Swiftech Apogee GT vs the D-Tek FuZion CPU blocks @ CES

[Petra]

Now wouldn't that make the block retailers happy?

Not in the slightest... because that would mean that we'd have to stock all (or most) of them. It would be bad for the same reason that these motherboard-specific chipset blocks are a PITA.

[gabe]

You can't argue with the results. Flat was 8c worse than bowed. The bowed base has also helped the FuZion have better performance as well. It works better with lapped CPU's as long as you're careful to get a good even mount.

Hello Gentlemen!

I think I should clarify one thing that seems to escape everybody. The first mount was absolutely equal in terms of pressure between the two blocks, yes -BUT the amount of pressure was grossly unsufficient -my fault: we damaged a mobo the night before by torquing these water-blocks to the max, and I was gun-shy. We were using regular nuts, counting turns, and measuring spring height. Torque value=unknown; whereas we normally use acorn (blind) nuts providing a torque value calculated to ~50 lbf, which is within Intel specs. So we cannot generalize and conclude that there always is an 8C difference between flat base and bowed base. In fact, our own test data measured in controlled environment shows an average of 2C at 100 Watts between flat base and bowed base. See preview of our tweaking guide here (it will be live tonight): http://www.swiftnets.com/products/apogee-tweaking.asp

This being said, I also want to take of advantage of this post to say that Danny's block is indeed a great product, and that Danny is a great guy. He and I come from the same [hard] school: we started the business from our garage and have many things in common. For all intents of purposes, the performance level of these two blocks is very comparable. It is at the peak of current water-block technology. I do wish D-Tek the best success with the Fuzion, just as I know that the Apogee GT will also please many many users :-)

Cheers

Gabe

[Cathar]

Not in the slightest... because that would mean that we'd have to stock all (or most) of them. It would be bad for the same reason that these motherboard-specific chipset blocks are a PITA.

Not to mention if the CPU manufacture gets streamlined, altering IHS flex characteristics slightly, and now you'd have to be stocking CPU batch specific blocks, not just CPU model specific blocks.

[Edit: I'm not intending to knock forwards progress - but I hold a high degree of reservations about a block that has been deliberately curved returning to a non-flat state and then being applicable for mounting on a different CPU]

[SiGfever]

Not in the slightest... because that would mean that we'd have to stock all (or most) of them. It would be bad for the same reason that these motherboard-specific chipset blocks are a PITA.

True, that would be a stocking nightmare.

[Turtle 1]

Bowed bases? Convex bases? Warped IHS's?

What is this? Send in a hunchback to do a hunchback's job?

What happens when the next batch of CPU's come out with a revised IHS design that is flatter?

This is exactly why I always preferred bare-die and/or die-sim testing. No such dependencies upon manufacturing variances to obtain solid repeatable results. Building blocks is already "horses for courses" enough, without then tweaking to perform the best on an singular individual CPU (or batch of CPU's) with their own inidividual quirks.

What is greater truth here? Shall we start making individual CPU specific water-blocks? Buy a CPU and get the matching w/b?

Ya I am getting confused by some replies here. If I clamp 2 flat solid surfices together. and have good contact between both. Than I get a better results using a bowed or convex surfaces. I am sorry there something else is occurring here. The hot rock theory holds true Fact. I can visualize a convex surface matting up to a flat ihs . If enough clamp pressure is used the convex surface should flatten out . But at the convex contact to the ihs that force would apply more pressure on the core at the convex surface. This is were the better heat transfer is occuring.

[gabe]

Bowed bases? Convex bases? Warped IHS's?

What is this? Send in a hunchback to do a hunchback's job??

We are dealing with an observable reality here.

Our rationale is simple: deliver a flat water-block as the pre-assembled solution, and provide extra gear to tweakers so they can do what they do best: tweak. Why not? If it is POSSIBLE i.e. if it works, why not do it?

If you condone the practice of giving different nozzles to adapt the performance of a given w/b to a particular die, why reject/question the practice of working on the TIM joint?

[Garrett]

Interesting... I assume the "bowed" base will become straight(er than the flat base) when tightening the block down... and indeed, talk about stocking nightmares

[Turtle 1]

Hello Gentlemen!

I think I should clarify one thing that seems to escape everybody. The first mount was absolutely equal in terms of pressure between the two blocks, yes -BUT the amount of pressure was grossly unsufficient -my fault: we damaged a mobo the night before by torquing these water-blocks to the max, and I was gun-shy. We were using regular nuts, counting turns, and measuring spring height. Torque value=unknown; whereas we normally use acorn (blind) nuts providing a torque value calculated to ~50 lbf, which is within Intel specs. So we cannot generalize and conclude that there always is an 8C difference between flat base and bowed base. In fact, our own test data measured in controlled environment shows an average of 2C at 100 Watts between flat base and bowed base. See preview of our tweaking guide here (it will be live tonight): http://www.swiftnets.com/products/apogee-tweaking.asp

This being said, I also want to take of advantage of this post to say that Danny's block is indeed a great product, and that Danny is a great guy. He and I come from the same [hard] school: we started the business from our garage and have many things in common. For all intents of purposes, the performance level of these two blocks is very comparable. It is at the peak of current water-block technology. I do wish D-Tek the best success with the Fuzion, just as I know that the Apogee GT will also please many many users :-)

Cheers

Gabe

Correct me if I am wrong . But your torque values is that 50lbF or 50 lbI. 50 flbs is pretty high . Could you link me to Intel specs for clamping force of the thermal block.

[Cathar]

We are dealing with an observable reality here.

Our rationale is simple: deliver a flat water-block as the pre-assembled solution, and provide extra gear to tweakers so they can do what they do best: tweak. Why not? If it is POSSIBLE i.e. if it works, why not do it?

If you condone the practice of giving different nozzles to adapt the performance of a given w/b to a particular die, why reject/question the practice of working on the TIM joint?

Design a large vs small die focus blocks is one thing. Got a single core-centric CPU? Use this block. Got some large split-die thingy? Use that block. I see nothing wrong with that. Do it right though, and in a world where almost all CPU's are now IHS covered, and we may not even need two blocks - one block with a sufficiently large cooling area would do. That's pretty much all that I asked for - was like a Rev3 of the jetted approach tuned for the modern CPU. Given that IHS-less CPU's are all but gone, could happily discontinue the old item.

Bending and warping a block design to suit is another matter altogether. Will the block remain unwarped? Has irrevocable warping occurred to the block's base? What if the user buys a different CPU and the IHS now demands a slightly convex base-plate to work better? Can the base-plate now be warped to be convex? What if it starts becoming rippled after all this bending in different manners? Just how much are we messing with the thermal interface layer across repeated remounts? Will performance now degrade as a result?

I'm not knocking a progressive approach to watercooling advances, but to me, fiddling around with base-plate warping is a near anathema to an individual waterblock's long term use across different CPU's, and perhaps, even on the same CPU across multiple remounts.

[STEvil]

Everyone seems to be forgetting that IHS's suffer from horrible deflection on both the top and bottom due to the manufacturing process used to produce them (they are stamped from a solid piece of metal).

Even if you lap the top side, the bottom is still convex/concave.

[Turtle 1]

Well I will bow to your guy's experiance. But I want my blocks flat along with my IHS. You can take this other stuff and what your calling more advance idea,s and tell these guys what you will . But this is just plane wrong. I want flat mateing surfaces for best contact area. IF AMD and Intel have problems with their IHS . They should fix them . Using a convex or a bowed block to correct IHS defiencies isn't progress its a step backwards. No amount of proven results will ever change my mind on what mechanical engineering has already proven to be correct.

[cdelong]

Everyone seems to be forgetting that IHS's suffer from horrible deflection on both the top and bottom due to the manufacturing process used to produce them (they are stamped from a solid piece of metal).

Even if you lap the top side, the bottom is still convex/concave.

we can do nothing about the underside of the IHS, but lapping the top does ensure the best possible contact patch with a flat WB base. as long as we are making comparisons using the exact same CPU/IHS in tests, the the underside profile does not matter for comparsons sake. Yes, temps may be better from one CPU to the next, but that would not be a fair comparison for testing.

[Barb]

Copper is relatively soft, but I initially didn't believe we were dealing with forces sufficient enough to warp the base to make such a difference.

People are arguing that a convex base will make better contact because it will tend to flatten out over a flat IHS.

Whereas a flat base and flat IHS wont be as effective because the mounting will tend to relieve pressure at the center of the interface.

Are we forgetting that the IHS is also made of copper?

