PDA

View Full Version : For Sale Oddity: Supermicro P3SCD i820 board with MTH for SDRAM



dionb
29th November 2018, 16:51
A bit of history:
In 1999 Intel bet the shop on Rambus RDRAM. The chipset designed to bring this to the masses was the i820 'Camino'. It was designed to offer better performance than the i440BX with new features like AGP 4x and ATA-66 and -100 support. Just two problems. Firstly RDRAM was hideously expensive. Secondly the i820 was plagued with major design flaws and bugs. It was such a mess it was called 'Caminogate' in the press. Boards with i820 were subject to recalls twice.
First, it turned out that there were unsolvable stability issues with three RIMM slots on a single channel. All i820 boards with three RIMM slots were recalled. While that was going on Intel realized that the price of RDRAM was simply too high and that they needed an alternative. The alternative was to put a protocol converter, called the Memory Translator Hub (or MTH) on an i820 board to allow the use of SDRAM. This made expensive boards even more expensive, and combined SDRAM's relatively low bandwidth with RDRAM's vey high latencies for awful memory performance.
And then, to make it worse, it turned out there were stability issues in the MTH too, which could not be fixed with a BIOS update as promised. So they had to recall all boards with an MTH, give a RDRAM-based i820 board without MTH in return and - worse for them - replace the SDRAM from the recalled boards with the same amount of hugely expensive RDRAM. This drama gave AMD the opportunity to claim the performance crown from Intel with their Athlon CPU which used plain cheap reliable old SDRAM. Intel moved back to SDRAM, delivered the i815 and later i845 chipsets and then did their best to forget Rambus ever existed.

So, want a piece of that history?

No boards with MTH should have survived the recall, but some slipped through. I have here one of the worst design ideas even from that era: a workstation/server board with i820 and MTH:

Supermicro P3SCD
ATX
Slot1 (Supports Coppermine CPUs)
Intel i820 chipset with MTH (that chip with the heatsink between the i82820 MCH and the DIMM slots)
2x SDRAM DIMM
AGP 4x
AMR
5x PCI
ATA-66, AC'97 audio
141916

Given that Intel withdrew support back in early 2000 when it tried to recall all these boards, good behaviour is most definitely NOT guaranteed, but I actually used this board to run Memtest86 for several GB of PC133 SDRAM and the only errors were from clearly defective DIMMs. It was also able to run a fairly heavy Linux desktop (slowly) with no problems. So maybe Supermicro's famous QA department ensured what left the factory actually worked...

Anyway, despite its historical interest, I have too much stuff here so want to get this a loving new home.

Asking EUR 30 25 20


Paypal and SEPA bank transfers accepted. Board is in the Netherlands, shipping (with track & trace) to nearby countries EUR ~16, outside of Europe probably EUR 26. If interested I'll look up the exact price

beorn
29th November 2018, 17:33
A bit of history:
In 1999 Intel bet the shop on Rambus RDRAM. The chipset designed to bring this to the masses was the i820 'Camino'. It was designed to offer better performance than the i440BX with new features like AGP 4x and ATA-66 and -100 support. Just two problems. Firstly RDRAM was hideously expensive. Secondly the i820 was plagued with major design flaws and bugs. It was such a mess it was called 'Caminogate' in the press. Boards with i820 were subject to recalls twice.
First, it turned out that there were unsolvable stability issues with three RIMM slots on a single channel. All i820 boards with three RIMM slots were recalled. While that was going on Intel realized that the price of RDRAM was simply too high and that they needed an alternative. The alternative was to put a protocol converter, called the Memory Translator Hub (or MTH) on an i820 board to allow the use of SDRAM. This made expensive boards even more expensive, and combined SDRAM's relatively low bandwidth with RDRAM's vey high latencies for awful memory performance.
And then, to make it worse, it turned out there were stability issues in the MTH too, which could not be fixed with a BIOS update as promised. So they had to recall all boards with an MTH, give a RDRAM-based i820 board without MTH in return and - worse for them - replace the SDRAM from the recalled boards with the same amount of hugely expensive RDRAM. This drama gave AMD the opportunity to claim the performance crown from Intel with their Athlon CPU which used plain cheap reliable old SDRAM. Intel moved back to SDRAM, delivered the i815 and later i845 chipsets and then did their best to forget Rambus ever existed.

So, want a piece of that history?

No boards with MTH should have survived the recall, but some slipped through. I have here one of the worst design ideas even from that era: a workstation/server board with i820 and MTH:

Supermicro P3SCD
ATX
Slot1 (Supports Coppermine CPUs)
Intel i820 chipset with MTH (that chip with the heatsink between the i82820 MCH and the DIMM slots)
2x SDRAM DIMM
AGP 4x
AMR
5x PCI
ATA-66, AC'97 audio
141916

Given that Intel withdrew support back in early 2000 when it tried to recall all these boards, good behaviour is most definitely NOT guaranteed, but I actually used this board to run Memtest86 for several GB of PC133 SDRAM and the only errors were from clearly defective DIMMs. It was also able to run a fairly heavy Linux desktop (slowly) with no problems. So maybe Supermicro's famous QA department ensured what left the factory actually worked...

Anyway, despite its historical interest, I have too much stuff here so want to get this a loving new home.

Asking EUR 30


Paypal and SEPA bank transfers accepted. Board is in the Netherlands, shipping (with track & trace) to nearby countries EUR ~16, outside of Europe probably EUR 26. If interested I'll look up the exact price

Great story :)

Then on the other side of the coin you have arguably the two best consoles ever to have used RDRAM - n64 and ps2 :)

dionb
29th November 2018, 19:50
RDRAM itself isn't so bad, it worked fine in i840 and i850 chipsets, the consoles you mentioned and quite a few video cards (Cirrus Logic GD5464, Chromatic MPact(2) etc.). The problem was price due to Rambus' excessive royalty demands - and the totally crap i820 chipset.

dionb
9th December 2018, 02:04
Well, it's a talking point, but no takers - so dropped price.

dionb
29th December 2018, 01:58
Bump again with price drop too.