Zalman Reserator 1 and GeForce 6800 Ultra

The alternative to direct air cooling

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
charliek
Posts: 25
Joined: Thu Aug 05, 2004 5:44 am
Location: 66500 France
Contact:

Zalman Reserator 1 and GeForce 6800 Ultra

Post by charliek » Thu Sep 23, 2004 10:57 am

Zalman Reserator 1 and GeForce 6800 Ultra – Watercooling the VGA

Temperature Diodes

First off, a note about temperatures. The figures I give here are in degrees Celsius from the following sources: Ambient case temperature is the average from two probes built into my Antec P160 case, one by the air intake and hard drive bay at the bottom, and the other by the top of the PSU, right at the top of the case. CPU and Motherboard Temperatures are given by my Athlon 64 3400+ and ASUS K8V SE Deluxe motherboard diodes, read by Asus Probe. The VGA temperatures are from the Gainward GeForce 6800 Ultra 2600 Golden Sample onboard PCB and GPU diodes, read by the nVidia 61.77 driver.

Diodes are inaccurate. If you and I had exactly the same equipment built in exactly the same way, the chances are the diodes would give different readings. It’s even quite possible that the amount by which a diode is inaccurate, changes as the temperature changes, so for example if I give readings of 30 degrees idle and 40 under load, the first might be out by +3 and the second by -1 on the same diode. So. Take temperatures with a pinch of salt. They are illustrative at best.

Before

I have been running my Zalman Reserator 1 on my Athlon 64 3400+ for a while now. It is silent and my cpu, which is at stock speed, has stayed cool. (With an ambient case temperature of 36, the motherboard/case temperatures have been between 40/38 idle and 40/44 (yes, at idle the CPU is cooler than the motherboard).

The VGA card, however has been warm to put it mildly. The Gainward GeForce 6800 Ultra 2600 Golden Sample is designed to run overclocked at 430MHz GPU and 1.2GHz RAM. It comes with a large copper heatsink on the GPU; an array of aluminium heatsinks around the six RAM chips, a small black ceramic heatsink on the MOSFETs, and an aluminium shroud over the whole thing with two fans blowing air through the heatsinks. The fans have variable speeds, depending on the heat of the card. They have red LEDs in them, too.

When it arrived, it turned out that the card’s shroud was too large for my motherboard, and I had to cut away the aluminium to accommodate the SATA ports. While I was butchering it, I figured I may as well check the heatsinks. I removed the GPU and RAM heatsinks and re-seated them using Arctic Silver 5. I also removed some thin plastic spacer/washers from the MOSFET heatsink as its thermal pad wasn’t in good contact with the little fellahs. All in all, this saved me a degree in idle and about three degrees in load temps.

The twin fans made a lot of noise, and didn’t strike me as being wholly efficient – the GPU temp readings were quite scary – high seventies under load. Additionally all of that heat was being dumped into my case.

So I downclocked the GeForce to stock temperatures 400Mhw GPU and 1.1Ghz RAM, and added a 120mm Pabst fan to my case’s air intake.

This brings us to Scenario 1 in the temp table below: Watercooled CPU, Aircooled GPU throttled back, five fans (two 60mm VGA, Two 120mm Case and one 80mm PSU). A fair bit of noise. The PSU, by the way is a Tagan 480w and it makes negligible amounts of noise.

So, now to fix this.

After

I got bored of waiting for Zalman to release a GeForce 6800 version of their ZM-GWB1 VGA waterblock for Reserator 1. They are working on one, or rather a mounting kit for the existing one. I decided to order the existing one (an anodised aluminium unit that is ideal for the rest of the anodised aluminium Reserator setup), and work around mounting it myself – it’s been done before.

While had my credit card out, I ordered a Zalman ZM-NB47J passive Northbridge cooler to replace the small ASUS passive heatsink – just in case I fancied doing a spot of overclocking when this is all over, and cooling the northbridge would help.

Basically it was easy to do with this card. The existing backplate for the Gainward’s heatsink had four bolts already attached to it, which solved the issue of the ZM-GWB1 mounting bolts being too fat for the 6800’s holes. I kept the existing backplate and used the Gainward’s screws and springs to hold the VGA waterblock down, with Zalman’s little rubber o-ring washers to keep it all tight. I pilfered the two mounting arms off the small ATI waterblock that comes in the kit, so I had four mounting arms for four backplate posts. All in all a nice snug fit, and Zalman’s mounting arrangements make it easy to move the waterblock around for the best fit over the GPU die.

I replaced the aluminium RAM heatsink with the stick-on ones that come in the Zalman kit. The Gainward RAM heatsink looks very good, is large, and can be seated with Arctic Silver, but it is too tight for comfort against the water block inlet pipe, and we’re not at home to Mr Piercing. Oh no.

Having mounted the waterblock, it was time to break into the cooling loop to fit it. I did this by putting the case on a desk and the Reserator on the floor, and carefully opening the higher of the two case inlet barbs, so that the residual water in the system flowed back to the Reserator (which is why they say not to fill it to the top).

I detached the CPU waterblock and raised it out of the case, making it the highest point, and then released the longer of the two pipes going to it (I’d left it longer for just this reason) and attached it to the VGA waterblock instead using the supplied pinch-collar.

I didn’t spill a drop of coolant, but if you want to try this at home, folks, you’re on your own.

I had already bought some decent 10mm Inner Diameter (10 ID) tube, and used a length of this to make the loop back up between the VGA and CPU waterblocks. This was good quality clear PVC 10ID/12OD tubing and, while kink resistant, was much more inflexible than the lovely Zalman silicone tubing.

I ran the whole thing, motherboard out and VGA card upside down, for a day to check for leaks and , once satisfied, put it all back together adding the northbridge heatsink and redoing the Arctic Silver 5 on the CPU.

Image

Pic – mostly reassembled, after testing for leaks

Having tested the rig, the temperatures were good, and the addition of the GPU to the loop seemed to add only a couple of degrees to the CPU cooling! Woo! The Reserator strikes me as warmer than before, mind.

Flushed with pride, I restored the Gainward GeForce 6800 Ultra 2600 Golden Sample to its default overclock of 430MHz GPU and 1.2GHz RAM. The temps stayed good, at CPU 51 degrees under load, and GPU cooler under overclocked load than it was aircooled at stock frequencies under load.

This brings us to Scenario 1 in the temp table below: Watercooled CPU, Watercooled GPU overclocked, two fans (One 120mm Case and one 80mm PSU). No detectable noise at all.

This is a fast, fairly hot rig, and I can’t hear it over the background noise in my room (which isn’t much). The rig sits under my desk on a wooden plinth, about six cm under my left elbow.

All in all, I’m most chuffed.

Code: Select all

Temperature Table

For each scenario, Case, Motherboard, CPU, Video PCB, and GPU

Scenario	Case	MB	CPU	vPCB	GPU

1) idle		36	40	38		50	61
1) load		36	41	44		53	72

2) idle		36	40	38		50	55
2) load		36	41	46		52	69

tinyrodent
Posts: 3
Joined: Thu Sep 30, 2004 4:24 am
Location: Canada

Similar system

Post by tinyrodent » Mon Oct 04, 2004 6:51 pm

Hi Charlie,

I have a similar system to yours: Antec P160 case, and Reserator cooling CPU and GPU. But there are some differences and I'm wondering if my system was put together properly. I bought it preassembled, all I had to do was hook up the tubes and add water.

My CPU and GPU are both slower than yours: A64 3000+ on MSI K8N NEO Platinum, with MSI 5700 video card. However I wanted the quietest possible system so I chose not to go nuts with the power. The PSU is a Silenx fanless model, and I can comfortably run the system with no fans at all.

However even though my case temperatures seem similar to yours, the CPU and GPU readings are way higher: idle around 60C and load around 70C. Diode differences, crappy install of the waterblocks, or something else?

Brett

charliek
Posts: 25
Joined: Thu Aug 05, 2004 5:44 am
Location: 66500 France
Contact:

Re: Similar system

Post by charliek » Tue Oct 05, 2004 4:49 am

tinyrodent wrote:Hi Charlie,

I have a similar system to yours: ... My CPU and GPU are both slower than yours: A64 3000+ on MSI K8N NEO Platinum, with MSI 5700 video card.... However even though my case temperatures seem similar to yours, the CPU and GPU readings are way higher: idle around 60C and load around 70C. Diode differences, crappy install of the waterblocks, or something else?
Hiya Brett,

The temps you quote seem a tad high to me, but within spec (although I don't know the 5700 too well). As I mentioned, there is really very little point in comparing diode readings with any hopes of accuracy, but this is quite a difference.

If I were you, I would probably be looking to improve the seating of the waterblocks, particularly with reference to the thermal paste (arctic silver have some good instructions on their site). Make sure that the waterblocks are mounted flat against the cpu/gpu, and that you haven't used too much goo. It is tempting to err on the side of too much, and that will increase your temps.

My method, with arctic silver 5, is to put a dab on the waterblock/heatsink and then rub it off again with a lint-free cloth (sometimes you can see a slight discolouring of the surface where microscopic traces are left in the 'valleys' on of the metal - this is fine). Then I apply a blob about the size of 1.5mm of standard wooden matchstick (or 1/2 a grain of uncooked long-grain rice) to the cpu/gpu die. this I smear over its surface with the edge of a credit card until it is practically transparent. Then I meet the two surfaces and give them a gentle wiggle together so they 'mate' well. Then I seperate them again - there should be a kind of resistance to pulling them apart like you'd expect from two panes of wet glass, and the surface of the paste should have rippled where there was good contact. Assuming this rippling covers at least almost all of the chip, I put them back together and bolt the waterblock/heatsink in place. If not, I add the tiniest quantity of extra goop to the bits that clearly didn't meet and re-try for a good fit.

There are more opinions about this than there are about how to make good toast, so read up!

Good luck

Post Reply