Watercooling your chipset can bring some benefits

MORE NEW mainboards now come with waterblock-ready North Bridge chipset cooling capability - look at the Asus Maximus Extreme, or the upcoming Foxconn BlackOps as examples.

It is a fact that the chipsets can these days match GPUs, and surely exceed the CPUs, in the temperature measurement department when overclocking your PC with high FSB and/or DRAM loads.

But, is it really that bad that water cooling is needed? Will it benefit when we push the fastest DRAM to its limits, and further tweak the NorthBridge latency?

Here was the configuration: the usual supercooled system with Intel QX9770 CPU running off the Asus Maximus Extreme mobo, with 2 x 1 GB A-Data Vitesta DDR3-1900 DIMMs, the fastest I have right now. These DIMMs did up to 7-6-6-13 at DDR3-1840 and 2 volts reliably within Windows, quite an achievement by itself.

I added the Thermaltake BigWater 760i, a simple internal 2-bay watercooling system we reviewed before on the CPUs, directly to Maximus Extreme chipset cooling system's water block connections. Committing a water cooler exclusively for the chipset may seem a waste to some, but it actually isn't - such water coolers are cheaper by the day, and anyway it lets us check the watercooling benefit under optimal conditions.

An unusual "observation point" on the watercooled North Bridge

I ran the system at two settings: 4.27 GHz / FSB1708, and 4.60 GHz / FSB1840. The first setting is "all settings green", a zero-risk long-term productivity run with CPU running at 1.388 volts, and the rest of the stuff under similar 'reasonable' values. The other one is the highest reliable speed I could get, with the CPU at 1.475 volts, and the rest up accordingly - all settings in the mid-range "yellow BIOS band", as seen in the table.

Also, I used an added fan on the extended portion of NB heatsink anyway, so this was never pure "passive cooling", mind you.

Now, with North Bridge running cooler, and the fastest memory, I could do something else too: reduce the Trd BIOS parameter to try and lower the chipset's own CPU-to-memory latency. A reduction by 1 does give some results in both cases, but it only worked at FSB1840 if the NB voltage went up really high - one more reason for water cooling here. Another 1 point in Trd reduction gave no benchmark benefits in either bandwidth or latency, and any further reduction simply didn't boot up.

The A-Data memory managed to keep the settings up all the way as well, even with the low Trd - it did become quite warm at the 1840 speed, though, guess it should be cooled more aggressively too. In any case, new memory record benchmarks here!

And those Celsius temperatures? As you can see, there is a benefit, but it is more pronounced when the push comes to the shove - at the highest FSB. To me, those 7 degrees Celsius saved are a lot if you're running the system for a longer period of time in those overclocked conditions.

In summary, if intending to overclock your FSB and memory to a high degree, water cooling the North Bridge seems a good proposition. Combine it with the best - and well cooled, too - memory to get the maximum speed up from your CPU. µ

.