Document: PCI chipsets list Maintainer: Patrick Duffy, duffy@theory.chem.ubc.ca Last Revision Date: August 20, 1995 Archived at: ftp.netcom.com, in directory /pub/ab/abe/ CompuServe (GO BENCHMARK) Web pages: http://warp.eecs.berkeley.edu/os2/workbench/work.htm This document may be distributed freely, provided it is distributed in its entirety and is unmodified. Here follows a brief summary of PCI chipsets. I've tried to keep it accurate, but if you spot any flaws please feel free to correct me, and if you have details on other chipsets, please feel free to let me know about them. I've included information at the end of this list to help people ID the chipset (assuming they have an Intel chipset on their motherboards). Where I know for sure that they do, I have indicated that a chipset will use parity RAM. PCI Vendor IDs first in HEX then in decimal appear in brackets beside the chipset names. Usefull numbers: --------------- ALI: (408) 764-0644 (sales) Intel: (800) 628-8686 (tech. support) PART ONE: 80486 Chipsets ------------------------- 1) The Aries Chipset (Intel: 8086/32902) ----------------- This is a chipset made by Intel for 80486-based machines, and used in boards (such as the Asus AVP4) where PCI and VL buses are to coexist. Unlike the Contaq chipset below, the Aries chipset implements PCI using a PCI to CPU bridge, and the VL bus is attached to this. This allows for full PCI performance, though I don't know what effect it has on VL (which likely will run no faster than the 33 MHz at which PCI 2.0 runs). The chipset has apparently been through at least two revisions, though I don't know of any specific problems with either one. 2) The Contaq Chipset (Contaq: 1080/4224) ------------------ The Contaq 82C599 is paired with one of their 486VL chipsets (82C596 or 82C597) and bridges directly from the 486 CPU to the PCI bus. Paraphrased from the Contaq spec.: The 82C596 system controller provides the CPU interface, VESA bus interface, ISA bus controller, etc. The 82C599 PCI controller provides the bridge between PCI master/slave agent and the ISA/VESA standard expansion bus; it arbitrates all the bus transactions between host CPU, PCI agent, VESA device, and ISA device. (Which sounds to me like the PCI bus is attached to the VL bus, rather than to the CPU, which will cause PCI performance degredation.) 3) The FINALI-486 Chipset (Acer Labs: 10B9/4281) ---------------------- This is ALI's chipset for 486 systems, consisting of the M1487 and M1489 chips. It supports EDO RAM and all the different 486 CPUs, and has IDE, a real-time clock, and a keyboard controller. I've no reports of success with boards using this chipset, and ACER 486's which use the board usually have slow cache controllers. Use CTCM to check your board's performance. 4) The Opti Chipset (Opti: 1045/4165) ---------------- Components: 82C822 (PCI functions) 82C895 (all other functions) This is Opti's chipset for 486-based machines. It's reported to work well with OS/2, although it implements PCI using a VL-to-PCI bridge instead of the other way around. Apparently this chipset will only allow 1 32-bit burst transfer per bus arbitration cycle, which limits throughput to 8 MB/s instead of (the maximum ideal transfer rate of) 132 MB/s. Boards which use this chipset are, therefore, to be avoided. 5) The Saturn Family of Chipsets (Intel: 8086/32902) ----------------------------- The Saturn family of chipsets is designed for use exclusively with 80486 and compatible processors, up to DX4s. They will, as of the latest revision, work with the P24T Pentium Overdrive processor. Boards which use them typically are of the combination ISA/PCI type, as I do not believe the Saturn chipsets were designed to handle VL extensions. The Saturn chipset has been through three revisions, numbered (oddly enough) 1, 2, and 4. Some brief comments on each: Rev. 1: Is now long-since discontinued. This chipset will only appear on older motherboards (perhaps pre Jan. '94?), and should not be on any motherboards of recent vintage. This chipset had problems (unknown to me), and so was put through its first revision and re-released. Rev. 2: Is still in wide use. This chipset has problems with cache integrity during PCI to CPU burst mode operations, as well as certain SCSI operations. Any board which uses this chipset today will still have these problems. There are BIOS "fixes" provided by various manufacturers, but what these usually do is disable