Fx 5900 U full w/c (pushing man & machine to their limits)

[BladeRunner]

...........Well at least pushing me with This Machine

With this build I wanted to have one inflow & outflow overall, but the block to cool GPU & Ram on both sides of the card. I Also want a single slot solution, being as compact and light as possible while still being all copper. To do this a one piece GPU & Ram frontside block was required, and this is very difficult to do, (at least with my mill), because the height difference between GPU & Ram needs extreme accuracy when milling the block base if it's to contact both effectively.

Anyway so far so good, the front block nearly finished, working on the backside ram block now that will contain the tube fittings, both blocks will bolt together sandwiching the card to become one similar to the Radeon 9800 Pro ram block I made a while ago.

some images;-

[1398342003]

Every time I think you've done the best, you out do yourself. It's very nice.

Is the hole in the middle for weight considerations, or looks? Does it work well?

[#Rotor]

"it's very nice".... WHAT? that is not just "very nice"..... THAT is awesome stuff
that is chocolate ice-cream on a hot summers night. [Pun intended]

[bigben2k]

Excellent!

Looking forward to a final, and details, of course.

[BO(V)BZ]

Can you show us a picture of the internal layout of the waterblocks too? Even if it's just an MSPaint drawing on top of a picture, I'd still like that =]

[1398342003]

Originally posted by #Rotor
"it's very nice".... WHAT? that is not just "very nice"..... THAT is awesome stuff
that is chocolate ice-cream on a hot summers night. [Pun intended]

I meant "VERY" and there's a after. Here's an excuse, I pressed CapsLock AND shift to make the letters bigger.

I'll try again.

That's the coolest, most amazing piece of WCing equiptment my eyes have ever laid themselves upon. I am awed by your metalwork and skill with a mill, great lord BladeRunner.

One other question, does the block work like the 9800 cooler you made? eg. two connections?

[pippin88]

Very nice.

A few questions: Why are you using so few of the available mounting holes? What's the internal layout like?
Is the block put wrongly on the card? It looks a few mm to one side of where it looks like it should be, for both mounting hole coverage and GPU coverage. Also if I'm right about it being misaligned, then more of the mounting holes line up.

[BladeRunner]

Yes the centre cut out was mainly for weight considerations and I thought it would look different too. Can't test it out until the backside block is done, so plenty of opportunity to mess it up yet! The way it will work when finished is very similar to the 9800 pro ram cooler.

here are a few construction images.

[BladeRunner]

Originally posted by pippin88
Very nice.

A few questions: Why are you using so few of the available mounting holes? What's the internal layout like?
Is the block put wrongly on the card? It looks a few mm to one side of where it looks like it should be, for both mounting hole coverage and GPU coverage. Also if I'm right about it being misaligned, then more of the mounting holes line up.

Cross posting as I was typing ,the other reply when you posted.

well spotted! To answer your questions, the intention was to use the reference heatsink holes only so that it should fit a later card that may not have all the holes, (at least in theory), BUT I messed up on the drawing/milling, and so one of the holes it uses is different. I didn't think it needed all of them for good retention anyway. The 5900 GPU is like a P4, having a heat spreader plate on top the core (being similar to the Radeon 9xxx if it were removed). The top of this GPU is slightly convex and this actually helps in a way ensure good centre contact, (at least is better than the concave Geforce 3/4's).

Due to how difficult it is to 100% accurately make all the holes in the same place as the card, and to allow adjustment of the blocks for when they connect together I used M2 screws. This allows a small amount of movement and is why it looks skewed in the pics as I just quickly fixed in place for the pics with a few screws. It sounds easy to just make six holes in the block line up in the same place as the card ones, but to get all of then in exactly the same place is not as easy as it sounds, (at least when not having access to production equipment), so smaller screws is my solution to allow some room for error.

Here's a few pics of it lined up better, the fact you can still see the edge of the GPU to the left is again another slight milling / calculation error on my part, but as this part of the GPU is not in contact with the block base, it doesn't matter other than it looks untidy. Where the GPU rises up to the contact surface is about 5mm further in anyway.

[pippin88]

I completely understand the problems with getting it all to line up. While my milling machine isn't as new or nice as yours it's still bloody hard on any manual equipment to get lots of various holes perfect.

