So finish, as rated by manufacturers, is irrelevant, due to other factors making more of a difference. It would also be of no use to test that, as optimally, you would want something to mate perfectly with the CPU, which is, of course, NOT flat, and not perfectly smooth. Having a perfectly flat, perfectly smooth surface would only be useful if the other surface was the same. That's what I'm seeing, I'm not sure if you're saying that, though. So, the only practical reason that manufacturers hand lap to a high grit would be look, so you get compliments, like Cathar up there? Now, with hand lapping, you can get a better finish, but a more circular, and thus less flat, surface, correct? I personally don't think that I could afford anything that could measure the temperature difference between a finely finished, flat surface, and a fairly well finished, fairly flat surface.

If you "dry lap" with 600-grit and push down hard, it will wear away the grit and give you a good mirror finish.

"Wet lapping" (I use low-smell kerosene) and low-moderate pressure will give an appearance like I described.

One thing I've been noticing is that the harder one pushes down, the more likely the surface will be non-flat, developing visible rippling which is easily apparant by sitting the block on a table, sitting about 1 meter away, and looking at the reflection of a straight object a few meters away. Move your head slightly and for a surface that has a mirror finish achieved through high-pressure lapping on 600-grit, you will have horribly noticable rippling. My suspicion is that this is caused by frictional heat flexing the metal, which expands and bows out-wards slightly. This "protruding" metal gets hotter 'cos it's now receiving the bulk of the lapping, and so gets worn away more. When you stop, the metal cools and contracts again causing the rippling.

I have had trouble lapping with grits higher than 1200-grit. Using kerosene tends to "suck" the block onto the paper firmly, making it very difficult to move the block about. Dry-lapping tends to generate the "rippling" effect. Lapping at 2000-grit paper requires a LOT of patience, and taking things very, very slowly. When I wet-lap with 2000-grit paper, I end up with a very dull surface. Not even remotely shiny. This, to me, indicates that it is probably fairly flat, as pulling the block off the wetted paper takes quite a deal of effort. If I then follow up with some Brasso/Silvo, the mirror shine comes out very, very quickly, and the reflection test indicates a surface as flat as the human eye is capable of determining. If I dry-lap, I will get a mirror finish from the moment I pull the block off the paper, but the reflection test will indicate rippling.

I use water with a bit of soap... I can see my face in it, but I can also make out the tiny scratches you spoke of. So they're correct in saying to not push down, to just let the weight of the block do the work. Doesn't polish take out some of the thermal conductance? Or is this to make people think its a better finish? (I keep forgetting that you're in marketing... Sneaky...)

I try not to polish my blocks. Doesn't mean that I haven't done it from time to time to attempt to achieve a mirror finish just for the hell of it.

Often I find myself re-lapping my polished mirror finishes again to take away the shine, as crazy as that sounds, it's true.

I haven't found any evidence that a mirror finish is any better than a non-mirror finish, so long as the block is flat. Of course, flatness is something that is extremely hard for someone like myself to measure. I've carefully lapped blocks with 240-grit paper and achieved similar performance to a 1200-grit lapped block.

Mirror finishes are "wank value" IMO. Flatness is where it's at. Doesn't mean that I won't try to provide a mirror finish if possible. I often find myself in conflict with what I believe to be adequate (within reason of my meagre ability to make a piece of metal "flat" around the CPU die area), and what people percieve as good with wanky mirror finishes.

Hey, if it's shiny, it has to be good! It's in our basic human nature, or why are diamonds so treasured? I'm no marketer. I'm just prone to the same desire for shiny objects like anyone else, despite what reason tells me.

Wait... What is your job then? Yeah, like I thought, it looks good... Yeah, you're buying the best block in the world... Why? Oh... It looks good?!

What I believe and what I provide for people's wants are two different things. A marketing person won't tell you that though. I do, and then I give them the option.

Joe has invited corrective comments, via email, in regards to the article.

Have fun guys.

Reticular implant aside...

You want to keep the flow speed under 5 feet per second, otherwise the pressure drop becomes significant.

