Waterblock Bench Testing Results

Some interesting observations – Bill Adams

This is a review of waterblock (wb) bench testing results utilizing equipment and procedures previously described in this article.

The focus will be on those performance parameters of significance to WCers in the selection of wbs and understanding the significance of these attributes with respect to the WCing system’s performance. The principal wb characteristics are:

• Pressure Drop as a consequence of the coolant flow rate,
• Thermal Impedance (or "C/W") in relation to the coolant flow rate,
• Thermal Impedance (or "C/W") in relation to the heat load, and

• Thermal Impedance (or "C/W") in relation to the coolant input temperature

The following graphs are from a variety of wbs and illustrate the data that can be developed. As they may also be from somewhat different test equipment setups, the values should not be considered as being directly comparable, unless of course the data appears on the same graph.

The wbs are treated as a ‘black box’ having a certain effect; no appraisal is made of the wb’s internals as to why it may have those performance
attributes. Waterblock design is a completely different topic.

A wb’s flow characteristics affect its specific performance and constitute a significant part of the total WCing system’s flow resistance (together with the radiator, hoses, and any fittings, reservoir, etc.) The pressure drop is not difficult to measure and easily understood by even neophytes to WCing.

But while it is tempting to think that ‘lower is better’, such may not be the case as different wbs’ cooling capability is often inversely related to their pressure drop.

Better cooling due to increased turbulence or surface area
for example, will always increase the pressure drop – one has to ‘dig
deeper’. The comparative pressure drops of some current wbs are shown
below.

When the wb’s thermal impedance is expressed as "C/W" and measured at different coolant flow rates, a curve will result with the "C/W"
decreasing as the flow increases. Due to the improved convection resulting from higher flow rates (actually the resultant increased velocity),

Higher flow will always improve a wb’s performance.

The
"C/W"s of the same wb’s as in the Pressure Drop graph (above) are
shown in Chart 2. Noteworthy is that the wb with the
lowest pressure drop (the Swiftech 462-UH), also has the highest "C/W" (poorest cooling expressed as a function of the coolant flow rate). The three mid-range wbs, in terms of their pressure drop are now seen to have the lowest "C/W" (thermal impedance).

From this, it can be seen that it is not simply flow that does the cooling:

A higher flowing wb is not necessarily better than a lower flowing one.

A WCing myth is shot down.
{mospagebreak}

Bill Adams

Additionally, it is clear that:

There is no such thing as a ‘low flow’ wb that is better at low flow rates than higher ones.

Another WCing myth is disproven.

So what happens if the "C/W" is plotted vs. the pressure drop?

This graph could be interpreted as a characterization of wb "efficiency", i.e. for a given pressure drop, which wb provides the best cooling (has the lowest "C/W"). Prudence is suggested with it’s application, as it is easy to misinterpret.

Here the big 462-UH can be seen to be advantageous, but only in that situation where a high volume pump has extremely low head capability; at higher pressure drops, almost any other wb does better. But
the rest of the ‘better’ wbs, despite their different designs, can be seen to perform quite similarly.

An interesting situation exists with the Swiftech 462-U and the 462-UH as they are the same wb, but with different connection sizes. Comparing their "C/W"s, it is easy to see that flow per se does not do the ‘cooling’, else their positions would be reversed.

To look at the effect of inlet flow velocity (more accurately die area impingement velocity), in place of the very large bore
(0.57 in. ID) inlet connection initially tested with the 462-UH, a Swagelok barbed connector with a 0.39 in. ID was substituted. The outlet was not changed and the standoff from the bp surface was the same for both (but greater than the 462-U).

It can be seen that the wb’s performance with the 0.39 in. ID inlet is substantially improved over the larger connection having lower impingement velocity.

Looking again at the pressure drop (below) for these three wbs, it can be seen that while significant improvement has been made to the 462-UH’s "C/W" (shown above), its pressure drop is still much lower than the 462-U’s. This wb would be an excellent choice for high-flow systems (where all of the other components were selected for minimal flow resistance – a good idea for any system in fact).

For reference a (rather modded) old-style 462 is shown as well.

