Good lord! That is one helluva pump! Nice, Its hard to imagine that in a computer. Nice jets!

HEHE,

No wonder you can get AMD mobiles to 2850mhz. !

Looking forward to seeing it done and a full comparison vs Cascade or others. I hope you'll also send pH a production version Cathar.

Great work man!

It certainly is awesome. :drool:

Maybe I can use this in my Mom's garden too :p

2850? I guess I didn't post anything here about it. Time to correct:

http://www.employees.org/~slf/images/3007.gif

Room temperature water-cooling. Not benchmark stable though.

Pump is strong no doubt. Calculated jet nozzle velocity there is up around 10.0-10.5m/s based on the flow rate being seen in that test. Peak possible velocity with the MD-30RZ is around the 12.5m/s mark, but only through restricting volumetric flow rates to a mere trickle (~1LPM) with the restriction required to achieve that.

Still, even with an Eheim 1250 in a full assembled 1/2" system the Storm block's jet velocity is around the 5m/s mark, while with the Iwaki MD-30RZ on a Cascade the jet velocities were around the 6m/s mark.

It's not really possible for me to eke out significantly more velocity without choking flow rates to a near stand-still.

Nice pump and that is one hell of a o/c....

Sweet. What Vcore did you hit that speed at?

Vcore was at 2.15v

do you always run these kind of voltages ?
how long do your cpu live?

In ~5yrs of pumping high voltages through CPU's, I've never lost a CPU yet due to over-volting it.

IMO, 2.00v full-time is the limit of where I'd feel safe with putting through the AthlonXP's.

To Cathar

Hi

I have some problem making my fins smaller than 2 mm wide, crappy quality and taking days to complete (using a NC mill).
Therefore I going to change to a CNC (3 axes) but I need some help.

Therefore I have some questions.

1. What size are your cutting tools (the drill).

2. What rpm?

3. Cutting speed (xy and z)?

4. Cooling?

5. Are there any thing else that you thing I should think about?

6. You are mixing different metals on the blocks, is that wise?

Best Regards
Andreas S

Hi Andreas,

I am not a machinist myself. There are others on this forum who do have industry experience with CNC machining and I would suggest starting a new thread to ask your questions.

Looking forward to your new Storm block. I am contemplating a setup with 2 bonne cores, 2 D4 pumps, a storm on the cpu and an mc50 sandwiched onto a piece of aluminum on a 6800GT video card.

i was going to put the bonne cores in parallel to reduce the load on the pumps which are in series before them. Will this setup be sufficient for the storm block?

I need 2 heater cores since the 6800 puts out SO much heat. It can actually exceed the heat limit of the noryl casing for the pump (140deg F)

Come to think of it maybe I have enough head with 2 pumps to put them in series between the 2 rads. That way the coolant from the vid card gets cooled before it hits the pumps. What do you think?

I was recalling a O/C of yours for a 2850 Stable O/C.......how high is your stable O/C now?

Thanks!

A nice max clock for ambient air cooled rads man! And arn't the mobile chips great! What sort of results can you get with settings for max performance? Settings?? 3D mark scores?

http://www.bwdesign.orcon.net.nz/wb2/Assem1.avi
Uses Xvid codec - I recommend VLC player (for everything actually, not just this avi)

Not what i envisaged (actually, i think this would perform worse, and has been suggested before) but it may get an idea rolling for you. Obviously not the full jets are shown, nor is that the full block base size; actually, the only reason im putting this up is because i said i would, i believe its next to useless.

WRT your jet testing, will less jets focused in the right places make for better cooling due to the velocity? The images you've show still show what appears to be quite a number of jets (more than the cascade, even?)

Would having less result in more water velocity, or have you increased the velocity to as high as you're able to (from what you've said, it seems so)

After looking at a die size, and the area of heat that it creates, the size of a key on a keyboard (14x14ish) should create enough cooling to let the rest be cooled by the outflow, would it not?

/random musings

Hmmm - jetting down onto the tops of the pins where the heat is furthest away from the CPU. Seems to me that what you presented there expends most of its jet energy cooling the wrong spot. Interesting idea I guess.

In my testing I strove for a balance point of pressure drop vs velocity vs cooling contact patch. The Storm block cools almost twice the area of the Cascade block's main contact patch. I did this mostly because it didn't affect jet velocity (in any significant fashion) to do so across a broad range of centrifugal pumping scenarios. i.e. if I had halved the number of jets and the main cooling contact area, overall jet velocity would not have increased by more than 5% for the remaining jets when given the same centrifugal pump (and for certain lower pressure pumps such as Eheims would increase by not more than 1-2%), but block restrictiveness would have doubled. Therefore it made sense to me to simply use that to best advantage and cool as large an area as could be supported without sacrificing jet velocity.

One thing which has not been lost on me in the design of this block is the almost uncanny symbiosis I'm seeing between typical CPU die sizes, and centrifugal pumping characteristics when pumping water. i.e. the design, coupled with the pump, lends itself to being optimised around cooling an area that just so happens to be about perfectly sized for cooling CPU dies and the sizes that they are. While the design could be scaled to much larger areas (like a 50x50mm TEC for example), we would start to lose some of the very nice balance between pump/pressure/patch.