IF copper really is that malleable, then wouldn't the the base and IHS connect equally well whether or not there were small differences in geometry?

[Turtle 1]

Copper is relatively soft, but I initially didn't believe we were dealing with forces sufficient enough to warp the base to make such a difference.

People are arguing that a convex base will make better contact because it will tend to flatten out over a flat IHS.

Whereas a flat base and flat IHS wont be as effective because the mounting will tend to relieve pressure at the center of the interface.

Are we forgetting that the IHS is also made of copper?

IF copper really is that malleable, then wouldn't the the base and IHS connect equally well whether or not there were small differences in geometry?

I am sorry but if you look at the D-Tek casting . mechanicly speaking the mateing surface and the mounting surface are on a parrallel but not the same plane. that casting has a vertical wall that meets the mating surface. Where the metering block mounts to the base is a differant plane. Its much more rigged than the apogge block . I can see where installing a larger diameter o-ring would cause deflection of its mating surface.With the apogge.
The same does not apply to the D-tek design . I highly suspect that the convex surface of the D-tek block is the result of the casting process. Fact is I believe the D-tek block when it comes out of its mold is actually concave . and is mechanicly straightened to a convex with a press. The reason I believe this is the mass of the pins . In the cooling process would cause warp towards the mass.

[gabe]

Bending and warping a block design to suit is another matter altogether. Will the block remain unwarped? Has irrevocable warping occurred to the block's base? What if the user buys a different CPU and the IHS now demands a slightly convex base-plate to work better? Can the base-plate now be warped to be convex? What if it starts becoming rippled after all this bending in different manners? Just how much are we messing with the thermal interface layer across repeated remounts? Will performance now degrade as a result?

I'm not knocking a progressive approach to watercooling advances, but to me, fiddling around with base-plate warping is a near anathema to an individual waterblock's long term use across different CPU's, and perhaps, even on the same CPU across multiple remounts.

Your personal objections are noted, and disagreed with.

Fiddling is in the nature of the tweaker. Resurfacing a block (or a CPU) on a glass table using various grit sand paper will provide acceptable results when done properly.

A genuine debate could be rigid base plate vs flexible one. Both have their advantages and disadvantages. Base thickness affects thermal resistance as you well know. Various means are available to get best of both worlds.
Economics need also to be factored in. Price of copper sky high encourages reduction in use of this material. CPU configs change rapidly. Water-block designs have to adapt. What was true yesterday is no longer true today, and will also change in 3 months.

We see something that truly works, we disclose it to the public, and allow people to try it, that's all. I was the first in the industry to deliver perfectly flat base plates remember? So I do like flat... but again, we are dealing with an observable reality here.

[darcon]

This is getting ridiculous. I would actually like to see real-world tests performed by "regular joes." What I'm seeing here is almost like a synthetic benchmark race to the highest numbers, where little tricks and tweaks are used.

[gabe]

Well I will bow to your guy's experiance. But I want my blocks flat along with my IHS. You can take this other stuff and what your calling more advance idea,s and tell these guys what you will . But this is just plane wrong. I want flat mateing surfaces for best contact area. IF AMD and Intel have problems with their IHS . They should fix them . Using a convex or a bowed block to correct IHS defiencies isn't progress its a step backwards. No amount of proven results will ever change my mind on what mechanical engineering has already proven to be correct.

Nothing is true forever. Real progress is understanding, then adapting. All things are transitory in nature.

[Cathar]

Nothing is true forever. Real progress is understanding, then adapting. All things are transitory in nature.

So you are then effectively saying that there should be a one-block one-cpu approach.

Where we are differing is in the extent of that adaption.

Sometimes there is such a thing as "too much rope".

Still, a successful commercial model argues that it is necessary to adapt to every twist and turn with a new or modified product. It's a self-fulfilling mechanism. I understand where you're coming from Gabe, but don't expect everyone to agree that warping a base-plate is the ideal approach to descend.

In that, we'll have to agree to disagree.

[gabe]

Even if you lap the top side, the bottom is still convex/concave.

Arguable. since joint between IHS and silicon is now soldered, amount of deflection in the contact patch area could remain negligible. Whatever the case may be, the problem is there is nothing we can DO about it :-(

[Turtle 1]

Your personal objections are noted, and disagreed with.

Fiddling is in the nature of the tweaker. Resurfacing a block (or a CPU) on a glass table using various grit sand paper will provide acceptable results when done properly.

A genuine debate could be rigid base plate vs flexible one. Both have their advantages and disadvantages. Base thickness affects thermal resistance as you well know. Various means are available to get best of both worlds.
Economics need also to be factored in. Price of copper sky high encourages reduction in use of this material. CPU configs change rapidly. Water-block designs have to adapt. What was true yesterday is no longer true today, and will also change in 3 months.

We see something that truly works, we disclose it to the public, and allow people to try it, that's all. I was the first in the industry to deliver perfectly flat base plates remember? So I do like flat... but again, we are dealing with an observable reality here.

Look this all came up because the casting process of D-tek block caused warpage in the casting that mechanicly had to be pressed to a convex surface inorder to fix a poor casting process that just happened to work well with a dual die core.

Don't wave a flag like it was engineered to work this way. It was luck plane and simple. Most of the customers of these blocks don't want to do a bunch of tweaking to get good results nor do they want to over torgue their components to get these results. Correct your casting process and supply flat surface parts.
I think the D-Tek block is great. I feel the same about the Apogge as well as others. As Cathar already mentioned copper has some very unusual properties. This high totted convex surface will dissappear very fast with just a little usage. One properity that copper is not good at is it memory. It will conform to its surface very rapidly. D-tek should be happy copper has low memory or it could revert back to its casting concave form.

[gabe]

So you are then effectively saying that there should be a one-block one-cpu approach.

why not? If we could make the block cheap enough? If watercooling goes mainstream cost of production could drop to a few dollars. Then OEM's would certainly adapt each solution to their particular needs.

I understand where you're coming from Gabe, but don't expect everyone to agree that warping a base-plate is the ideal approach to descend.

In that, we'll have to agree to disagree.

That's the nature of a good debate isn't it?

[Praxis1452]

I'd actually agree that if performance is what you want then a one block one cpu aproach could work just fine so long as of course it's cheap as gabe mentioned.

The convex base was used with the different gasket.

If cpu's perform better with convex waterblocks then I'm all for it. It doesn't matter what principle you believe in only results actually matter. If you don't want a convex base either lap it flat or change the gasket. Lower preformance is your choice to make.

[Graystar]

You can't tell me that dipping a small portion of a hot rock (IHS) into a pool of liquid (WB), would cool the hot rock better than dipping the entire hot rock (IHS) in the pool of liquid (WB).

Just as it would be silly to say that hot water placed in a freezer will freeze faster than cold water placed in the freezer...yet under certain conditions that's exactly what happens.

We live in a complex world. Nothing should simply be dismissed outright.

oh...what if your pool of liquid was hotter than the rock? The the statement would be true!

[gabe]

Look this all came up because the casting process of D-tek block caused warpage in the casting that mechanicly had to be pressed to a convex surface inorder to fix a poor casting process that just happened to work well with a dual die core.

For the record (phylly_boy can confirm) I announced the bow technique first and clearly presented it as a one week old discovery. No engineering flags here. We plain and simple stumbled on this, then decided to hand it over to the community. After I made that announcement, Danny remarked that his block was also bowed, and provided some explanations that do not coincide with your above assumptions, but I will let Danny answer this if such is his inclination.

Most of the customers of these blocks don't want to do a bunch of tweaking to get good results nor do they want to over torgue their components to get these results.

From a statistical point of view, probably true. But I also know for a fact that the top 30% of our customers are tweakers and will spend uncounted amounts of time to try to improve the perf of their systems by any means. The mere scale of what we have recorded (2 to 3C for us is huge) warrants that we disclose it to our extreme crowd.

Correct your casting process and supply flat surface parts.

Our parts are not cast. D-Tek parts are cast. Our parts are suplied flat in stock config.

This high totted convex surface will dissappear very fast with just a little usage.

"Disappear" is wrong choice of words. "Reduce" is likely to be a more adequate description.

[WeStSiDePLaYa]

this principle applys to ANY CPY with a IHS. and does NOT have to be tweaked to a specific ihs.

it is easy enough to view the principle, in the easy way i shown.

also, the thinner the base on a block, the more it will benefit from this.

gabe, seems to understand, but he appears to work for a company. (swiftech? )


but if you understand this theory, you will see how it applys to pretty much every water block. all you need is a cd and a glass to see for yourself.