Using Hazen-Williams and a friction factor of 140 (smooth copper), here are the pressure drops at various flow rates, for a 3 foot long section of tubing:

1.0 gpm (60 gph)
1/4": 34" H2O pressure drop, 6.5 fps (feet per second) flow speed
3/8": 4.7" H2O, 2.9 fps
1/2": 1.2" H2O, 1.6 fps

1.5 gpm (90 gph)
1/4": 72" H2O, 9.8 fps (!)
3/8": 10" H2O, 4.4 fps
1/2": 2.5" H2O, 2.5 fps

2.0 gpm (120 gph)
1/4": 122" H2O , 13.1 fps (!)
3/8": 17" H2O, 5.8 fps (!)
1/2": 4.2" H2O, 3.3 fps


It's really not hard to see (even without an implant!) that loosing 10" of pressure would be a real waste of pumping power, when 1/2" tubing would only drop 2.5 inches. Given that our pumps are relatively weak, especially on the pressure side, IMO, every bit counts. As a bonus, if one ever upgrades the pump, 1/2" tubing can easily handle up to 3 gpm, dropping about 9" H2O of pressure, resulting in a flow speed of 4.9 feet per second.


Note: Hazen-Williams may not be the best formulae to calculate this, but should be pretty close to real figures.

PS: I believe I've posted the Excel sheet with this formulae. If interested, I'll post it again.

Ben I would counter that flow rates above 1.5-2GPM are sometimes still beneficial but no longer practical. Let's assume that you are willing to increase radiator size and fans to keep water temp constant and remove that from the equation. Even still, the cost/performance benefit for going from 0.5-1GPM is usually large, the benefit for increasing from 1-2 is much smaller (and may be completely offset by pump noise), and the step from 2-3GPM is probably not worth it to water coolers.

I did the calculation too, tho i used Haaland's correlation to calculate the friction factor (which change with Re and hence velocity) and used a e=0.0015mm (drawn plastic tubing) and got results similar to yours. But that was not my point. My question is rather why the 1.5gpm mark? Why not 1.3? 1.4? 1.6?... and so on. I mean, for pretty much any flow rate, the loss in a 3/8 tubing will be around 4 times what you will have with 1/2 tubing. The 1.5gpm mark you choose seems a little arbitrary. The best that can be done with results like these, imo, is to show the curves you calculate and let others choose for themselves.

I'm going to go out on a limb here. With certain waterblocks I have seen overclocking stability benefits that outweigh the apparant difference in CPU temperatures at various flow rates.

For example, say I run at 4LPM vs 10LPM, and observe a ~2C CPU temperature difference between those two flow rate points at the same water temperature. I then set my CPU overclock to a near unstable point with 25C water. If I turn off the fan on the radiator and let the water temperature climb for both, the 4LPM flow-rate will become unstable at a lower CPU and water temperature than the 10LPM experiment. I've observed the 10LPM flow-rate to generate stable overclocks with up to 3C warmer CPU and 5C warmer water temperatures over a 4LPM flow rate.

I do not observe this behavior with thick-based blocks, but with thin-based impingement blocks.

Why?

Now overclock stability is a shaky thing at best to run with as the rate of instability can be statistically mapped out as a Poisson distribution with the chance of a computational error occurring being time-based, but if I can repeat the experiment multiple times (and I have) then this lowers the statistical chance that what I'm oberving is just a co-incidental anomaly.

Indeed, it would seem that I absolutely require 2+GPM flow rates to pull off some of the more extreme overclocks that I am able to eke out of my CPU's, which I admit is part of the reason for my hunt for good strong 12V pumps that I can over-volt. I believe it enough to spend money on it, despite the temperatures not being dramatically better, the overclocks certainly are.

^Hmm, I've never heard of this. Maybe your impingement blocks ability to take heat from the die area faster than thick-based blocks, coupled with the high water flow makes the cpu more stable regardless of temps, maybe takes the 'harmful heat' (if there is such a thing, lol) away quicker?

My best guess would be the old standby "CPU heat isn't uniformly distributed". Perhaps there are some local hotspots which limit overclockability and which are not next to the diode of CPU? The complication here of course is that baseplate temp is needed to understand what's going on but in a thin base wb that can't be obtained...

and taking the question one additional step:

the rational behind a thick bp is to provide thermal buffering from transient heat spikes, and the benefits of said thermal capacitance in the bp can be shown in terms of the attainable oc

as Cathar has observed, another means of accommodating thermal spikes is with a thin bp, BUT with 'rather' high flow (certainly as compared to that flow rate needed for a thicker bp)
- this high flow/convection rate is necessary to inhibit the temp rise of of the bp in response to spikes

worth noting is that it can be the case that a thicker bp with a higher CPU temp
may in fact yield a better oc when compared to a thinner bp

many ways to skin a cat

Bill and Cathar:

Have you played around with watching die temperatures rise/fall upon turning pumps on and off? It's remarkable how quickly that the thin bp waterblocks respond.