the high-performance options on the chipset. Rev. 2 was never 'fixed', and there is no such thing as a newer release called rev. 2. Because of the bugs, however, a new version of the Saturn chipset was released. Rev. 4: With this release of the Saturn chipset, Intel seems to have finally fixed the problems with the earlier revisions. This chipset (also called Saturn II) also supports all the green features when used in combination with the right BIOS. It is distinguished externally from its older cousins by the last two letters on one of the three chips. Check for a 'ZX' to positively identify the Saturn II chipset. This chipset, when presented with a device which transfers 100 Mb/s, doesn't seem to let the CPU run at all. 6) The SIS chipset (Silicon Integrated Systems: 1039/4153) --------------- Components: 85C496 and 85C497 SIS makes a separate chipset for 486-based boards which is commonly used in systems which also have VL slots. Unfortunately, it sounds as though the chipset implements PCI by bridging it to VL (instead of the other way around), so PCI-based systems which have VL slots will suffer a degredation of PCI performance. Still, however, boards manufactured by AMI and Asus have been reported to work well with OS/2. There have been five revisions of this chipset. A4 (the earliest one) supports IDE up to mode 2 but apparently was not stable with caches on. B2 had on-board IDE support but did not support mode 3 well. Revisions B3 and later apparently work better. The varisous chipset revisions can be identified by their labels: SiS 85c496 MU & 85c497 MW (A4) SiS 85c496 NU & 85c497 NS (B2); SiS 85c496 NV & SiS 85c497 NS (B3); SiS 496 NV & SiS 497 NU (B4); and SIS496 OS and SIS 497 OT (B5). 7) The VIA GMC chipset (Vendor unknown) -------------------- This chipset includes the VIA VT82C486A-F with a built-in 8042 keyboard controller and a VIA VT82C505-D chipset for the VESA to PCI bridge. Specifically, the chips are as follows: 82C486A - cache/memory controller + VLB to ISA bridge 82C482 - VLB to ISA bridga (why there are two I'm not sure) 82C483 - DRAM controller VT82C505 PCI to VLB bridge A board using this chipset has been unstable (even under DOS/Win), and did not work with an Adaptec 2940 SCSI controller under OS/2 at all. Boards based on this chipset are therefore to be avoided. I have, however, had one report of success from someone using revision G of this chipset, so it could be that the new revision fixes problems with older rev's. Designers with whom I've corresponded indicate that improperly designed boards which use this chipset may have unstable caches. In addition, the cache controller reads the data into the cache SRAMs first, then into the CPU, increasing latency and reducing throughput. PART TWO: Pentium Chipsets --------------------------- 1) The Aladdin Chipset (Acer Labs: 10B9/4281) ------------------- This is a chipset (M1511/12/13) which supports Intel, AMD, and Cyrix CPUs. It has an advanced programmable interrupt controller, supports dual processors, all the different RAM and cache options, and includes an IDE controller which busmasters. I've no reports of success or failure with OS/2 and this chipset, and apparently early versions of this chipset had a cachable DRAM size of only 32 MB. 2) The Ariel Chipset (Intel: 8086/32902) ----------------- This is Intel's chipset for notebooks. It has most of the same features as Triton (below) and supports all the advanced power management features common to notebooks. 3) The Genie Chipset (Acer Labs: 10B9/4281) ----------------- This is a chipset by Acer Laboratories Inc. (ALI) which is intended for use in multiprocessing systems. Specifically, the chipset will support up to four P54C, P55C, Cyrix M1 or AMD K5 processors, though I don't know if you can "mix and match" the CPUs (though I doubt it). I've had no reports of success or failure with boards which use this chipset. 