Steve Foster's Water Cooled Peltier Dual CPU System

Update 3/20/00:

Two CPUs at over 1 GHz – will the motherboard melt?

Persistance has its rewards: A little BIOS rewrite and now the 1/4PCI settings are available from the W177H PLL-IC Chip. Now with the PCI @ 1/4 133MhzFSB and 143MhzFSB will boot into NT. Ran two Seti@Homes concurrently to see if the system would be stable. Had to raise the Peltier voltage to 13.8V and add 1/2oz of Redline Water Wetter to the coolant, to get -15C at the CPU core, -27C at the cold plate.

The hardest part was changing the voltage on the CPU, just had to finger nail polish VID1 (A120) and VID2 (A119), the problem was VID3; it had to be connected to Vcc so I had to solder a little wire from B120 (VID3) to B119 (VID0) to get 1.75V at the core.

At 133MhzFSB the AGP is running at 88.67Mhz, but the G400 Max seems to live with it. Ran both Seti units in 6.2hrs and they downloaded new work units.

At 143MhzFSB the AGP is running at 95.33Mhz, could get it up, run Seti, but when I loaded 3D Studio Max, CRASH AND BURN… DAMN AGP!!! Next step is trying to figure out the serial write from bios to get me the 1/2 AGP that is available with the W177H PLL-IC, then maybe I can get to the 148Mhz FSB and the 153Mhz FSB settings.

It would be kick-ass to get these 700 Coppermines to run at 1071Mhz on a dual board without the aid of the +$1000.00 *ntel i840 dual boards or the Server Works Chip Set boards that have only 100Mhz FSB and 133Mhz FSB settings. I am pumped and will shoot for the moon, gonna order a pair of 800Mhz Coppermines to see what they can do……..:)

Update 3/17/00: Tons of e-mail on the “What’s the case?”; go here:
Cal PC Super Server

Also email on the PCI slots; go here:
13PCI slots

Finally, Page 2 has some more PIC’s of the whole system:

on to page 2…

Note: Two CPUs @ 866 MHz each – Smokin’!

One of the most difficult “overclocking” projects is that of a dual CPU configuration. They tend to have the greatest problems due to the GTL+ bus. I have had some success with the TYAN S1837UANGR Thunder mobo and peltier/water cooling. Since this machine is used for rendering projects and video production editing, I needed major PCI slots. With the Aurora XP-7 PCI expansion board and the TYAN, I have 1 AGP, 13 PCI (11 usable), and 1 ISA. With all the cards in the system and WincrapNT, getting all the drivers and equipment to run in a “overclocked” configuration is a REAL challenge.

The CPUs are buried in the duct work.

The TYAN Board uses the Cypress (IC-Works) W177H PLL-IC which allows for 16 FSB settings up to 153Mhz. One of the problems that TYAN incorporated into the Bios is that when you selected anything over 100Mhz, the PCI divider remained at 1/3. Since the W177H had eight settings over 100 MHZ with a 1/4 PCI divider, it was either recompile the AMI Bios or Serial write to the PLL.

In comes H. Oda’s SoftFSB. After contacting Cypress, I was able to get the data sheets on the W177H and write a custom PLL-IC file for SoftFSB. So far we have run super stable at 124Mhz FSB setting with the PIII 700Mhz Coppermines or 866Mhz. Once I have figured out how to disable the IOAPIC1 clock, just keeping the IOAPIC0 clock, I should be able to run at 133Mhz FSB, this is verifiable since I can load and run WINSh*t95 at 133MhzFSB but not WincrapNT 4.0. Haven’t tried WinDump2000 since I don’t have all the drivers yet for all the hardware.

Lots of insulation to prevent condensation.

I started out with 4 TE Distributing ICE-71 peltiers and custom heatsinks and cooling shrouds but ended up with water-cooling and custom exchangers to get the CPU’s at a stable -10C. I am using 2 Sanyo 120MM fans (127CFM) with a used Yamaha motor cycle radiator, 2 pass, 2.28″ deep. It will run at 10C at 124Mhz FSB by setting the voltage at the peltiers to 8.5V, but once I get the PLL settings fully figured out, I will up the core voltage to 1.7V on the CPU’s and see if I can push over the 133Mhz FSB settings.

Detail on the CPUs.

The GX chipset should be able to run at 153Mhz without problems since Tyan located the Memory slots and CPU across from each other, substantially reducing the trace lengths on the Thunder Mobo. The only drawback to the Thunder Mobo is that the slot 1 connectors are placed so close together (to reduce trace lengths). Overclocking with just “big” heatsinks just ain’t gonna happen.