The i860 Conspiracy

I’ve been thinking of acquiring a board with the Intel 860 (Colusa) chipset. This chipset is historically interesting because it was Intel’s first chipset for NetBurst Xeons, and–at least according to Intel–the only chipset that supports the original Foster Xeon DP processors with the 180nm Willamette core.

The platform is interesting because it was Intel’s first dual-socket Pentium 4 implementation, and the i860 chipset was also the first with support for certain modern amenities like Message Signaled Interrupts (MSIs), enabled by the switch from a dedicated APIC bus to interrupt delivery via FSB messages.

The catch is that the i860 chipset was relatively short-lived, having been cursed with RDRAM. The i860 was introduced in May 2001, and in February 2002 it was already superseded by the DDR SDRAM-based E7500 (Plumas) chipset, which also coincided with the release of 130nm Prestonia Xeons based on the Northwood core.

The i860 was apparently so short-lived that Intel did not manage to release its own board based on the Colusa chipset. There were apparently only four board vendors who did: There was Supermicro P4DCE and P4DC6 (I am guessing that P4DC stands for Pentium 4 Dual Colusa); there was Tyan Thunder i860 (S2603); there was MSI 860D Pro; and there was an obscure Iwill DX400-SN.

The MSI and Iwill boards appear to be very hard to find. The Supermicro and Tyan boards are not, but there’s a catch.

There are currently several sellers offering Tyan i860 boards on eBay. The prices range from $650 to $1,025. The Supermicro P4DC6 is available at prices from $475 to $804. The Supermicro P4DCE is comparatively cheap, at “only” $337 to $382. And it’s not one seller, it’s several, in several different countries.

Now seriously, what is up with that? These boards are not only old but weird. They are almost 20 years old and not, in any typical sense of the word, useful. Who is crazy enough to pay $500 or $1,000 for such a board? It does not make sense.

But wait, it gets even weirder. Supermicro and Tyan weren’t the only companies selling i860-based boards in larger quantities. There was Dell (Precision 530), Compaq (W6000), and HP (x4000).

The HP x4000 workstations reportedly used Tyan Thunder i860 boards. There is currently one complete HP x4000 on eBay for $29, no doubt with a Tyan i860 inside. Compaq W6000 boards are available from $39, typically well under $100. And there’s one Dell Precision 530 board on offer with CPUs and RAM with a starting bid of $9.99. Now what is up with that? Those price levels are of course much more in line with expectations for this sort of gear.

But really, what’s going on with the prices of the i860 boards from Tyan or Supermicro? Is this some conspiracy, trying to squeeze desperate business buyers who need to replace old hardware and don’t know any better? Just random attempts at profiteering? I don’t know. But I’m quite sure the Supermicro and Tyan i860 boards are not made of pure gold, which might justify the prices.

As an aside, the the Rambus-based i860 boards had decent memory bandwidth, but they were severely limited in the amount of RAM they could support. The usual maximum was 2GB RAM, and only boards with a memory repeater hub (MRH) could support additional 2GB RAM for a total of 4GB. Pentium III based Xeon boards could already handle 6GB or more at the time, and Intel’s E7500 (released 9 months after the i860) supported up to 16GB DDR SDRAM.

At any rate, I’m planning to get a Compaq W6000 board with CPUs and RAM for a reasonable price (under $100 for the whole lot). Which again makes me wonder who would pay north of $500 for an i860 board…

Update: My Intel 860 (Colusa) chipset board has arrived, complete with two Foster Xeon 1.7 GHz processors. The 1.7 GHz variant was the fastest in the first batch of NetBurst Xeons released in May 2001. The board should be upgradable to Prestonia Xeons at 2.8 GHz; there’s a 3.0 GHz Prestonia model with 400 MHz FSB, but that wasn’t supported in the W6000, likely due to its higher power/thermal requirements. The board came with 1,280MB ECC RDRAM which I promptly upgraded to 2GB (the maximum the board supports).

I had to get a matching WTX power supply for the W6000 board. There’s a 24-pin and a 6-pin power connector, and they look a lot like ATX but aren’t. Then I had to guess how to turn the power on—there’s a poorly marked 11-pin header with a single row of pins (pin 2 is not populated), located right next to the IDE connectors, possibly labeled ‘*P5’. Shorting pins 6 and 7 does the trick and turns the system power on and off. I could not find the pinout documented anywhere, either because it simply wasn’t ever documented or because I didn’t locate the right manual.

Minor surprise: The W6000 has an onboard Ensoniq ES1373 chip, also known as Sound Blaster 128. It has no ISA slots and only 3 PCI slots, but it already has onboard Ethernet, IDE, and SCSI.

It took me a couple of tries to find a graphics card that the W6000 board likes. Since the W6000 was a workstation, not a server, it has on-board audio but no video; it has an AGP Pro slot, something that’s not at all common on server boards.

Fun fact: A “full boot” takes more than five minutes to test 2GB of memory during POST. If the boot logo is enabled, there is absolutely no progress indication and the system looks completely stuck, because the memory test is done before the keyboard is initialized.

This entry was posted in Intel, PC hardware, PC history. Bookmark the permalink.

15 Responses to The i860 Conspiracy

  1. Chris M. says:

    ebay is part of the problem. They have the whole “pay no listing fees until it sells” deals, so overpriced crap clutters your search results. Supermicro boards in particular attract this stupid high pricing. I have a rather ordinary P6SBA that ebay sellers will post for hundreds of dollars. Its a bog standard 440BX board that they sold tons of.

  2. rm says:

    I am surprised you would call it i860. i860 is this: https://en.wikipedia.org/wiki/Intel_i860

  3. Michal Necasek says:

    Some call the chipset 860, some call it i860. The confusion with the Intel 860 processor is hard to avoid either way. I certainly did not make up the designation, it was common back in the day. See here or here or here.

  4. Michal Necasek says:

    I bet you’re right, they’re playing lottery because it’s free.

    I wonder if Supermicro as a very business-oriented brand attracts this kind of behavior because sometimes someone really needs a 20yo board and is willing to pay two or three times the price of a new one. Tyan is similar with its server boards.

  5. John Elliott says:

    From a retrocomputing point of view, the P6SBA is great because it can disable the floppy controller and use a plug-in FDC, and the BIOS works fine with pre-VGA video cards. But I suspect that isn’t what’s driving the eBay market — were they perhaps used in some embedded industrial application and now there’s a demand for spares?

  6. Yuhong Bao says:

    Also note the ICH2 B4 stepping, which fixed an errata allowing I/O APIC to be enabled in all desktop PCs as required by PC2001. (The original Pentium 4 in November 2000 launched with the ICH2 B1′ stepping without I/O APIC enabled)

  7. Michal Necasek says:

    Ah, that’s erratum 19, “Intel ICH2 Entry into S1/S3 (Intel 850 and Intel 860 Chipsets Only)”. Implication: “Effects [sic] the use of the I/O APIC feature in the ICH2 when running compute intensive, high priority application under an ACPI and APIC enabled OS resulting in a system hang or delayed entry into S1 and/or S3.” And “Fixed in B-4 stepping of the ICH2.”

    The APIC and I/O APIC was not used all that much in single-processor systems (99% of them) around 2000, so they no doubt decided disabling the APIC was acceptable. With the dual-socket 860 chipset machines it was a different story.

  8. I was hoping this was something like that ancient dual i386/i860 motherboard thing. I guess it’s only hope was the NTen project, but apparently performance of NT on the i860 was so horrible it never even made it to the 1991 pre-releases.

    I couldn’t get my dual P2 board to fire up at all, but now I’m wonder if it was WTX, not ATX….

  9. Michal Necasek says:

    It’s kind of hard to imagine how the i860 performed when NT ran like treacle on a 386. I actually wonder when they dropped the i860 support in NT, I don’t think there’s any sign of it in any of the pre-releases, let alone the released stuff. Which would imply that NT on i860 didn’t last very long.

    Hmm, was your board an L440GX+ or something? They had some really weird power switches IIRC. Pretty sure the power supply wasn’t WTX, but it might have been something ATX-but-not-quite.

  10. GL1zdA says:

    IBM also made a 860 workstation, the IntelliStation M 6850. And there was a bigger Compaq, the W8000 with 860 (i guess it used the MRH-R compared to the W6000).

    The problem was of course RAMBUS price. That’s why some smaller PC manufacturers put together workstation with Athlon MP (since 2001), others pushed Pentium III with dual CPU capable VIA chipsets, some tried to use the ServerWorks GrandChampion chipset (in 2002). 2001 was also the year, when Intel pushed its Itanium workstations, all based on the Intel’s reference design (Dell 730, SGI 750, IBM IntelliStation Z 6894, HP i2000, Fujitsu-Siemens CELSIUS 880). So there were quite a few designs competing with the 860 and most were simply cheaper. It wasn’t until Intel released the E7505 (Placer) chipset, which could use DDR RAM, that they regained control of the workstation market (not everywhere, NEC used the GrandChampion in their Express5800 workstations until 2004 with Gallatin Xeons).

  11. Richard Wells says:

    The RISC i860 resulted in some very funny stories on how inaccurate benchmarks could be. The 30% boost in Dhrystones because strcpy was aligned on 8 byte structures was representative. IIRC, the theoretical performance claimed by Intel was about 10 times the performance most people could get.

  12. Michal Necasek says:

    Yes, the Compaq W8000 supported 4GB RDRAM and must have had the memory repeater. It also had 64-bit PCI slots, so it must have had the additional 64-bit PCI hub.

    The one interesting thing about the W6000 is that initially it was available with Foster Xeons up to 1.7 GHz, but later supported up to 2.8 GHz Prestonias. That should have greatly increased the system’s performance, which I suspect was not all that stellar at the beginning.

    ServerWorks was interesting, and one of the most interesting things about them is all the Intel boards for both Pentium III and Pentium 4 class processors with ServerWorks chipsets. The 860 chipset is a good illustration why, it could only support 4GB RAM with the additional memory repeater, while ServerWorks boards supported 6, 8, 12GB RAM easily with no extras required. Though I suppose with the RDRAM prices being what they were, 4GB was the most anyone could afford, anyway 🙂

    We all know that Intel chipsets were a big mess around 2000, with large holes in several market segments (Pentium III chipsets? Who needs that?). A lesser company might easily be sunk by so many bad decisions, but Intel’s deep pockets saved them time and again.

  13. Yuhong Bao says:

    AFAIK 4GB with the memory repeater means 512MB RDRAM RIMMs based on 288Mbit RDRAM, which was more expensive per bit than 128/144Mbit RDRAM in 2001.

  14. Michal Necasek says:

    Yes, 512MB RIMMs were the biggest supported by the 8xx chipsets (I can’t figure out if bigger RIMMs were ever available). There are 4 slots on the MCH and optionally another four slots on the repeater.

    I’m pretty sure there were 1GB SDRAM DIMMs available at the time. For example the Intel SDS2 board (dual PIII, also from 2001) supported up to 6GB with 6 slots right on the board. And there were certainly 1GB DDR DIMMs, supported e.g. by the VIA Apollo Pro 266. The Supermicro P3TDDE, also available in 2001, supported 4GB with 4 slots on the board. I suspect that Tualatins could easily hold their own against Foster Xeons, but not against faster-clocked Prestonias.

  15. Yuhong Bao says:

    AFAIK there was 576Mbit RDRAM chips available, but Intel’s chipsets never supported it.

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.