This Is How to Feed It

The question of how to feed an Intel SBT2 board has been answered thanks to a very kind blog reader. Interestingly, there are two different answers. The official one is this:

Delta RPS-350B

The Delta Electronics RPS-350 B power supply comes from a SC5000 chassis and is exactly what the SBT2 board calls for. It has the right 24-pin ATX-style power connector and the right 10-pin special I²C power connector, as well as the uncommon auxiliary connector. It’s a redundant power supply, meaning that the system should survive the failure of one of the two PSUs within the housing (notice the two handles on the back).

With the SSI switch panel (likewise from a SC5000), my SBT2 started up right away (well, almost, see below). Integrated graphics, 4GB RAM (this is a board from 2000!), 1GHz Xeon.

Not content to leave well enough alone, I wanted to see what happens if I replace the redundant SSI power supply with a regular ATX part. And surprise surprise… it worked! But only with the magic switch panel:

SC5000 Switch Panel

Interestingly, the switch panel has L440GX stamped on it—Intel’s L440GX+ server board was the predecessor of the SBT2 (it was a dual Slot 1 board for Pentium II and III processors), and the SC5000 chassis was designed for the L440GX+ board.

I have not been able to power up the SBT2 board by shorting the power-on pin, which is something that works fine with other Intel server boards from the same era, and more or less every other ATX board. On the other hand, it’s enough to connect the P4 connector of the switch panel to get things going.

So what have we learned? The SBT2 board does not require a SSI power supply, and neither the I²C nor the auxiliary connector are required for basic operation. However, there are a few caveats:

  • It takes about 10 seconds before the board reacts to the power switch. Right after turning on the PSU, there is no reaction whatsoever.
  • There are no LEDs or any other indication of whether the board is ready to power up or not, or really whether it’s doing anything (notably no CPU fans).
  • The SSI switch panel does something unusual which probably can be replicated but isn’t at all obvious.

I also learned that ServerWorks was originally called Reliance Computer Corporation, which explains the ‘RCC’ logo and markings on the SBT2 chipset. Newer boards, such as the Intel SCB2, have ServerWorks chips with a different logo and markings.

One thing that still mystifies me is why Intel built Cascades Pentium III Xeons with up to 2MB L2 cache and 100MHz FSB, but the newer 133MHz FSB Xeons only had 256KB L2 cache. It would be interesting to know what the relative performance was of a 900MHz/2MB/100MHz FSB Xeon vs. a 1000MHz/256KB/133MHz FSB Cascades Xeon. Most likely there wasn’t a clear winner.

Acknowledgements: Big thanks to Ivan Veršić for donating an SSI power supply, an older SBT2 specimen with two 933 MHz Xeons, and most importantly the switch panel from an old SC5000 chassis.

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

14 Responses to This Is How to Feed It

  1. Yuhong Bao says:

    I think these 133Mhz Slot 2 Pentium III “Xeon” are dual socket only and was targeted at Intel 840 chipset motherboards.

  2. GL1zdA says:

    The 133 Slot 2 Xeons were workstation CPUs. Identical to the corresponding Coppermine Pentium III, only the connector was different. The 100 MHz CPUs with 2 MB cache on the other hand were successors to the “Tanner” Pentium III Xeons, intended to replace them in all these 450NX servers. Comparing the two doesn’t make sense, because the 2 MB 900 MHz Xeon did cost over $3600 when it was released in August 2000 and the 933 MHz Xeon did cost $794 when it was released in May 2000 – no one thought about choosing one over the other. The x86 workstation/server market split in two lines when the Cascades Xeons were released, before many manufacturers did workstation with the big cache parts. The market remains split today with the dual capable E5 CPUs and scalable E7 Xeons, but it seems Intel wants to merge the two lines after nearly 20 years with the Purley platform.

    The nice thing is, the 900 MHz 2 MB Xeon will often work in old Slot 2 motherboards based on the 440 GX or Reliance chipsets. I have upgraded my SGI Visual Workstation 540 (with the custom SGI Cobalt chipset) with 2 of these and it works even if SGI never sold the workstation with anything faster than the 550 MHz Xeon. Now I need two more heatsinks two add another pair of 900 MHz Xeons to it to make it the ultimate workstation.

  3. Michal Necasek says:

    Of course comparing the two makes sense. Who cares about the pricing, what matters is the performance. There’s no rule that a more expensive CPU has to be faster than a cheaper one…

    Anyway, the above still doesn’t answer the question why there were no big-cache models with 133MHz FSB. It’s hard to imagine that the 133MHz version wouldn’t be faster than the 100MHz one, so there had to be some reason.

  4. Michal Necasek says:

    As a case in point… remember those Mendocino Celerons with 128KB full-speed L2 cache that were for some tasks faster than the much more expensive Pentium IIs with bigger but slower L2 cache? This case is different, as it’s the FSB that’s faster rather than the L2 cache, but it will still make some difference.

  5. GL1zdA says:

    You can compare it for the sake of comparison, but such comparison would be useless in real life.

    There were no big-cache models with 133 MHz FSB because there were no 133 MHz high-end server chipsets. You had the 450NX and Profusion, none of which was 133 capable. The ServerSet HE could probably do 133, but no one sold such motherboard with Slot 2 connectors and there were no servers offering it.

  6. Yuhong Bao says:

    Yea, a quad socket platform needs to last longer, and I think Intel deliberately decided not to do a 133Mhz version.

  7. GL1zdA says:

    @Yuhong Bao Yes, I suspect it was deliberately, because they hoped to push Itanium into the MP market. When it flopped they offered the Xeon MP with RCC Grand Champion Chipset in 2002, it wasn’t until 2005 that Twin Castle (E8500) was released.

  8. Yuhong Bao says:

    The point is that quad socket platforms generally need to last longer because development and validation costs are higher.

  9. Yuhong Bao says:

    “The ServerSet HE could probably do 133, but no one sold such motherboard with Slot 2 connectors and there were no servers offering it.”
    You mean the ServerSet III LE? LE and HE-SL was targeted at the dual socket market and supported 100/133Mhz FSB for normal P3, HE was targeted in the quad socket market and supported 100Mhz FSB for P3 Xeon.

  10. Richard Wells says:

    The 133 MHz Coppermines ran into problems as seen with the 1133 model.

    I see two potential problems with the 2MB Xeon with a 133 MHz bus:
    1) Would the larger cache be stable with the higher bus speed? Of course, the cache could be slowed but that gives up much of the benefit of a fast bus.

    2) Power consumption will be increased a lot. The 2 MB of cache add about 10 watts to the chip. Increasing bus speed to 133 MHz would increase it more, perhaps to about 50 watts total. Can Slot-2 handle that power draw? I don’t know and haven’t found a quick reference to it.

  11. GL1zdA says:

    @Yuhong Bao
    According to ServerWorks the HE supported the 133 bus:
    The Tyan Thunder 2500 is also a HE motherboard with 133 bus, but it’s Slot 1
    The HP VISUALIZE A1280 and A6034 x- and p-class workstation also did 133 with Slot 1 and HE.

    @Richard Wells
    The 2M version of the 450 MHz Pentium II Xeon consumed over 50 W. The 2M 900 MHz Pentium III Xeon consumed over 47W, but the power section was a bit different. For the Xeons with off-chip cache you had to use 3 VRMs for 2 CPUs – the third VRM was shared between the cache chips on both cartridges. For Xeons with on-chip cache, the 3rd VRM was not used (I haven’t tested it yet, but my Visual Workstation 540 should work with two 900 MHz 2M Xeons with the 3rd VRM bulled).
    I don’t think 100 vs 133 MHz would make much difference. If you look at Intel’s “VRM 8.4 DC-DC Converter Design Guidelines” (Order Number 245335-006), you’ll see the Pentium III 700 drawing 14.8 A and the 733 drawing 15.4 A. Even if the Xeon uses more power, there shouldn’t be more that 1-2 W difference between the 900 MHz 2M and the hypothetical 933 MHz 2M Xeon.
    The L2 cache in Pentium II/III Xeons is on the back-side bus – it runs at full speed, but this means it runs at full CPU speed not bus speed. So I don’t think there would be much difference between running it at 900 and 933. And definitely there should be no problem running it at 733.

  12. Michal Necasek says:

    Good stuff. Thanks for writing this down.

    I should point out though that the 133 MHz FSB Cascades Xeons go up to 1 GHz. The SBT2 manual documents frequency jumper settings up to 1067 MHz, but such a CPU presumably never materialized.

    Needless to say, a performance comparison of the 2M Xeons with Tualatins would be interesting too. At least the HE-SL chipset supports those, and they have only 1/4 the L2 cache size but up to 55% higher clock speed.

  13. Michal Necasek says:

    I think you may well be right. See here:

    Intel withdrew the 900MHz/2MB Xeons for a while due to “thermal problems”. That suggests those parts were at the limit and faster CPUs wouldn’t be reliable. And that’s likely why they don’t exist.

Leave a Reply

Your email address will not be published. Required fields are marked *

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