4) The Mercury Chipset (Intel: 8086/32902) ------------------- The Mercury chipset is designed for use primarily with 60 and 66 MHz Pentium processors (so you won't see 486 motherboards with it, and you usually won't see 90/99 MHz boards with it either, though there are always exceptions). It has, apparently, been through two revisions as well. The first revision is reported to have the same problems with posted buffered writes as Neptune I. This problem was fixed in Mercury II. This chipset will use parity RAM. 5) The Neptune Chipset (Intel: 8086/32902) ------------------- The Neptune chipset is designed for use with 75, 90 and 99 MHz Pentium processors (so you won't see 486 motherboards with it, or 60/66 MHz boards with it either). The Neptune chipset has been through two revisions, both of which will use parity RAM. Rev. 1: This chipset was in boards shipped by Intel to vendors up until about the end of July 1994. It has/had problems with posted buffered writes, which would manifest themselves most prominently with SCSI devices (which used this feature extensively). Recent releases of the AMI flash BIOS which Intel ships with their Premiere II (90/99 MHz) boards (the latest is 1.00.13.AX1) switch posted buffered writes off on the chipset when rev. 1 of the chipset is detected. Rev. 2: This chipset is in boards shipped by Intel to vendors as of about mid August 1994. It has no reported problems (and works well in my system). 6) The OPTI chipset (Opti: 1045/4165) ---------------- There are two OPTI chipsets in use: One is used in Pentium boards which also have a VL bus. This from the OPTI spec. sheet (paraphrased): One of the other OPTi chips provides CPU interface, VESA bus interface, ISA bus controller, etc. The 82C822 PCI Bridge bridges between the ISA/VESA and the PCI; it arbitrates the bus requests between host CPU, PCI masters, DMA/ISA masters, and Refresh. (The other chips in the chipset are the 82C596 and 82C597.) This chipset is reported to yield very poor VL/PCI performance, as an asynchronous clock is used for the local bus, resulting in synchronization delays. Apparently CPU/cache performance is acceptable. (Which again sounds to me like the CPU is hooked up to the VL bus and the PCI is bridged to that.) There is apparently a problem with this chipset which causes the BusLogic and Adaptec SCSI adapters to not work properly, as the PCI controller is not being properly programmed. The other OPTI chipset is apparently new and nicknamed "Viper". It consists of the 82C556 (data buffer controller), 557 (system controller), and 558 (peripheral controller) chips. It supports busmastering IDE and type F DMA, and Intel's Native Signal Processing (NSP), which means that it will allow the Pentium to perform DSP functions. It also supports all the latest power management features, and has the ability to run the PCI bus at 33 MHz irrespective of processor speed. There is a notebook variant of this chipset (an "N" is suffixed to the chip model numbers above) which supports all of the same features. I've no reports of success with this chipset and OS/2. Apparently the BusLogic SCSI controller has difficulty with one or both of the OPTI chipsets, in that the chipset can generate spurious interrupts, causing the card to not claim interrupts that it should. There is apparently a workaround for this in the latest BusLogic driver. 7) The Samsung chipset (Samsung Electronic Co., Ltd.: 1,099/4240) ------------------- Samsung makes a three-piece Pentium chipset. The KS82C531 is the cache (it will support synchronous caches) and RAM controller (EDO or DRAM). I've no more details on the chipset, nor reports of any success with it and OS/2. 8) The SIS chipset (Silicon Integrated Systems: 1039/4153) --------------- SIS makes a chipset for Pentium-based PCI motherboards which will apparently support (with the right BIOS) some non-spec. PCI operations. There are settings, apparently, which will allow the PCI bus to run at CPUCLK/1.5, which means that for a 90 MHz CPU (60 MHz external clock), you could run your PCI bus at 40 MHz. What affect this has on everything else remains to be seen, but the chipset is reported to work well with the Asus motherboard which uses it. 9) The Symphony "Rossini" Chipset (Symphony Labs: 10AD/4269) ------------------------------ This is apparently a low-cost alternative to the Triton chipset, as it operates with up to 66 MHz external clock rates, up to two CPUs, piplined or non-piplined, synchronous or [conventional] asynchronous SRAM cache, EDO RAM, dual-port busmastering IDE. It will, apparently, adjust the voltages to its various (CPU, PCI, cache, RAM) buses to suit their requirements, and will control up to six PCI masters. It consists of the SL82C551 cache/memory controller, the SL82C522 data path controller, and the SL82C555 system I/O controller. 10) The Triton Chipset (Intel: 8086/32902) ------------------ This is the latest PCI chipset to come out from Intel, and is PCI 2.1-compliant. It will also increase PCI-to-memory bandwith from 40 MB/sec (in Neptune) to 100 MB/sec, and features an enhanced ISA bridge and built-in EIDE support. Drivers for all major operating systems to take advantage of the busmastering capability of the Triton chipset may be found in a self-extracting Zip file called triton.exe, available at all the major FTP sites. (Note that one person has reported that his system locks when using this driver in combination with fixpack 9). It also supports EDO RAM, which will allow access to RAM in a 3-2-2-2 burst rather than the conventional 7-4-4-4 burst. There is also now a 12 word buffer between the PCI bus and memory as opposed to the 8 word buffer in Neptune. The chipset does not check parity, although parity RAM may be used (without any benefit) in systems which use the chipset. Systems which use the chipset appear to run OS/2 very well. 11) The Wild Cat chipset (VLSI Technology Inc.: 1004/4100) -------------------- This is a chipset by VLSI. It will handle EDO RAM, drive DRAM chips and some ISA slots without external buffers, byte merging (for writes), and the ability to read data from the 64-bit data buffer without first writing it to memory. It will support all the AMD and Cyrix chips as well. I've no reports of success with this chipset and OS/2, though the performance of the chipset is reported to be better than Neptune II but not as good as Triton. PART THREE: P6 Chipsets ------------------------ 1) The Mars Chipset (Intel: 8086/32902) ---------------- This is a chipset by Intel which supports many of the same features as the Triton chipset. It will be the general purpose chipset for use with mainstream PCI P6 motherboards. 2) The Mercury Chipset (Opti: 1045/4165) ------------------- This is a chipset by OPTI. 3) The Orion Chipset (Intel: 8086/32902) ----------------- Orion is a six- to nine-chip set (depending on the implementation) by Intel. It supports multiprocessing, up to four memory interface controllers (each of which can support up to 1 GB of DRAM), and the new 64 MB DRAM chips. This chipset will be used in higher-end systems (with prices to match). Useful Information: ------------------- How to find out your motherboard PCI chipset type and revision: (Requires DOS support to be installed.) THE FOLLOWING INFORMATION IS SUPPLIED WITHOUT ANY WARRANTY, EITHER EXPRESS OR IMPLIED, OF ANY KIND. UNDER NO CIRCUMSTANCES MAY I (PATRICK DUFFY) BE HELD LIABLE FOR ANY DAMAGE RESULTING FROM USE OF THE INFORMATION GIVEN BELOW. YOU (THE READER) ASSUME FULL RESPONSIBILITY FOR ITS USE AND THE CONSEQUENCES THEREOF. The following commands will identify various aspects of your motherboard PCI chipset. Type each command as it appears (and press enter, of course). I've tested this on my own motherboard and it seems to work. Under the result column, '-' means that nothing will be shown. I do not know what the result codes will be for the Contaq or Aries chipsets, so please feel free to let me know. Command Result Comments debug Enter debug. O CF8 F0 - Open PCI interface. I C000 86 Chipset manufacturer ID: `8086' means I C001 80 Intel, any other value implies the other bytes are nonsignificant. I C002 DeviceID_LSB I C003 DeviceID_MSB Chipset's ID I C008 RevisionID Revision number O CF8 00 - Close PCI interface quit Exit debug. The chipset's ID can be 0483 (Saturn), 0486 (Aries), 04A3 (Mercury/Neptune), or C557 (OPTI Viper). For the Saturn chipset, revision 1/2 is for Saturn I, rev. 4 is Saturn II. For Mercury/Neptune: revision 1/3 is for Mercury I/II, and revision 10/11 is for Neptune rev. 1/2. Anybody who's IDed the Triton chipset please do let me know what the appropriate responses are. How to identify which revision of the SIS 486 chipset you have: There are five slightly different revisions of the SIS chipset used: A4, B2, B3, B4, and B5. The A4 does not have onboard ide support. All the B-level chipsets do, however, with B5 being the best chipset to have. The various chipset revisions can be identified by their labels: SiS 85c496 MU & 85c497 MW (A4) SiS 85c496 NU & 85c497 NS (B2); SiS 85c496 NV & SiS 85c497 NS (B3); SiS 496 NV & SiS 497 NU (B4); and SIS496 OS and SIS 497 OT (B5). So that's what I know, as clearly as I know how to say it. If you have any questions or notice any ambiguities please do let me know and I'll do my best to correct them. -- Patrick Duffy, duffy@theory.chem.ubc.ca -- I am Homer of Borg. Resistance is fu-- mmmmm... donuts...