Your hunt for pumps cathar... have you tried car pumps, thet are 12 volt pumps, aren't they. I do not know almost anything about car cooling pumps but... maybe sb does? It could possibily be a very cheap source of pumps (car scrap yards).
Second thing, I was entertaining thought of buildong a water pump from scratch? Has anyone ever tried it (oc community)?

yes, one tends to focus on getting the switching sequence in order

especially when an actual CPU is being cooled. Certainly much different from watching the effects of fan delay on a thick bp copper heatsink

Yes, I've tried this. Thin BP blocks don't offer much in the way of thermal buffering, however, they still offer plenty of time (many minutes) for any thermal protection circuitry to kick in.

Bill is right in that under certain circumstances that a thicker bp will offer higher overclocks than a thinner bp, but I find that this is also quite closely linked to the flow rate and the design. I regularly test from 1LPM - 10LPM, and have yet to observe a thicker bp block offering a higher overclock than a finely structured thin-bp block in that flow-rate range at anything over 2LPM. For unstructured thin-bp blocks, definitely, that's the case. By "unstructured" I mean basically largish expanses of flat plate convective area.

This is one of the things I'm playing with for the XXX, that being an increase of the bp vs "cup base expanse" ratio, and balancing that off against the thermal resistance of the material in an attempt to more evenly "smooth" out any temperature spikes.

With most things for this waterblock gig, there are many ways to approach the problem, and it is apparant that there is a great deal more going on than just a bulk C/W value when it comes to overclock stability.

Almost all car pumps are belt driven from the main motor rotational action. Very few regular cars actually use 12V electric pumps to circulate the water. Some of the more modern cars will use small pumps to circulate water after the engine has turned off, but such pumps are typically quite noisy and not really rated for continual use. They are only meant to kick in for a few minutes at a time when the car is turned off.

The one 12V car pump that's commonly available is the Davies-Craig EBP, and I actually own one of these pumps. Performance wise it's just a little better than an Eheim 1250 (~2.4mH2O pressure head, ~22LPM peak flow). It's actually rated for continuous use, being designed as a booster pump to continuously boost water flow for high performance engines instead of driving the flow from a belt pump attached to the engine.

Davies-Craig also make the EWP, which is intended as a serious water pushing system for high-performance engines and has some pretty impressive stats, but it's only rated for about 6 months of continuous use, is pretty expensive (~$250US), draws a lot of power (comparitively) and is quite noisy.

well, yes but not quite
the newer Intel thermal design guides discuss CPU heat sensor controlled fans (they are coming);
and make the point clearly that NO fan implementation can respond appropriately to thermal spikes as they occur too quickly

the thermal protection (by Intel) cuts in within cycles, literally faster than the temp response of the bp

what is not so clear to me is if the OCing measurement is also being diddled by the cut clock cycles
-> I suspect that Prime (for example) will run w/o errors while being 'temp protected' by cutting clock cycles;
so if clock cycles are being cut, is the indicated operating freq being reduced correspondingly ?
(a rather lot about this that I do NOT know, eh)

anyone messed with this ?
pHaestus - did we not discuss something about this some months ago ?

We mostly return to the same topics ad nauseum with no resolution (x equations, x-1 knowns will lead to this).

I'm not TOO familiar with Intel chips and AMD has no throttling. AFAIK there is no program that both maximizes heat generation and checks for errors. Intel used to have a good heat generating app but I suspect they are a bit gunshy w/ current state of Prescott. The discussion related to pelt cooling a P4 to a range of temps and benchmarking right? Would be even more interesting with current infernos...

There are also fuel pumps. There was a little review on overclockers.com quite some time back, but the thing was noisy and not very effective (I think it wasn't effective... There was some other problem other than noise)

I was of the understanding that the cycle-reduction style of thermal-throttling offered by Intel CPU's could be disabled within the BIOS for many of the modern overclocking focussed boards. My Socket 478 P4 board is a little older and doesn't have that BIOS option, but I've been assured by many of the 3DMark overclocking fiends that they have the option in the BIOS and turn it off regularly as a matter of ensuring that they get the highest possible benchmark scores.

I was, however, more referring to critical thermal shutdown, as in 135C internal die temperatures for P4's, in that a CPU still does have sufficient time to react to critical shutdown temperatures.

For AMD, the Abit NF7-S board does offer voluntary thermal throttling of AMD CPU's through monitoring of the CPU internal diode, but it does not allow you to read what that diode is reporting. See section 4-4 of the NF7-S manual.

By far the majority of my testing is done with non-thermal-throttling AMD CPU's.

Found it. Read section 3-6 of the Abit IC7-Max3 mobo manual which may be downloaded here.

Basically this allows you to turn thermal throttling on/off for P4 CPU's.

Yes, I picked 1.5 rather arbitrarily, in gpm flow rates, in 0.5 increments, for the sake of round numbers, but see below for more

If I was going to be picky, I'd say 1.7 gpm: that where the flow speed in 3/8" tubing still falls under 5 fps, but the drop at that point (over three feet length) is still 11 inches.

The original figure I found was 8 fps, and that was a recomendation as the upper limit, for PVC tubing (source: http://www.ppfahome.org/pdf/pvcpipewaterspec.pdf ). Also: "A flow speed of 5 fps is the guideline for 1 inch ID and higher.". Then, "uncle" Dave reminded me that "This guideline works fine for industrial pumps, but would cause most aquarium style pumps to struggle. Low speed is the weak pump's friend. Save the high speed for where it is needed, and that isn't in the tubing.", and having run the numbers, I've found that 5 fps makes for a pretty good guideline, considering the pressure drops involved.

Otherwise, I'll still stick to my original figure of 1.5 gpm. As most of us know, 1.0 gpm is "typical" of an average loop (ref: the OC article). Any one of us who puts more care in component selection (Johnson pump aside...:rolleyes: ) will easily achieve 1.5 gpm and that, IMO, dictates the use of 1/2" ID tubing. Plus, as I've said before, 1/2" fittings are much easier to find locally, so it's really a natural choice.

I agree with pHaestus though: at 3 gpm, the cost of the system becomes unreasonable (aka unjustifiable), but then water cooling itself isn't cost justifiable, unless one can put a price on "peace and quiet".


Back to the evolution of this topic...

Yes, a thicker baseplate gives you more "buffering", for those heat spikes. I particularly enjoyed reading Cathar's figures, on OCAU. Cascade's response time is, relatively, frighteningly low, but should still allow for a thermal protection to kick in. It's still much better than direct die cooling... :rolleyes: :p . It's a trade-off: reduce the baseplate thickness, and depend more on the pump to remain operational, not to cook the CPU.

Has anyone messed with Intel's thermal throttling, to see what temps they can hit?

Ben
when you describe a point on a curve as a limit you will be questioned - and found foolish
and when you persist in defending YOUR value judgments against more accurate descriptions, . . . .

a point on a curve is just that, why not call it such ?
(because then you could not sound like an authority spouting 1.5 or 1.3 or 1.8 infinitum ad nausium ?)
this is the same old re-state and re-phrase stuff you used to do (and I guess wish to continue doing)

no | limits | Ben, just associated incrementally higher 'costs'

I hear you Bill.

I'll put together a graph of the pressure drops, for 2', 3' and 4', at 0.5 to 3.0 gpm, for 3/8" and 1/2" ID tubing.

No there is no "set point", it's really down to a design decision. 1.5 gpm is a rough figure I've arrived at, based on the proportion of the pressure drop from the tubing, relative to the total head of the pumps used to achieve those flow rates (reticular implant?).

To qualify my statement, I could compare a few blocks, cores, and pumps, in their various combinations, and present the differences with 3/8 and 1/2". An interesting challenge, but I believe that the wide availability of 1/2" fittings is more important.

Still, might be worth doing...

shit Ben, doing ANYTHING at all is better than continuously spouting what you think
we look forward to some real data and thoughtful analysis
(finally)

Edit: oops was reading this thread and and another and accidently posted to this one instead of the other one :mad:

But I laughed all the way through this article too... OC usualy has much better articles. Too bad he didn't drill through the motherboard! That would have been perfect.