IMO for focused cooling patch designs, if more than 10% of the total heat flux is being dissipated in areas outside of the main cooling patch, then the design is flawed (for that application). For that 10% or less, this amount of heat can be very easily "mopped up" by the most basic of "outflow" effects due to the relatively large metal convectional surface area that it is encountering. Even here though I did spend some effort on ensuring that the outflow was being fairly well used to assist in cooling the metal outside of the main cooling patch.

/offtopic

what PSU do you use?

Etacovda-> Same as Cathar, If you are useing solidworks as I think, then make a termal simulation with Cosmos Works.

Cathar-> Whats the original speed?

PSU is an Antec TrueControl 550W

CPU is an XP-M 2600+. Original speed is 2.0GHz I think. :shrug:

Perhaps it would be a more interesting concept if the jets were directed between the pins rather than on top of them.

Thanks for the explaination on your jet testing Cathar.

Next question (probably the big question)

Do you think you've hit a 'peak'? or is there something else that you have spotted that may be worth working on.

I believe you said the cascade was probably the peak (ie, as far as you can go with block design), but you proved yourself wrong - think it'll happen again?

mwolfman - unfortunately i dont have those plug ins

In terms of impingement, one thing left to try that may offer small benefits. Possible very-small room for improvements with the Storm design. They're on my mind, but the cost of doing such things weighs heavily simply due to the small size of the tools required. I am starting to reach the limits of what's possible on a practically machinable level. So yeah - still some more small gains (<0.5C for scenario below) - but how to make it cheaply?

If (and it is definitely an if) the theory I'm working with holds up, then for a 100W 10x10mm square heat source and a copper base-plate, then I can't see more than ~2.5C left to gain in that scenario practically over a G4, and no more than ~2.0C over a copper G5. Investigating higher thermal conductivity materials starts to become a more practical way to make things better.

Theoretically one could gain yet a further 3C, but such gains come with severe insurmountable physical impracticalities attached (i.e. fantasies of a "perfect" waterblock).

I've said it before, at this stage am pushing up against a theoretical hyperbolic slope to improve performance yet further.

hitting the top of a logarithmic curve

good job on the block, it looks impressive already without any focused performance tests

If you were to make several thousand of the Storm blocks could the price drop to around $50?

No, unless it were being made in China, in which case, definitely.

A quick comparison:

Made for Quality

Raw cost of materials (copper, delrin, aluminium) is about $7 US.

The G4 has about bang on 35 minutes of total CNC run-time, of which exactly 18 minutes presently gets spent on the copper base-plate. These times exclude part change-over, which is about an extra minute per piece.

Copper plates are then ground to quickly remove the dirty and marked finish of the raw extruded copper bar. Copper plates are then machine lapped at an average of 5 minutes per plate.

All up we're looking at a total of around 45 minutes of machine time per block.

Add in quality nickel plated brass barbs, aluminium anodising, stainless steel bolts, springs, and the rest of the mounting kit made from quality sourced components, and that adds about a further $8 US to the price.

So all up, around $15 US for the raw components + 45 minutes of machine time. Machine time in the USA typically costs around $1/min, so you're looking at a base cost of manufacture of $60 US. Assume a block maker wants a measly 10% margin, and so does the end-retailer, plus shipping costs in-between, and you're looking at a final product cost of around $75US even in bulk.

Made for Kwality

Instead, could substitute Delrin for a bodge cheap plastic, and reduce raw material costs to ~$3 US per block

Mass produced cheap-ass steel and plastic pieces for mounting kits => ~$4 US per block

Copper machining done cheaply/inaccurately (no guarantee on critical alignment for certain important elements) could be done in ~15 minutes per plate including a basic ground finish flat to around 30-40 microns (instead of 5 microns).

Could get the plastic plates injection moulded with only follow-up machining for the details that are too fine for a mould to accurately reproduce. Once moulds are payed for, looking at $1/per plate + 5 minutes of machine time.

Cheap Chinese labor of $1US/hr + CNC machine time of ~$20US/hr, total production cost of around $20US/blk. Final sale price could be around $35US/blk.

Performance would vary from block to block due to the shortcuts.

Why not have the bases cast to save on tooling? That's good enough for Starrett. Some may even better trust the machining on cast parts (I do - it looks purposeful).

If your not planing to sell any of the blocks then you can download it for evalation....

Cathar
1 Why the M processor?

2 Laser cuting? (can cut down to 0.1 mm thick parts)

Now how do those numbers add up for the simpler Storm designs? That's what, 1-2C worse? How does the overclock compare? I believe that some of the lower budget enthusiasts may be interested (if the difference is a fair amount. Saving $5 for a fair rise in temperature wouldn't be popular.)

Think it's cheap to do precision parts in a cast? Starrett seems to use fairly straight and large pieces and in HUGE numbers.

Its funny how we all have different areas of expertise, eh?

A Mobile processor is cream of the crop, designed to run on lower voltages (and therefore lower heat output for a given clock freq.) than regular processor.

A 2600M+ is the best processor you can get for max overclock.

My 1700+ was the best you could do for % overclock, i can do about 190% of stock.

Yes. Even one-offs like plaques and trophies will be first cast, then machined, etched, or otherwise detailed. The body of your wristwatch and the blade of your best kitchen knife were probably cast, to begin with. You don't see many commercially viable products made like waterblocks are for a reason.

Maybe your next project goal could be a price performance leader. I think that currently goes to the MCW6000 series from Swiftech so you'de have to compete with only $40 dollars available. It could be an interesting departure from your current and previous designs which focused on maximum performance at relatively high costs.

<-- Frugal :)