supplied by an AGP slot depends upon the card used. The maximum current drawn from the various rails is given in the specifications for the various versions. For example, if maximum current is drawn from all supplies and all voltages are at their specified upper limits, an AGP 3.0 slot can supply up to 48.25 watts; this figure can be used to specify a power supply conservatively, but in practice a card is unlikely ever to draw more than 40 W from the slot, with many using less. AGP Pro provides additional power up to 110 W. Many AGP cards had additional power connectors to supply them with more power than the slot could provide.
Legacy use
By 2010 few new motherboards had AGP slots. No new motherboard chipsets were equipped with AGP support, but motherboards continued to be produced with older chipsets with support for AGP.
Graphics processors of this period use PCI-Express, a general-purpose (not restricted to graphics) standard that supports higher data transfer rates and full-duplex. To create AGP-compatible graphics cards, those chips require an additional PCIe-to-AGP bridge-chip to convert PCIe signals to and from AGP signals. This incurs additional board costs due to the need for the additional bridge chip and for a separate AGP-designed circuit board.
Various manufacturers of graphics cards continued to produce AGP cards for the shrinking AGP user-base. The first bridged cards were the GeForce 6600 and ATI Radeon X800 XL boards, released during 2004-5. In 2009 AGP cards from Nvidia had a ceiling of the GeForce 7 Series. In 2011 DirectX 10-capable AGP cards from AMD vendors (Club 3D, HIS, Sapphire, Jaton, Visiontek, Diamond, etc.) included the Radeon HD 2400, 3450, 3650, 3850, 4350, 4650, and 4670. The HD 5000 AGP series mentioned in the catalyst software was never available. There were many problems with the AMD Catalyst 11.2 - 11.6 AGP hotfix drivers under Windows 7 with the HD 4000 series AGP video cards; use of 10.12 or 11.1 AGP hotfix drivers is the recommended workaround. Several of the vendors listed above make available past versions of the AGP drivers.
Protocol
An AGP bus is a superset of a 66 MHz conventional PCI bus and, immediately after reset, follows the same protocol. The card must act as a PCI target, and optionally may act as a PCI master. (AGP 2.0 added a "fast writes" extension which allows PCI writes from the motherboard to the card to transfer data at higher speed.)
After initialization, AGP transactions are permitted. For these, the card is always the AGP master and the motherboard is always the AGP target. The card queues multiple requests which correspond to the PCI address phase, and the motherboard schedules the corresponding data phases later. An important part of initialization is telling the card the maximum number of outstanding AGP requests which may be queued at a given time.
AGP requests are similar to PCI memory read and write requests, but use a different encoding on command lines C/BE[3:0] and are always 8-byte aligned; their starting address and length are always multiples of 8 bytes (64 bits). The three low-order bits of the address are used instead to communicate the length of the request.
Whenever the PCI GNT# signal is asserted, granting the bus to the card, three additional status bits ST[2:0] indicate the type of transfer to be performed next. If the three bits are 111
, the card may begin a PCI transaction or (if sideband addressing is not in use) queue a request in-band using PIPE#.
In-band AGP requests using PIPE#
To queue a request in-band, the card must request the bus using the standard PCI REQ# signal, and receive GNT# plus bus status ST[2:0] equal to 111
. Then the card asserts the PIPE# signal while driving the AGP command, address, and length on the C/BE[3:0], AD[31:3] and AD[2:0] lines, respectively. (If the address is 64 bits, a dual address cycle similar to PCI is used.) For every cycle that PIPE# is asserted, the card sends another request without waiting for acknowledgement from the motherboard, up to the configured maximum queue depth. The last cycle is marked by deasserting REQ#, and PIPE# is deasserted on the following idle cycle.
In-band requests are always sent at single data rate (1× speed).
Possible request codes are:
AGP 3.0 dropped high-priority requests and the long read commands, as they were little used.
Side-band AGP requests using SBA[7:0]
If side-band addressing is supported and configured, the PIPE# signal is not used. (And the signal is re-used for another purpose in the AGP 3.0 protocol, which requires side-band addressing.) Instead, requests are broken into 16-bit pieces and sent across the SBA bus. The possible values are:
0aaa aaaa aaaa alll
10cc ccra aaaa aaaa
110r aaaa aaaa aaaa
1110 aaaa aaaa aaaa
1111 0xxx
, 1111 10xx
, 1111 110x
1111 1110
1111 1111
Sideband address bytes are sent at the same rate as data transfers, up to 8× the 66 MHz basic bus clock. Sideband addressing has the advantage that it mostly eliminates the need for turnaround cycles on the AD bus between transfers, in the usual case when read operations greatly outnumber writes.
AGP responses
While asserting GNT#, the motherboard may instead indicate via the ST bits that a data phase for a queued request will be performed next. There are four queues: two priorities (low- and high-priority) for each of reads and writes, and each is processed in order. Obviously, the motherboard will attempt to complete high-priority requests first, but there is no limit on the number of low-priority responses which may be delivered while the high-priority request is processed.
For each cycle when the GNT# is asserted and the status bits have the value 00p
, a read response of the indicated priority is scheduled to be returned. At the next available opportunity (typically the next clock cycle), the motherboard will assert TRDY# (target ready) and begin transferring the response to the oldest request in the indicated read queue. (Other PCI bus signals like FRAME#, DEVSEL# and IRDY# remain deasserted.) Up to four clock cycles worth of data (16 bytes at AGP 1× or 128 bytes at AGP 8×) are transferred without waiting for acknowledgement from the card. If the response is longer than that, both the card and motherboard must indicate their ability to continue on the third cycle by asserting IRDY# (initiator ready) and TRDY#, respectively. If either one does not, wait states will be inserted until two cycles after they both do. (The value of IRDY# and TRDY# at other times is irrelevant and they are usually deasserted.)
The C/BE# byte enable lines may be ignored during read responses, but are held asserted (all bytes valid) by the motherboard.
The card may also assert the RBF# (read buffer full) signal to indicate that it is temporarily unable to receive more low-priority read responses. The motherboard will refrain from scheduling any more low-priority read responses. The card must still be able to receive the end of the current response, and the first four-cycle block of the following one if scheduled, plus any high-priority responses it has requested.
For each cycle when GNT# is asserted and the status bits have the value 01p
, write data is scheduled to be sent across the bus. At the next available opportunity (typically the next clock cycle), the card will assert IRDY# (initiator ready) and begin transferring the data portion of the oldest request in the indicated write queue. If the data is longer than four clock cycles, the motherboard will indicate its ability to continue by asserting TRDY# on the third cycle. Unlike reads, there is no provision for the card to delay the write; if it didn't have the data ready to send, it shouldn't have queued the request.
The C/BE# lines are used with write data, and may be used by the card to select which bytes should be written to memory.
The multiplier in AGP 2×, 4× and 8× indicates the number of data transfers across the bus during each 66 MHz clock cycle. Such transfers use source synchronous clocking with a "strobe" signal (AD_STB[0], AD_STB[1], and SB_STB) generated by the data source. AGP 4× adds complementary strobe signals.
At AGP 4× and 8× speeds, it is possible for a request to complete in the middle of a clock cycle. In such a case, the cycle is padded with dummy data transfers (with the C/BE# byte enable lines held deasserted).
Connector pinout
The AGP connector contains almost all PCI signals, plus several additions. The connector has 66 contacts on each side, although 4 are removed for each keying notch. Pin 1 is closest to the I/O bracket, and the B and A sides are as in the table, looking down at the motherboard connector.
Contacts are spaced at 1 mm intervals, however they are arranged in two staggered vertical rows so that there is 2 mm space between pins in each row. Odd-numbered A-side contacts, and even-numbered B-side contacts are in the lower row (1.0 to 3.5 mm from the card edge). The others are in the upper row (3.7 to 6.0 mm from the card edge).
PCI signals omitted are:
Signals added are: