ISA go Bragh™
Last modified: 2023-05-28 08:41
In the PC System Design Guide (1997–2001), Intel and Microsoft made the case that ISA should be not merely deprecated, but completely eliminated. The reasons were "To improve system performance, reduce customer support costs, and ensure true ease of use in PC systems and peripherals."
Although the guide had teeth only through the "Designed for Microsoft Windows" logo program, ISA slots disappeared from all consumer (non-industrial) motherboards not long after the final version of the guide was published. The entire ecosystem of hardware and software that had grown up around ISA was orphaned, and anyone who found the ostensible PCI replacements to be a big disappointment was left with only bad options going forward.
One of the less worst options is to maintain a collection of working hardware and spare parts from the last days of ISA. This page lists the last chipsets from each manufacturer that supported fully functional ISA slots for which I have been able to find example motherboards as evidence. Please email if you have additional evidence or firsthand experience that would improve this list.
| Chipset | ||||||||
|---|---|---|---|---|---|---|---|---|
| Intel | AMD | [Industrial] | VIA | SiS | ALi | Other | ||
| C | Intel | ☑ | ☠ | ☑ | ☑ | ☑ | ☑ | ↓ |
| P | AMD | ☠ | ☑ | ☐ | ☑ | ☑ | ☑ | ↓ |
| U | VIA | ☐ | ☠ | ☑ | ☑ | ☑ | ☐ | ↓ |
| Other | → | → | → | → | → | → | Other | |
Disclaimer for "max CPU" quotations
VIA southbridge complaints
ALi AGP instability complaints
Other buggy DMA bridges
Other ways to get burned
Dell's shiny heat sink dwarfs the Pentium II cartridge.
Tabor3
Gateway OEM board: Slot 1, 100 MHz FSB, max
384 MiB RAM (PC100), max 600 MHz Pentium III (per Gateway).*
ASUS CUBX: Socket 370, 100 MHz FSB (overclockable), max 1 GiB RAM (PC100/133), max 1 GHz Pentium III (per CPU-Upgrade).
Abit BX133-RAID: Socket 370, 100 MHz FSB (overclockable), max 768 MiB RAM (PC100/133), max 1.13 GHz Pentium III (per hearsay).
All have AGP 2×, but AGP is overclocked when FSB exceeds 100 MHz.
* I have a Tabor 3 motherboard running a passively cooled 700 MHz Pentium III.
440BX was the last Intel desktop chipset to include ISA support. It enjoyed an extended life due to market rejection of the early 8xx chipsets that followed it, and for other reasons. In fact... it still lives.
The most commonly occurring 440BX is a Gateway, Dell, or Compaq OEM Slot 1 board with a 450–500 MHz Pentium II or Pentium III. The fastest compatible CPU frequency on these OEM boards is something less than 1 GHz.
The slower Pentium IIIs are able to survive with a large, passive heat sink and good case airflow.
Related chipsets:
Revised 2015-02-03: Thanks to Jaymes M. Wilks for noticing
that I had misread what the Intel white paper said about LPC/ISA bridges.
Revised 2023-05-01: Added info from Andy Tait.
Example with ISA: MB865: LGA775, 800 MT/s FSB, AGP 8×, max 2 GiB PC3200 DDR RAM, max 3.8 GHz P4 (per manual).
ISA support was dropped by Intel in i810. Their white paper Implementing Industry Standard Architecture (ISA) with Intel Express Chipsets explains what the options are/were for specialty manufacturers to add ISA support onto later chipsets:
* In 440BX, Intel supported ISA using a PCI/ISA bridge that was integrated with the PIIX4E southbridge.
I'm aware of two success stories using 9xx or later chipsets:
865G, 865PE, and 875P boards supporting AGP 8×, dual channel DDR RAM, and FSB frequencies of 800 MT/s were manufactured but are now hard to find. Related chipsets according to Wikipedia:
845-series boards with working ISA DMA remain relatively common in the used market. From Wikipedia plus personal experience:
"ADD" stands for AGP Digital Display. The slot is physically indistinguishable from AGP, but ADD cards just provide a DVI output for the integrated graphics; they have no on-board memory or GPU. It's a trap!
Regarding the compatibility of third-party ISA bridging, my conclusion from personal experience so far is that it is close enough for some things but not for others. It's kind of like using a VIA chipset instead of Intel: when something is glitchy or completely broken, you cannot prove that the chipset is to blame, but circumstantially these same things work better on other systems. If you want 440BX compatibility, you know where to find it.
Searching the forums turns up lots of testimonials that a Sound Blaster worked with this or that board, but also reports of trouble with other devices, plenty of confusion about ISA bus mastering versus ISA DMA, and scary tidbits about what else can go wrong when you demand DOS compatibility out of P4-era hardware and firmware. The following list of relevant threads won't be updated, so do your own research.
Builders of 8xx chipset motherboards with ISA slots include industrial manufacturers BCM, Adek, iBASE or ATTRO? (the MB865 manual doesn't name its maker, and both companies list it as a product), Kontron, NIXSYS, and Bressner (UK), as well as familiar names Intel, MSI, Supermicro, DFI (or DFI-ITOX), EPoX (calling itself IPOX for EPoX Industrial), Legend (Lenovo), and SOYO.
2018-06: New industrial boards from random Chinese manufacturers are now available through AliExpress.
Example with ISA: Soyo SY-7VBA133U: Socket 370, 133 MHz FSB, max 1.5 GiB PC133 RAM, AGP 4×, max 1.4 GHz Pentium III (Tualatin) (per CPU-Upgrade).
Apollo Pro 133T = VT82C694T northbridge + VT82C686B southbridge.
VIA's chipset one-upped 440BX on specifications, but widely reported southbridge complaints obviated that marketing advantage.
Related chipsets: PLE133T appears to be Apollo Pro 133T with integrated Trident graphics. The ones I found did not have an AGP slot.
Example with ISA: ECS P6STMT: Socket 370, 133 MHz FSB, max 1 GiB PC133 RAM, max 1.266 GHz Pentium III (per ECS). No AGP slot (integrated SiS graphics).
Example with ISA: PC Chips M754LMR+ "GFX PRO:" Socket 370, 133 MHz FSB, max 1 GiB PC133 RAM, max 1.13 GHz Pentium III (per manual). No AGP slot (integrated Nvidia graphics).
ALADDiN TNT2 was followed by several more Socket 370 chipsets. Have not found an ISA motherboard with an ALADDiN-Pro 4/5[T] or CyberALADDiN[-T] chipset for Intel CPUs.
Example with ISA: Biostar M7MIA: Socket A, 266 MT/s (133 MHz DDR) FSB, max 2 GiB PC2100 DDR RAM, AGP 4×, max 1.4 GHz Athlon 1400C (Thunderbird) CPU (per Biostar).
AMD-760 = AMD-761 northbridge + VIA 686B or AMD-766 southbridge.
None found.
This blog post from September 2012 says that an ISA/PCI bridge with working DMA is still available for embedded systems using AMD G-Series APUs; however, "DMA is achieved by virtualizing the DMA programming interface using firmware imbedded in the BIOS."
The following are Socket A, 266 MT/s (133 MHz DDR) FSB, max
1.5 GiB PC133 RAM, AGP 4×:
Abit
KT7A (KT133A): Max 1.733 GHz Palomino Athlon
XP 2100+ CPU (mobo v1.3 only, per CPU-Upgrade).
DFI AM75-TC (KM133A): Max 2.133 GHz
Thoroughbred Athlon XP 2600+ CPU (mobo rev. DH0 or later only, per
CPU-Upgrade).
KT133A = VT8363A northbridge + 686B southbridge. KM133A appears to be KT133A with integrated S3 Savage4 graphics. The cited DFI example also has an AGP slot.
Although some KT133A boards work with Athlon XP CPUs, it is more common for them to stop at the 1.4 GHz T-Bird Athlon 1400C.
2014-07: One EBayer is selling a AXDA2800DKV4C processor (Barton 2800+) with a Chaintech 7AJA2 motherboard. I have no corroboration from Chaintech or CPU-Upgrade that this is compatible. If it works, it would run at the same frequency as the Thoroughbred 2600+ (2.133 GHz) but it has a larger cache. It is the last Athlon XP processor specced for a 266 MT/s FSB.
Related chipsets:
Example with ISA: ASUS P5S-VM: Super Socket 7, 100 MHz FSB, max 768 MiB PC100 RAM, max 450 MHz K6-III CPU (per manual and CPU-Upgrade). No AGP slot (integrated SiS graphics).
The available multipliers on the P5S-VM would allow for a 500 or 550 MHz K6-III if such exists ("AMD chose not to sell a 500 MHz or faster K6-III after the rare 500 MHz K6-III had been immediately recalled..."), but compatibility is claimed only up to 450 MHz. FSB overclocks up to 133 MHz can be attempted.
Have not found an ISA motherboard with a SiS chipset for Slot A or Socket A (SiS 730).
Example with ISA: Gigabyte GA-5AX: Super Socket 7, 100 MHz FSB, max 768 MiB PC100 RAM, AGP 2×, max 550 MHz K6-III CPU (per manual, with caveats).
The manual lists settings for 500 and 550 MHz K6-III processors, but with an asterisk saying "these processors have not been tested yet!" (In fact, only the 400 MHz version appears to have been tested.) FSB overclocks up to 140 MHz can be attempted.
Caution: Nvidia says that AGP in the 1541 chipset is unusable due to severe system stability problems.
Have not found an ISA motherboard with an ALi chipset for Slot A or Socket A.
None found with declared support for C3 CPUs + ISA slot except industrial boards (q.v.). Only i810* and i815* appear on the C3 verified motherboard list.
Caution: There is no information either way on whether any of these boards support ISA DMA.
AAEON IMBM-700: VIA CN700 chipset, 800 MT/s FSB (advertised), max 2 GiB PC2-8500 DDR2 RAM, max 2.0 GHz C7[-M] (advertised). No AGP slot (integrated VIA graphics). Since a 1.0 GHz, 400 MT/s CPU is soldered on, the chipset's capabilities for FSB and CPU frequency are academic.
The following Socket 370 industrial boards with ISA slots are said to support the C3 in some places, but that information is sketchy and not corroborated by manuals or other evidence:
Example with ISA: Gigabyte GA-6VTXE-A: Socket 370, 133 MHz FSB, max 1.5 GiB PC133 RAM, AGP 4×, max 800 MHz C3 (per manual).
FWIW, 800 MHz is the limit of the Samuel 2 (C5B) core. Higher multipliers are available, but definitive compatibility information is hard to find. Gigabyte doesn't include the C3 in its CPU compatibility list at all even though it's in the manual.
Related chipsets: PN133T (a.k.a. TwisterT) appears to be Apollo Pro 133T or something like it with integrated S3 graphics and maybe a VIA CPU soldered on. An example with ISA slots is DFI GVPC3.
Example with ISA: ECS P6STMT: Socket 370, 133 MHz FSB, max 1 GiB PC133 RAM, max 933 MHz C3 (per ECS). No AGP slot (integrated SiS graphics).
None found with declared support for C3 CPUs + ISA slot. Only M1651T/M1535D+ (ALADDiN Pro 5T / MaGiK 1) appears on the C3 verified motherboard list.
Nvidia nForce chipsets: None found with ISA. The first nForce chipsets arrived in 2001.
Chipsets by OPTi were already obsolete.
The Socket 7 CPUs Cyrix MII, IDT WinChip, and Rise mP6 were also in circulation, but are unlikely to be of interest. The Cyrix MII and IDT WinChip are ancestors of the VIA C3 (originally VIA Cyrix III) Joshua and later cores respectively.
AFAICT, Transmeta CPUs didn't appear in normal desktop PCs.
The maximum CPU supported depends on the specific motherboard model, the hardware revision of the motherboard, and the version of BIOS that is on it. Also, it may be the case that only certain steppings (hardware revisions) of a given CPU will work. Therefore, both the reported maximum and the actual maximum may vary significantly, and buying the fastest CPU that was reported by somebody to work is always risky.
The safe CPU options are those that are declared to be compatible in the motherboard's manual. However, the manuals are finalized before the hardware is, so they tend to lag behind. More responsible manufacturers will keep a CPU compatibility list updated on their web site with clear indications of which revisions of what you need to have in order for it to work. Unfortunately, few manufacturers keep that information up indefinitely for vintage parts.
Whether it was the root cause or merely the innocent victim of Intel-centric engineering by other manufacturers, VIA hardware of this era was often problematic for users. The 686B southbridge that was prevalent at the twilight of ISA became the most notorious of all after an epic conflict with the Sound Blaster Live!.
Some chip revisions were luckier than others. For the unlucky, successful workarounds often required drivers and patches that were made only for Windows, starting with the famous 4in1 drivers (more accurately described as a service pack).
I have an MVP3 motherboard with a 596B southbridge and three different motherboards with 686B southbridges. All four have serious problems related to DMA. Merely playing a 44.1 kHz WAV file from the hard drive under Windows 98 SE or Linux is an effective test. Without software workarounds, the simultaneous use of PATA UDMA and ISA or PCI DMA produces obvious stuttering/crackling in the audio or a hard lockup every time.
The Linux kernel's workarounds are not as effective as the 4in1 drivers. Detuning the hard drive down to PIO mode (kernel parameter libata.dma=6) avoids hard lockups with ISA sound cards at least, but then the hard drive performance really sucks.
This site suggests messing with the BIOS setup options for CPU to PCI Write Buffer (= Disabled), Latency Timer (= 12 or lower), and AGP Fastwrite (= Enabled). That site suggests messing with the BIOS setup options for PCI Delay Transaction (= 0), PCI Master Read Caching (try both), PCI Device Timer Latency (= Disabled), and PCI Delayed Transaction (= Disabled). In my experience, monkeying with these settings has occasionally sparked false hope as the lockups would randomly disappear for one session, only to reappear on the next boot.
AnandTech wrote, "From what we can gather, the VT8231 South Bridge is actually identical to the 686B and is traditionally called the 8231 only when used in a ProSavage chipset."
The following two excerpts are quoted from the NVIDIA Accelerated Linux Graphics Driver README and Installation Guide rev. 2010-11-22 as found with version 173.14.37 of the 32-bit Linux drivers.
Chapter 8. Common Problems
There are some known timing and signal integrity issues on ALi chipsets. The following tips may help stabilize problematic ALI systems:
- Disable TURBO AGP MODE in the BIOS.
- When using a P5A upgrade to BIOS Revision 1002 BETA 2.
- When using 1007, 1007A or 1009 adjust the IO Recovery Time to 4 cycles.
- AGP is disabled by default on some ALi chipsets (ALi1541, ALi1647) to work around severe system stability problems with these chipsets.
...
Problems that Will Not Be Fixed
ALi chipsets, ALi1541 and ALi1647
On ALi1541 and ALi1647 chipsets, NVIDIA drivers disable AGP to work around timing issues and signal integrity issues.
Excerpted from drivers/pci/quirks.c rev. 2013-04-12:
/* The VIA VP2/VP3/MVP3 seem to have some 'features'. There may be a workaround
but VIA don't answer queries. If you happen to have good contacts at VIA
ask them for me please -- Alan
This appears to be BIOS not version dependent. So presumably there is a
chipset level fix */
static void quirk_isa_dma_hangs(struct pci_dev *dev)
{
if (!isa_dma_bridge_buggy) {
isa_dma_bridge_buggy=1;
dev_info(&dev->dev, "Activating ISA DMA hang workarounds\n");
}
}
/*
* Its not totally clear which chipsets are the problematic ones
* We know 82C586 and 82C596 variants are affected.
*/
DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_VIA, PCI_DEVICE_ID_VIA_82C586_0, quirk_isa_dma_hangs);
DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_VIA, PCI_DEVICE_ID_VIA_82C596, quirk_isa_dma_hangs);
DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82371SB_0, quirk_isa_dma_hangs);
DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_AL, PCI_DEVICE_ID_AL_M1533, quirk_isa_dma_hangs);
DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_NEC, PCI_DEVICE_ID_NEC_CBUS_1, quirk_isa_dma_hangs);
DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_NEC, PCI_DEVICE_ID_NEC_CBUS_2, quirk_isa_dma_hangs);
DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_NEC, PCI_DEVICE_ID_NEC_CBUS_3, quirk_isa_dma_hangs);
Although VIA is a repeat offender, Intel's PIIX3 southbridge (used on 440FX) and ALi's M1533 (used on ALADDiN IV or IV+) made the list as well.
Watch out for this stuff:
Tiido Priimägi noted that the IO signals on all post-Pentium motherboards are 3.3 V with 5 V pullups while on older motherboards they were just 5 V. The pullups introduce some lag. This is a possible explanation for why some very old or quirky ISA cards just refuse to work in newer motherboards. Most cards should trip at 2.4 V because of TTL compatibility.