danger den rbx test at overclockers
 

http://www.overclockers.com/articles886/

he tested it with the ehiem 1260 but tested all the others with a 1048.

how can you compare the rbx in the same test results as the others when they use differant pumps ?

Well, he does at least provide the pressure drop measured for the different test results.

IF you ignore the heat put into the system by the pump, there is some basis for comparing the numbers.

The way I see it, the RBX performs about the same as the Swiftech MCW5002 IF you provide seven times the pressure drop. Not good, especially considering the prevalence of GPU blocks these days.

It looks to me like, the RBX is not a contender.

My comments copied from the liquidninjas shitstorm thread:

And keep in mind JoeC tested with the LEAST restrictive of the inlet inserts!

I wish Joe would have tested out all of the nozzles and with the same pump in all cases. His 1.0GPM starting flow point is in a spot where most waterblock C/W curves are in their "knee" and so a small difference in final flow rate can have a fairly big impact on the C/W. If the RBX only moved say 0.6-0.7 GPM with with the 1048 then it might perform a lot worse.

Re: danger den rbx test at overclockers
 

Well, ignoring that the 1260 might put in more heat to the water, his testing apparatus makes them perform the same. He's able to control the flow, so that all the blocks were tested at 1GPM, so they are on even footing for his results.

No DMOS. All blocks are tested by adjusting flow rate without the waterblock to 1.0GPM. When the block is added then flow rate decreases at a varying amount (depending upon flow resistance of block). So by using a higher pressure pump (1260) for the RBX it is very difficult for me to wrap my head around how I can compare the data with earlier tests (using 1048). Most of the other blocks have a fairly low decrease in flow anyway; the RBX is much higher even with a higher pressure pump though.

it would be ok if he used both pumps to test all blocks as the other blocks tested could give better results on a 1260 limited to 1gpm like the rbx would give very bad results on a 1048.

what we want to see is how a range of blocks work with the same test setup each time. if it's not the same then it make comparing very difficult.

What we want to see is EVERYTHING
When we want to see it is NOW
What we want it to cost is NOTHING

Gotta make some compromises. From the standpoint of reader and interested third party I don't find JoeC's testing completely satisfying. But when wearing the reviewer hat it makes a hell of a lot more sense than what I am doing (I just spent 8 days reviewing a single heatsink).

wont the 1260 have more pressure than the 1048 even though it's limited to 1gpm ?

I thought that JoeC's results were measured by subtracting the water temperature from the die temperature? The amount of impact that the larger pump would have would then be relative to the secondary heat losses of the system as a result of the warmer water flowing through the loop.

What is important is that one is just gathering a data point at a particular flow rate. That's about all you can take away from the results. Agreed that the results don't really lend themselves to flow vs pressure predictions like BillA's data did, which is really what you're asking about here.

thats whats throwing me looking at joes results and treating them like the flow vs pressure predictions of BillA's data.

I had a grasp for why JoeC was starting with 1.0GPM sans block and then recording how much flow rate dropped along with performance at the resulting flow rate. It isn't the way I personally do it, but it seemed like a good practical approach to streamlining testing. If a block A drops more in flow rate than block B then it's obviously more restrictive. But introducing another variable (a different pump used now) really makes it difficult for me to follow what JoeC is trying to convey with the results. For example the switch from 1048 to 1260 would have effects on final flow rates for RBX>>PolarFlo>MC5002>Slit Edge. Are these effects enough to shift C/W values? Dunno.

Or am I reading JoeC's tests incorrectly? I am assuming from the tables of data that he begins with a flow rate of 1.00GPM without block and that blocks will reduce flow rates by differing amounts depending upon resistance. Or are these two different tests?

The problem with doing what Joe has done is that different pumps obviosuly have different PQ characteristics. How one pump reacts when artificially restricted to 1GPM with a subsequent resistance added is going to be different to how a different pump reacts artificially restricted to 1GPM with the same resistance added.

JoeC needs to settle on either using a fixed pressure drop across the block (ala BillA), a fixed pump (fairly useful to many people with a similar pump - not so useful for flow rate predictions though), or a fixed flow rate and measure the PD at that fixed flow rate.

He's created a bit of a mish-mash of data here. The flow-rates are all close enough to be roughly comparable, but Joe seems to be reluctant to settle on any of the above three scenarios, and has instead settled on something weird that sits in-between.

This is exactly my feeling Cathar; JoeC has a very good test setup but I can't quite figure out how to make full use of the results because of so many variables.

