GPU water block. How do u want it?

[Bruno Facca]

Ok, you probably read (at least some) of my thread called "The ideal water block" (if you didn't please do ), I said I would make water blocks for the northbrigde and the GPU, some people convinced me that I should do the GPU block first. Now it's the time to do it because I'm making the model to cast the final version of the CPU block so Iwant to make the model for the GPU while I'm at it.

I never really made a GPU block but I learned in these boards that GPUs don't dissipate much heat, so the block has to focus mainly on restricting as little flow as possible. My first idea is to make a block like this, very simple, should restrict very little flow and exchange enough heat to cool a overclocked GF4, well I think so, if anyone knows how many watts a GF4 dissipates when it's heavily overclocked that would be usefull info, I guess this block I'm thinking of can cool up to 50 watts without any problems. I made a sketch in photoshop but my FTP client is in my other HD, wich is not installed now so I'll post it later. It consists of a "box" with 3 or 4mm walls, the inlet and outlet in one of the sides and a division in the middle that forces the water to go around it, so it circulates in the whole block, these channels will already be pretty large, if flow restriction is still a problem the block can be made tall to make the channels even bigger.

This block will be made to be compatible with the Geforce 3 and 4, I have a GF2 myself but I'll have to upgrade once I have to begin testing the block.

I would like to know your opinion on the following:

1 - Measurements of the block so it fits on the place of the original GF3 or 4 HSF.

2 - What you think about the design I talked about, this can wait intil I post the sketch I made

3 - How fast is your Geforce running and how much heat does it dissipates.

Thank you
Bruno Facca

[gmat]

1 - It should fit a GF3 *and* a GF4 through a separate hold-down mechanim. A simple metal plate would do the trick.
2 - I don't know really without any picture. What i can say is, use either 45° barbs or have them stick out quite high above the bottom (like Aquacomputer Twinplex), this because of RAM sinks clearance problems.
3 - It's a GF2 GTS, it produces lots of heat. I don't have any useful figure though, apart from it needs active cooling, and badly.

(side note: if by "at least some" you mean the 6000+ hits on your other thread, then my GF2 produces "at least some" heat )

[redleader]

It's a GF2 GTS, it produces lots of heat. I don't have any useful figure though, apart from it needs active cooling, and badly.

Actually the GTS/MX series was the coolest of the entire line since the TNT days. I belive the typical was 6w. Barely half my old GF DDR. The Geforce 3 and 4 run hotter as well. As I recall even the GF4 MX runs hotter, though I don't have a link.

[redleader]

Thought about it, and I'd do a 90 degree turn. Goes in one side and exits through the adjecent side. No maze or crap, just an open channel a little bigger then 1/2 at the center where it bends and then 1/2 barbs.

Something like this:



Very high flow straight over the core of the chip, so it should have little impact on system flow and still cool nicely.

[MadDogMe]

I'd just like an open flow block like a swiftech for GPU & NB cooling. Barbs on top to start,(so it can be used for GPU &NB)...

[dreamie]

here's a preview of my GPU/NB block.



perhaps u would have more turbulence in ur GPU.. i o/ced my GPU not long ago and the whole card became very hot. the air surround it was very warm...

btw, my GPU is still air cooled!

[Bruno Facca]

This is very useful info, please keep posting. I'm late for something so I'm sorry for not being able to give a more detailed answer.

I couldn't post the sketch I made in photoshop yet because my web space is still having some problems, actually my webmaster is in the process of switching hosting companies so you get the picture.

Thanks
Bruno

[koslov]

I'm thinking microchannels... like Cathar's. Just a mass of copper with some gaping channels ain't special. It'll do the job, but at the expense of decreased performance and increased weight. Microchannel blocks can be lightweight and allow for increased surface area close to the heat source. And they need not wreak havok on flow either.

I don't know if this is doable with your casting process, as microchannels require a high degree of precision.

1/2" barbs are a must for easy compatibility with the majority of other WC components. I have a Gemini GF4 block with 3/8" barbs sitting around because I can't figure out a way to connect it to the cooling loop without killing flow.

Whatever you decide, good luck with it.

[Bruno Facca]

just got a Prolink Geforce 4 MX440 especially for this project, so I have a way to test the GPU block.

Here's my idea:


I have some questions about block position (inclinated to allow bigger size or "aligned" limiting the size to something around 1cm bigger than than the GPU chip itself). I'll post them tonight because I'm late for something now.

@koslov: Thanks for the suggestions, I will make both 1/2 and 3/8 versions, just like I'm doing with the CPU block. I don't like the idea of microchannels very much, IMHO they tend to make the flow laminar, sometimes the boundry layer even takes the whole area and that's not very good for heat exchange, it sould also add some unnecessary flow restriction. Well, even if I wanted to make something like this, as you said it's not doable with the cast process.

Thanks
Bruno Facca

[gmat]

I do not understand fully your question. Whatever solution you choose, make sure the GPU block (*with* stiff tubing) does not block more than 1 PCI slot. Blocking 2 PCI slots is not acceptable...

[Bruno Facca]

In the following pic I'm trying to illustrate how a block that's made to be placed "in a 45 degree" angle related to the video card's edges can be bigger than one that is aligned with the edges. the drawings were made quickly and they're not in scale but they show where the holes on my GF4 MX440 are placed (related to each other and the chip, not the PCB itself because it doesn't really matters..).

The last drawing is my idea for the block (outside part now), it has to be conical to cast nicely and that may be an advantage because the hose fittings being inclined will probably put less strain on the board (because it doesn't has to hold the weight of the hoses, they'll be coming from below).

The "top" (mobile part) of the block is actually the part that goes in contact with the chip, the support will probably be better with the retention part of the block being very close to the board itself, to understand it better look at the outside of my CPU blocks in the other thread and imagine them being placed on the CPU (in out case the GPU) backwards.



I think this post has gotten pretty confusing, my english is not good enough to explain mechanical details like these, I hope you understood what I meant.

Blocking 2 PCI slots is not acceptable...

I agree but wouldn't that also depend on the size of the PCI card you'll be using in there? Ay least with the incluned hose fittings it will.

Thanks
Bruno

[gmat]

Yes, the angled barbs is a good idea.
You may want to go like on "top view 2" because of RAM sinks restrictions on many boards
AGP/PCI slots clearance is standard, check http://www.formfactors.org , ATX spec 2.1 page 12.
With angled barbs you should cover 1st PCI slot but not the 2nd one, so it's ok.

[Bruno Facca]

I'll probably start working on the model for the GPU block tomorrow, I hope to have the external walls done by tomorrow afternoon, I'll do the inside (division) the day after tomorrow, so post all of your ideas before it's too late

@gmat: Thanks for the useful info. Do you know what boards would have that kind of restrictions so I can take a look at them and work something out?

thanks
bruno facca