SUMMARY: Cool-Computer’s chipset waterblock can yield significant gains in FSB speeds.
The good guys at Cool-Computers sent me a sample of their chipset waterblock to try out. This little gem is sized to cool one of the performance stumbling blocks on your motherboard – the Northbridge chipset. Installing it is no big deal (outside of removing your mobo from the case), as shown below:
After removing the Lasagna cooler I had on my Iwill KK266, I used small bolts to hold the waterblock to the board. You MUST use non-conductive washers on the back side of the board! This waterblock uses 3/8″ nipples, so it fits in perfectly in my system – I spliced it so that one hose from the CPU block goes to the chipset waterblock, then the return back to the radiator from the chipset waterblock.
In testing the Cool Computer’s unit, I tested with and without a peltier. Now, there is a reason for this, as shown in this email from Benjamin Berglihn, from Norway:
I haven’t read much on the net yet about KT7A’s that have reached high FSB’s, but i think mine does quite well 🙂 Anyways, here’s my setup:
Abit KT7A (no RAID)
Duron 800 (unlocked by me)
CardExpert GF2 MX (5,5ns RAM, does 220/220)
128 MB 133MHz CL3(?) Mushkin Mosel RAM
+++ the normal stuff
As you can see, I’m no millionaire, but I like to see myself as this hardware’s worst nightmare! 🙂
I’ve read a little bit about KT133A vs DDR chipsets, so i wanted to get the FSB as high as possible, of course. To my surprise, this 133MHz CL3(?) RAM did Turbo/CL2@150MHz!! Had to set the I/O to 3.6V – no problems.
At 152, I had to set CL3, but then I couldn’t get any higher.
RAM? Chipset? Hmmm, I had to find out, and since theres no PC150 RAM available around here (Norway), I decided to use my FOP32/52W peltier on the KT133A northbridge! Lot’s of mods to the FOP32, but it fits, right next to the FOP32 on my Duron – looks like a Dual-something!
Anyways – FSB did 155, then 160; normal I/O 3,8V – 165 to 170!!!! Thats a 340MHz system-bus, not bad!! But, Windows registry error; had to reinstall the whole thing! But I’m not sure: Was it the RAM or the Chipset (145 + 25, HD maybe??) I want a PC150 stick!!
So thats my experience with the KT7A. What do you think, is 165 FSB and stable any good? Or is this people’s average results maybe?
Now this is interesting! We all know that cooling the Northbridge is a MUST, but will supercooling do us better? So I accepted the mission and proceeded to do this:
What you see is a 40 watt TEC installed directly on the chipset – I have a copper coldplate around someplace, but I wanted to try this out so I slapped it on without one. Also notice there is no insulation – you will need it if you run this setup much below ambient. Anyhow, the objective here was to find out what benefits, if any, we can get from better cooling this little hot spot.
And these are the results from Si Sandra 2001, below :
The Iwill KK266 will now post at 170 FSB with Crucial’s PC133 at CAS 3, although I can’t represent that it is absolutely stable. At CAS 2 and the fastest settings in BIOS, I can run at 160 FSB. With just plain water cooling, I can run at 155 FSB, which is about 3 MHz more than what I can do air cooled. I measured chipset temps with a probe on the back of the motherboard; with the TEC, back temps at 170 FSB were 10-11 C, and with just watercooling, 40-41 C.
AlSO NOTE that running at 170 CAS 3 gives you LESS THROUGHPUT than running 160 FSB CAS 2.
A LOT of money is being spent on RAM that purports to be PC150, PC166, PC170 and even PC180 by our readers, with the expectation of significant performance improvements.
There are two issues that are working against significantly higher FSBs:
- The higher the FSB, the more likely peripherals (NICs, video cards etc) will not work due to out of spec PCI speeds;
- Unless the Northbridge is supercooled, it will most likely be the stumbling block before your RAM poops out.
Ed wrote up a piece about PCI HERE – it’s some common sense about what to expect when you really push FSBs hard. I have done some RAM tests, and the results I found were bounded by what you can expect on your motherboard, NOT what the RAM can do in some isolated RAM tester. After all, we run RAM on motherboards.
What YOU can expect from your motherboard with a particular slice of “high performance” RAM may be better or worse than what we find in our tests; however, MOST of you will find results similar to what we find. I am going to do some more “extreme” RAM testing with the TEC cooled Northbridge to see how these sticks perform. However, unless you’re really aggressively cooling the Northbridge chipset, good quality PC133 CAS 2 is probably going to do you fine.
So I am indebted to Benjamin Berglihn for alerting me to just how key Northbridge cooling is – I’m sure some of you have similar experiences, so send them on in and let’s see what’s possible.
Thanks again to Cool-Computers for sending the chipset waterblock to test out. At a minimum, water cooling the Northbridge should give you more stability than air cooling. If you have a crummy OEM chipset fan and use air cooling, waste it and get something like a Lasagna chipset cooler.