These types of experiments are easy to conduct (but time consuming) and rapidly sort the ‘good ideas’ from the ‘foolish notions’. More important is the ability, over multiple trials, to identify an optimum, or to evaluate the tradeoff between two interactive effects – which always the case with wbs. Note that for the above case, the really useful graph would be "C/W" vs. flow velocity vs. standoff – it’s been done too.
{mospagebreak}

Bill Adams

Much effort has been expended discussing the relative merits of one wb (design) over another when used for ‘low’ or ‘high’ powered CPUs.

The graph below of the response of the Cooltech WB75, a small, circular labyrinth type wb, should end such debate. The superimposition of one power level curve on top of another conclusively demonstrates that the
"C/W" of a wb is unaffected by the applied power (so long as the die size is the same). Many wbs have been tested, and they all respond quite the same.

There is no such thing as a ‘good’ low power wb being relatively worse at higher heat loads.

And yet another WCing fable bites the dust.

The same data used in the preceding graph can be also be plotted (below) as the die temperature vs. the heat load, a representation of perhaps greater interest to users interested in predicting CPU temperatures. The CPU’s temperature rise above the coolant temperature is easily known by subtracting the (standard) coolant temperature of 25.0°C from the values shown.

And again, presented to show the effect of changing the pump size/flow rate, the same data can be plotted as the die temperature vs. the flow rate:

Composite graphs tend to be a bit confusing, but can often enable one to ‘see’ a relationship that might be obscure. In the case of the Surge wb shown below, looking at the decreasing temperature differential between the die and the wb bp at higher temperatures, one can understand that the
thermal impedance of the TIM joint (it accounting for the difference between the two temperatures) is non-linear. It is somewhat inversely related to
temperature.

Otherwise, the graph illustrates nicely the interplay between flow, pressure and temperature. The actual wb and die temperatures are the values shown plus 20 and 43°C respectively, which was subtracted for graphing. (Note that the intersections of the curves have no significance.)

The general effect of temperature on "C/W", T/W, and the TIM joint C/W can be seen (along with some other parameters) in another composite graph below. Of interest is the virtually unchanging die/coolant temperature differential, even over a 25°C span. From a testing perspective this indicates that the absolute coolant (inlet) temperature is not too important, so long as that temperature is held constant, of course.

As has been illustrated in Chart 6, this may also be said concerning the heat load when plotting "C/W" vs. flow rate. But observe that the pressure drop does, of course, increase as the coolant temperature is decreased due to the changing viscosity.

The decrease in "C/W" as the coolant inlet temperature increases above the ambient temperature of 26°C is largely due to the increased convection from the wb body as more of the applied heat goes to the ambient. This accounts for the T/W change as well. While a change is also apparent to the TIM joint’s C/W, the reasons for such are not now known.
{mospagebreak}

Bill Adams

As previously noted, some of the rather widely perceived ‘truths’ in the WCing world are incorrect. What has been demonstrated is:

• A wb having a lower flow resistance is not necessarily better than another with a greater resistance
• More flow through a given wb will always increase ‘cooling’
• Greater coolant impingement velocity will improve a wb’s effectiveness (where such is applicable)

• A good wb, or a poor one, is such at any and every applied heat load (for the same sized heat source)

The practical import of this testing is that given the appreciable benefit possible due to even moderate coolant flow rates (over low rates), every effort should be expended to identify and select components having low flow resistance.

The present emphasis in pump selection on ‘flow rate’ must shift to an increased awareness of ‘head capability’.

There is little point in running a pump at 1/5th its free flow rating because it does not have sufficient pressure capability. Those WCers interested in performance must accept the need to understand – and use – pump P-Q curves to evaluate what will be most effective.

Performance costs money, why would WCing be any different than other ‘go fast’ activities?

Where the purpose of WCing is to decrease noise, then the criteria for pump selection shifts slightly, but the very same factors will determine the system’s performance.

It must be stated again that this kind of testing does not identify a ‘winner’, although clearly some wbs are inferior in some (or even all) respects as compared to others. It is up to the WCing system designer to identify those wb characteristics of significance, establish their relative importance and to then optimize the tradeoffs and compromises necessary to select complementary components of the WCing system.

Again, thanks must be expressed to the reviewers of this article, and in particular Antoine Dechaume for his insistence upon clarity and completeness.

be cool

Bill Adams