You could look at the valve been set to 1 gpm as the resistance of an artificial rad. I wonder what pressure drop the valve is equivilent to? Though changing the pump would throw that idea out of wack.

The biggest problem I have with the test is the difference in order between the PolarFlo and the split edge waterblock. On BillA's tests the spilt edge is 2C better than the PolarFlo while Joe's test has the PolarFlo 1C better than the split edge block. If the order of blocks aren't consistant on die simulators, then what chance do they have of being consistant on actual cpu's? If there is no consistancy, then what good is the data?

Another problem I have is that the temperature gradient across the surface of the cpu may be alot greater than with a die simulator. It may possibly even change patterns with different load programs. One block could perform better than another with prime95 and perform worse with cpuburn. I haven't seen any data that proves the cpu temp gradient is consistant.

This has been discussed before. BillA's test of the PolarFlo was with the initial versions with the rather "casually" lapped bases which weren't that flat, as was noted in Bill's review of the block. Joe tested the PolarFlo with the new "superfinish". The difference between the two blocks in terms of the lapping would easily explain a ~2C difference.

Temperature patterns are not consistent across a CPU die at all, however this tends to be more of an effect on the final overclock, which is something that cannot be measured. One block may deal better with "CPU hot spot" activity than another block (typically due to using finer scale "furniture" within the block), yet both may give fairly similar CPU temperatures. Since CPU overclock stability is often closely linked to how hot the hottest bit of the CPU gets, then this is probably the real side-effect of measuring different blocks on a fixed die-simulator.

Unsure of how large such an effect is. I know that Les has made a few efforts of calculating such for different block design types.

For now though that's a problem that's probably best left in the "too hard" basket.

Thanks, I must have missed that disscusion. It's interesting that lapping made such a large difference (more like ~3C). Think I need to go buy some sandpaper.

Edit: I just reread BillA's test and it seems to contradict what you just said. Do you have a link to where it was discussed that the block wasn't that flat?

You will have all sorts of problems when you use CPU loading programs that don't run completely in the cache. Then you are at the mercy of the motherboard's communication with hdd and ram and other complications. Reproducibility isn't so good at that point.

And as the main difference between Prime95 and K7Burn is that the first accesses system memory, I doubt that one would result in a different ranking than another on Athlons or P4s.

It brings up an interesting point though: How about the integrated memory controller of the A64? How hot does that get, and could whether a program accesses system ram or not make a big difference in temps? Something I hadn't thought of before (wish AMD were as free with tech docs as Intel is).

I believe that I tested a version of the PolarFLO prior to the current one
- one would have to ask Steve at Horse of Iron

Steve, BTW, is in a class with only N8 as the other member

Hmmm, must've been thinking it as I looked at the pictures. Bill indeed said that it was slightly convex and very flat. (My apologies for misquoting you Bill). I do notice that it does fall away from the middle quite rapidly.

Compare BillA's MCW5000A picture (somewhat flattish all over), to BillA's PolarFlo (which has a rapidly falling gradient of non-flatness away from the very middle).


MCW5000A

http://thermal-management-testing.com/S5fringes.jpg


PolarFLO

http://www.thermal-management-testing.com/PF3fringe.jpg

Maybe it's just a matter of personal interpretation. I don't pretend that I can do a whole lot better myself when I lap blocks, but I do believe that this is an area that does need more exploration.

Have heard some conflicting points of view on the matter of what's desirable. Some say convex, especially for the IHS chips. Some say very slightly concave for bare-die CPU's as CPU die's are very slightly convex themselves, and some say that as flat as possible is the best tradeoff.

I tend to agree with flat as possible,

the PolarFLO shown is convex, rather much so
but it is only the area in the center that is of interest
and if one starts counting fringes the two shown are pretty equal

with an IHS, convex will still work - but I too prefer flat

After re-analyzing all the data, it appears that the split edge is the block that's most out of sync with BillA's and Joe's results, not the polarflo. A warped baseplate may be the culprit. Player0's test also seemed to show a poor result for this block.

If the above statement holds true, then I expect the RBX to receive a good asswhooping from the WW by about 1.5C (or more).

EDIT: Cancel that perdiction. I just read Joe's update on the RBX.

I wonder how man blocks DD has put the "accelerator plate" in sideways? http://www.overclockers.com/articles886/ has anyone talked to them about it?

and lets hope he uses the 1048 to test this time.

Thanks BillA :)

the test has been updated, puts the RBX at a c/w of 12:drool: