Tweakmonster BGA RAMsinks ReviewBosco - February 15, 2003
: PHAM Computer
Price: $25 USD For 8
I'm not sure whether this is a performance review or a modding review. What the heck, let's look at it from both perspectives. We recently received a set of 8 BGA (video memory chip) Tweakmonster RAMsinks from Jerry at PHAM Computer. I'm going to install them on my Gigabyte Maya II 9700Pro and show you how they look and what, if any, performance gains they give me.
The real thing
They are sold in sets of 8, they arrived in 2 blister packs of 4 each. The blister packs prevent the sinks from bumping each other and possibly chipping off the chrome finish. PHAM Computer's shipping department further did a nice job of packing them in an oversized box with bubble packaging and everything arrived in perfect condition.
They are constructed of a copper alloy (alloyed with what, I'm not sure..) and are quite heavy relative to their size - they measure 12mm X 15mm X 9mm high (.6" X .5" X .375" for the metrically challenged). I don't have a postal scale so I'm not sure the exact weight. The size is small enough to not interfere with most GPU cooling solutions - for example, DangerDen's R9700/Ti4600 waterblock will still work with these RAMsinks installed on the memory chips.
(Picture courtesy PHAM Computer)
The RAMsinks are very good looking and SHINY. They are finished by chrome plating - it's a decently heavy plating that even if they were knocking one another in the packaging, I highly doubt they would chip. Overall the finish of the chroming is very smooth on the upper/visible surfaces of the sinks. However, on the bottom there is a residual ridge from the plating process - this is really the only negative I encountered with them and as you shall see in a moment, 20-30 minutes of light work will easily alleviate this concern.
Mostly, we are in the practice of reviewing articles "as is", without modifications. In this case, because the adhesive I am going to use is permanent in nature, I'm going to go ahead and lap the bottoms of the sinks, to ensure that very "square" contact is made between the RAMsinks and my video card's RAM chips. It's not a big, difficult or particularly time consuming job. I grabbed a sheet of 1000 grit wet/dry 3M sandpaper and a small bowl of water and set to work. It was recommended to me by a machinist friend to work in a "figure 8" pattern any time you are working for a flat surface and so, 3 or 4 minutes on each sink, with very moderate pressure, and a nice flat surface was easily achieved. It does remove the plating from the bottom, as far as I'm concerned this is actually a positive as the copper alloy will now be in direct contact with the adhesive - I have never read about chrome being a great conductor of heat (rimshot).
OK, unfortunately my camera just doesn't have the resolution to give a good close up. What I'm trying to show in the shot above is the difference, 4 are lapped and the closest 4 are not. Lapping does remove the chrome off the bottom, there is also a very thin layer of what appears to be pure copper just under the chrome and it ends up being partially removed also in the process of getting an extremely flat finish.
To ensure that I had the right process in mind, I paid a visit to the @installation guide on Tweakmonster's website (PS: for the non-squeamish, check out the pics of what NOT to do with a knife while modding, linked from their front page). They recommend swabbing down both the RAMsinks and the surface they are to be applied to with alcohol. Check. They do make a statement about lapping the sinks with 600 grit sandpaper if so desired. Well, I used 1000 but in any case, check. And lastly, they recommend using a dielectric (NON-conductive) thermal adhesive. At this point I'm going to deviate from their recommendations.
Silver based Epoxy
I spent the equivalent of almost a full day phoning and visiting various computer, electronics, hardware, and other stores in Edmonton trying to find some AS3 adhesive or equivalent in a dielectric version. Other than a few quizzical stares, what I got was "Nope, sorry, too specialized/expensive/wierd for us". I ended up settling on Circuitworks Conductive Epoxy - exactly what I didn't want, and I certainly don't recommend you try it. On the plus side, it's a silver based epoxy (looks just like Arctic Silver 3 when mixed) and so should have similar thermal properties, besides having the ability to screw up my $500+ video card if I slip and bridge some electronics. And so, with shaking sweaty hands, I set to work. The first thing I did was cut up some narrow strips of masking tape and mask off the areas immediately surrounding all 4 sides of each BGA chip.
As many of you will know, epoxys are usually (always?) a 2-part solution. One of the parts contains a catalyst that sets off a chemical reaction and "cures" the adhesive. The package says to mix the 2 parts at least 2 minutes, and apply within 5 minutes. Not much of a window of opportunity, and I can now tell you from experience, after that 5 minutes are up it's time to make a new batch. When freshly mixed, it has a very similar texture to Arctic Silver 3, slightly thicker than generic thermal grease. After 5 minutes, it's like your little sister's play-doh that's been left out of its container overnight - pretty tough to get a thin, smooth layer. I should also mention, the stuff I used is TOXIC AND CORROSIVE TO SKIN, according to the package. These, particularly the last, I did not feel compelled to test and report to you on.
Moving right along, apply the epoxy to the RAMsinks, NOT directly to the video card chips. I used 7 of the 8 sinks, 4 on each of the backside BGA chips (Gigabyte is kind enough to provide a cooling solution for the front chips but not the back?!?!) and 3 on top of what is itself already a heatsink, tho that sucker gets almost too hot to touch and I figure some more cooling certainly won't hurt and may help. I worked quickly but carefully, applying the adhesive to the RAMsinks over a paper towel and only when satisfied with a smooth layer did I move over to the 9700Pro and position them. Fortunately I got a very clean job out of it, didn't even get any epoxy on the masking tape.
The instructions are to let the epoxy set for 4 hours at 24C (75F) or "for faster curing times, maximum conductivity and adhesion, heat the bond to between 66-121C (150-250F) for 10 minutes and allow to cool." I grabbed my wife's 1500W hair dryer (hey, I'm not sticking the card in the oven, YOU try it if you think you're so smart) and heated everything up at max setting for 15 minutes, then let set for 5 hours anyways. At this point, I can a) pick up my video card by a RAMsink no problem, and b) brace the PCB and twist very firmly on the RAMsink with no evidence of movement. Time for the fun part.
Whoops, fried the video card.
Bosco just had a heart attack when he read that hehehe. Everything's fine. This is going to be some pretty simple testing, as I have well established the overclocking limits of the Gigabyte 9700Pro by now. I have tried numerous tweakers and overclocking utilities but have found my best *clock results* come using the Gigabyte utility that came with the card, known as V-Tuner. As we are focusing on the memory here, I won't discuss the GPU core overclocking - other than to say that the core speed was kept constant throughout testing.
Location, location, location
The V-Tuner utility, as with most videocard overclocking utilities, allows you to set a desired memory speed number then "apply" it. This then generates a number that is almost always slightly different than what you set it at. For example, setting the memory speed to 321 then hitting "GO" results in a memory speed of 325. I can 100% always run 3D Mark SE 2001 build 330 at 325 memory speed. Depending on the weather (yes, you read that right - my computer sits with the side open, next to an open window - temperatures over the past 2 months have ranged from -30C to +5C) I can get the occasional run in at 332 MHz, which is the next jump up in V-Tuner.
Veritech, can I borrow your camera? :)
After installing the RAMsinks, I have seen a small increase. 332 I can now call a stable speed - it's at 0C outside as testing has been performed, which is quite warm. I made 2 artifact free passes at 339, which is yes, the next jump up in V-Tuner, and is a speed I had not been able to get even once before. I also made quite a few unsuccessful attempts at that speed and was not able to get a clear run at 345. 332 MHz compared to 325 MHz makes what amounts to a statistically unimportant improvement in my 3DMark scores and so I won't bother charting it - but it WAS an improvement. And maybe I can get a run in at 339 when everything's pumped to the max and finally crack 17k 3D Marks (sitting at 16,932 for now).
I like them. They are going to look great and fit in with the theme of a major mod project that is nearing completion (keep watching the site for lots of details in the coming weeks). However small, I did get an improvement in the memory speed of the video card they were installed on. My only gripe is with the flatness of the bottom, and like I said before, 20-30 minutes and that was taken care of. Are they for everyone? Probably not. Are you looking to get every last possible memory bump, or to put a nice finishing touch mod in your box? I can definitely and without reservation recommend that you pick yourself up a set.