November 26:

The waterjacket, having had the silicon set overnight, was first tested in the sink.  Success!  No leaks... until we turned it upside down.  An hour and some superglue and silicon later, we tested it again.  No worries this time - not a single air bubble coming from the jacket at all. 

The next big worry was hooking up our water jacket to the peltiers, cold plate, and finally the CPU.  Given that the pipes were sealed onto the water jacket made this a bit of an effort.   Instead of using cable ties to hold everything together, we used jumbo rubber bands.  It sounds pretty crap, but it worked a treat, holding things together beautifully, and making it easier to take things apart again.

Using our lovely thermal paste, we smeared the Celeron 300A and stuck a cold plate to it.  (An offcut of the aluminium sheet ;)) We continued our smearing fun with the peltiers, and slapped them down on the cold plate.  Finally, with great trepidation, we attached our waterjacket, and put rubber bands around the whole thing.  Inserting it into the slot was easier than expected, and even the pipes seemed to sit nicely in the case.

The system in the end... notice the rubber bands are gone
(hehe, dodgy and ugly, the only way!) 

We installed the second power supply inside the case, and wired up the peltiers, by ripping apart old fan attachments, this made the removal and installation of the peltiers very simple.   Unfortunately it meant having two power supplies to switch on, but we could live with that for the moment.  Firstly we turned on the PC without the peltiers or pumps running, and set it to 504MHz, a setting we'd previously been unable to get no matter how many fans we put in the PC.

The moment of truth was upon us.  We turned on the power supplies.  It booted.  It was stable.   Life was good.  Doubly so cuz there was no leaks from the water jacket.   (Glad wrap was placed around the motherboard in case this did eventuate).  The next step?  598MHz of course!  Unfortunately, it appeared koen's 66MHz RAM wasn't up to the task of a 133MHz FSB.  Or anything else above 112MHz.  Never mind - there was still the Celeron 400.

Unfortunately, after being sidetracked by pre-purchasing a Dreamast, we decided, hell, why stop there?   Going out to trusty local PC shop, (Centre Com), we scored a Celeron 433 Retail with PPGA converter for $Aus140 all up.  Time to go burn this baby in.

Sandra showing our benchmark (BTW, here's another image that shows sandra running at 650, but this is taken in win2000, and the result is different from 98.  I'm more inclined to believe the win98 score, as sandra isn't even meant to run under NT.) 

Here's the really exciting bit.  A brand new chip, with 66MHz RAM, with room temperature water cooling... posted at 650MHz straight off.  And with enough time for the peltiers to cool it down enough to frost the cold plate, it not only posted, but it ran with a great deal of stability too.  More so when we chucked in a couple of icy cold VB cans to the esky.  Life was indeed extremely good, considering L2 cache was still enabled (unfortunately at a latency of 10, but we're working on that).

So there you have it.   First day of testing, and already at 650MHz.  We hope to break 700MHz tomorrow.

November 27:

Today we reassembled the equipment at koensayr's house.  Thankfully everything is working, we had a little scare last night with the 433 not booting at all but everything *seems* fine now.  However, at first 650MHz was anything BUT stable.  It'd barely post, let alone be rock solid.  Nevertheless, we soldiered on, and assembled a much cooler water reservoir using a frozen milk bottle of water and a couple of tray fulls of ice. 

Our water reservoir

We set up the pump (which kept popping its connection, which was very annoying).  Another problem we had was that only one peltier would work, which was very very odd, as they both worked the day before.  For some reason, it kept shorting out the power supply.  Cheap-arse $10 A.T. PSUs just aren't what they used to be dammit :)  More fiddling and we solved that problem.

Next we went out to Bunning's Warehouse and bought some cable ties to replace the pathetic rubber band setup we had.  Much better!  Not only did it mean the setup wasn't liable to fall apart, and didn't move out of place from gravity, but it also meant we got a tighter fit, and thus better heat conduction  (see the top photo to see what the system looked like in the end).  With both peltiers, and the cable ties, 650MHz was no longer a problem.  In fact, we were able to get it running VERY stable at 650MHz *with* L2 cache enabled at default latency!  We were really impressed by this point, and 3D Mark and SiSoft Sandra had no worries in running at all.  It wasn't until we noticed how quickly the water reservoir was heating up that we started worrying about stability.   It was fine until over 15 degrees celcius and then stability slowly slid from our grasp.  At 18 degrees celcius, the system crashed running the Synthetic CPU Benchmark in 3DMark 99.   However, before this tragedy occured, we were able to post at 728MHz! 

Posting at 728MHz (mmmm, 728MHz.  BTW, that's me (Vision) holding the digicam and koen leaning over my shoulder ).

With just 10 degree celcius water!  Once we get the bar fridge and the coolant system working we're sure to get 728 MHz to run.



Site design by Michael Davies. Site maintained by Vision and koensayr.