In fact 1GHz benchmarkable is not a problem! That's more than a 50% overclock from the default 650MHz! However, the memory on the card is not as overclockable. I've had trouble trying to hit 1GHz on the memory, the memory would get unstable all of a sudden and crash the system, even when there no visual artifacts on screen. First thing I've tried was to increase Memory voltage which did not help at all. Next I tried soldering some SMD capacitors to see if it helps but that didn't work as well. Next, I decided to try cooling the memory better. RAMsinks are usually used for this purpose, but I was looking for something more cooling... a lot more cooling. I could throw some Dryice on the memory but that would entail the hassle of ordering dry ice everytime I want to do some benchmarking. So I decided to try out something extreme, using refrigeration phase change on the graphics card memory!

I decided to go with a simple design as it is not known whether extreme cooling the Graphics card memory will help. Like that I would not waste a load of time and effort. So, I came out with this:

Don't get it? Well the next few pictures should give you a clearer picture. Basically, it's using a coil of copper pipes to transfer the cold to a copper plate which will then cool the memory. As seen above, I cut out a template for the copper plate from a piece of paper.

On a thin piece of copper plate, I cut out the plate with a pair of shearers.

As seen above, I also drilled out the mounting holes to mount the chiller onto the card.

Next, I brazed on a thin roll of 3/16" copper pipes onto the plate. My first job ended up with me overheating the pipes and breaking part of it.

So I did another one.

After brazing:

You see that the capillary tube runs into one end of the pipe and the refrigreant flow exits through another end. For a more indepth on how refrigeration phase change works and is built you can check out some guides here.

Next, I test-fitted it onto a card.

Perfect!

I wanted to have a piece of arylic to clamp down on the plate so I cut one out:

I also insulated the RAM chiller with lots of cork tape and foam tape:

Next it was time to fit it onto a phase change system! I made a pair of these for Dual VGA cooling and connected them to a dormant chipset cooling system which I have not been using for a long time. It was a dual capillary tube feeder system.

I charged up the unit and off it went! I used refrigerant R22 and rather big capillary tubes to try to aim for -20C on the copper coils.

Frosting action!

In the end I reached -15C on the copper coils and around 10C on the whole plate. Good enough for my use!

The whole setup after insulation:

The wagon of black snakes!

It was time to put the chiller onto the card. As I wanted to space insulation, I decided to use small aluminum plates in between the chiller and the memory:

The aluminum plates are stickied on using thermal tape.

Next, the card was insulated nice and proper:

The RAM chiller was then mounted onto the card:

Further insulation was done and the core evaporator was mounted:

Black snake action!

Did it work? Yes, the memory got chilly! Did it help with overclocking the memory, sadly, not at all for this card, the cooling on the memory itself does not help, cooling on the core helps to overclock the memory better instead.

But if it does help it would be nice to see how phase change on the GPU memory would increase the overclocks wouldn't it?