How deep are the channels, and how thick is the copper?

I'm looking at making a block for my 9700 that goes all over.

[Skeptic]

I have a question I hope you can answer.

I have a Gainward GeForceFX 5900 Ultra 256mb Golden Sample card. It takes up 2 slots on the mobo, 1 agp, 1 pci even though it doesn't conect to the pci.

My questions - How the hell do you remove the heatsinks and fan combo pice, and is it easy to do? Sorry for the question, but I don't have an extra 500 bucks to shell out for this card again.

I am also curious to know if I could use just regular ramsinks combined with a Maze4 GPU block on the gpu and get good overclocking abilities that way.

Any help, instruction would VERY MUCH appreciated.

[pHaestus]

Simply amazing work BladeRunner. Not surprised to see that you've raised the bar again but always impressed by the end product. You might convince me to buy a 5900 over an ATI 9800 if I knew I could pick one of those blocks up to put on it

[BladeRunner]

pippin88

The holes are not really a problem if you use an M2 screw. This has other drawbacks of course as tapping an M2 thread in copper is asking for trouble. It is extremely easy to snap a tap.

The hardest thing overall though from my point of view is holding on to the piece whilst milling it. The centre section I removed at the final stages helped here. The other problem is levelling the mill table so surface milling across the piece is uniform. The vice was too small and useless for this anyhow. What I did in the end was securely bolt a large thick plate of alu to the mill, set it up with the surfacer and surface the alu plate. I now had a milling surface that was set to the milling head, so as long as the head bolts were not loosened it should remain this way, and drilled holes in the alu and tapped them so a copper plate could be bolted to it securely.

I started with an 8mm thick piece of copper plate, marked the design on it and the hole positions, then milled all the channelling, (that varies in depth slightly from GPU to ram area). I then milled part way through the plate, (about 5mm), around the outline, remounting the plate at 45 degrees where required. when this was done it was turned over and the back side skim surfaced until the plate was about 6mm thick. The Gpu area was then skim surfaced to 1.6mm lower. the shaped part was then cut out with a hacksaw most of the areas holding it to the plate only being 1mm thick.

It was still far from easy to do, and the channel milling / levelling would hopefully be a fairly simple operation on production CNC class equipment. I'm also investigating water-jet cutting as an effective way to cut out odd block shapes.

I just finished the O-ring holes & M2 retention bolt holes so this front part is basically done other than final finishing and lacquering. That will be done after it's used for lining up etc in the making of the backside block.






Skeptic

If you mess around with any hardware, you should accept the risks of possibly killing it.

The sinks are pretty easy to remove, (assuming your card is a reference example like ALL the initial release versions).

Accepting it takes up a PCI slot the reference cooler is actually quite a decent cooler. The GPU and ram sinks on the front side are separate items, and the ram sinks at least well attached. The GPU uses the normal plastic pin retention that just require the barb part to be carefully squeezed. the cooler, (or at least mine), was a little stuck on by the chewing gum Tim but came away with a gentle pry. The Ram sinks are removed simply by undoing the screws. Note that the wet tissue / wax Tim on these is easy to damage and the sinks wont fit back on properly without the Tim's in place, (the backside ram cooler makes contact with other PCB surface mounts before touching the ram chips with no Tims in place.

Depending on the GPU waterblock design you may be able to use the stock ram sinks with a waterblock, (Geforce 4 Ti4600 hole spec), as they are separate units. They may however hit each other so I wouldn't guarantee it. A simple Custom GPU block could be made so the standard ram sinks could be used however. As to how well it works would depend on so many specific factors, (your card, ambient, case airflow etc etc). I know when I tested my card in a fanless but cool environment, the reference cooling did a very good job the card sinks being only warm in 2D and warm / hot in 3D. This compares well to Radeon 9700/9800 pro both of which would not run 3D apps without locking up or artifacts in similar conditions due to finger burningly hot ram chips.

To see the card with sinks removed have a look at this review HERE

[Blackeagle]

VERY IMPRESSIVE Bladerunner!

I really like the cut out area for weight savings. And the channel to cool all the RAM chips is about as perfect as it can be.

I do have a question regarding the GPU cooling finned area.

Why allow part of the water to bypass around the finned section?

[Pritorian]

[Skeptic]

Thanks for the reply, and link. I think I am ready to give this a shot then. As long as I can have my block independant from the ramsinks, I should be ok.

[Neo2Neo]

Excellent work Bladerunner,
Ive been checking out your website Zerofanzone and i think
all your work is good but the link for the ramsink cooler for the 9800 pro is out
of action, any chance of a few pics from that?

[BladeRunner]

Thanks for the positive comments and feedback all


pHaestus

I'm working towards making full card coolers in a limited, (small scale), production in the future, but my current methods are far to labour intensive & time consuming. Each cooler I make teaches me more about various important aspects, so it is very useful to me anyway. I pretty much think soldering is out, so a channel with o-ring is the next best thing. This would mean with a two part block (lid & base), could have cnc channel milling in the base part, with the cutting out done to datum points by water-jet cutting.



Blackeagle

The GPU finned part is not really how I intended it to be but I snapped the second of my two 2mm milling bits and didn't have any more, so decided to just finish it like that as I wanted to get on with it. The plan was to mill similar slots at 90 degrees to the others turning the slots into diamond shaped pins. I really don't think it will make a difference as long as there is flow of cool coolant, and it will favour the least resistant path which is through the more central slots anyway, not the right angled edges. 5900 U has GPU temp reporting in the drivers so I'll be able to see how well the GPU is cooled when it is finished.



Skeptic

The card having plenty of holes around stuff helps and I'm sure some commercial GF4 block will fit while still allowing use of the standard ram sinks. I can't confirm this but I heard Danger Den use the four holes close around the GPU for their water-block, but these are different spacing on 5900 U. Danger Den know about this and are making a new block or mounting system. This might be worth checking out because if it used the four holes it might not foul the ram sink.



Neo2Neo

Yeah I'm trying to get the projects written up for my site, but limited spare time usually gets used up making coolers, as that's what I prefer doing The 9700 Pro build was detailed on Deviant PC and they have ceased now. The 9800 Pro I've not yet written up but there is some about it on Pro Forums HERE

[pippin88]

Originally posted by BladeRunner
pippin88
The holes are not really a problem if you use an M2 screw. This has other drawbacks of course as tapping an M2 thread in copper is asking for trouble. It is extremely easy to snap a tap.

I have an 8ba tap and nice little screws that I plan on using for a block like this sometime. Scared I might snap the tap though, its so thin.

[MadDogMe]

What's the flow like?, the square 'lug' for the mounting screw in the middle of the channel looks like it'll be very restrictive. I'd of maybe cut into the sides around it a bit and deepened the channels at it's sides(depending on how much 'baseplate' is there?)...

Luverly pic of the even impressions of heatpaste, pics do sometimes challenge a thousand words ...

[BladeRunner]

Not able to properly test flow out until the backside is done, but I'm sure, (going on past experience) it will be more than sufficient flow wise to cool the chips. The 'lug' area is bypassed both sides and is not the most restrictive part, although in hindsight maybe I should have used the hole above, but when designing it I wanted to use the same holes as the reference sinks use, not all the holes in the card are used by the reference cooler.

The initial inflow outflow area is probably the most restrictive, where is enters and leads down to the main channel & the same on exit. I have just made a testing block out of a Perspex offcut so I can leak test the block. pics should explain this, just bolts to the block in the same way the backside part will and seals against the O-rings. The other side of the Perspex has some 6mm Festo for testing, (backside cooler will have 8mm Festo).







I also like that heatpaste pic too and was happy to see it contacts all the chips after all the intricate set-up work making it entailed. The actual height difference between GPU to Ram chips is 1.62mm, (best I can measure it), so when the block is bolted down firmly it just slightly bends the card in the gpu area, giving good contact pressure on the GPU.

[TerraMex]

Great work, as always.

Just a question , weight. What do you think will be the full weight , with water (im assuming you already have a good ideia) ?

[Brad]

*hands bladerunner an honorary doctorite in watercooling and cool shit*

Only one question - when can we buy?

[BladeRunner]

Originally posted by Brad
*hands bladerunner an honorary doctorite in watercooling and cool shit*

Only one question - when can we buy?

I'm working on it just trying to image what I could do with some production quality CNC machinery

[nicozeg]

You could start by transforming your mill to CNC. That's a good project