Date: December 11, 2006
Author(s): Matthew Harris
One of the drawbacks of full graphics card water blocks is that they don’t always fit the cards being used like they should. There are tolerance issues depending on the height of the ram versus the height of the GPU. Swiftech offers a cure to those ills with the MCWRamcool.
When performance is your goal you often go to extremes to wring as much of it from your equipment as you possibly can. Quite often it involves the black art of overclocking and this is where your thermodynamics can make or break you. Face it, if your case won’t cool well you could have the noodliest equipment with the wildest headroom around and it won’t OC worth a damn.
A mediocre CPU will out clock an extremely good core at near ambient temp if the good core is at 50C or above. We’re starting to see ram that benefits from good cooling such as active cooling or water cooling by rewarding us with very good overclocks. I’ve heard a lot of debate over water cooled ram and I have to say I’m sold! I took a box stock 7950 and pushed it to the point that it went into "Emergency 3D Mode" on air and surpassed the clocks that it failed at previously by a very good margin by switching to full water cooling on the card.
I’ll get into greater detail on this later in the review. First let’s hear what Swiftech has to say about the MCWRamcool.
The MCW-Ramcool series are extreme duty waterblocks designed to fit nVidia’s GeForce series VGA memory modules. The design concept is a standalone solution mechanically distinct from the GPU waterblock. For this reason the MCW-Ramcool waterblocks guarantee superior thermal joints with the memory modules compared to stock or to other aftermarket "monolithic" cooling solutions. This can translate into superior performance and overclockability of the memory modules.
In order to provide universal compatibility with existing liquid cooling systems, we have designed the Patent Pending "F" fittings which allow the MCW-Ramcool to operate with 1/2", 3/8" and 1/4" cooling loops. These fittings provide a clean tube routing, with a minimum of pressure drop in the system by linking in parallel the 1/4" MCW-Ramcool ports to the VGA water-block ports.
I won’t go into greater detail because it gives you the general idea but if you’re interested, Swiftech has more details and a flow chart showing pressure drop data.
The box is a simple corrugated paper (commonly known as cardboard, but it’s not…really) with a printed label showing a line illustration of the block. For some reason this is typical of the higher end American water cooling companies out there. The stuff from overseas uses flashier box art but box art doesn’t mean much.
It’s what’s inside the box that matters. Inside there are 3 pair of "F" fittings, clamps for the "F" fittings, a short section of 1/4" tubing, thermal paste and the block and mounting hardware. The "F" fittings are unique, there are no 90’s to slow down flow. Instead the water exits at a 45* angle to the original direction of flow. This creates a marginal decrease in flow velocity but the benefit is twofold, the smaller tubing exiting at that angle gets flow sheared off as it flows over the oblong hole with the sharp pronounced corner that is created as a result and the main tubing going into that bend creates enough back-pressure to help multiply that effect.
On the outlet side we have the pump drawing water into itself and that bend helps to cause the pump to draw water through that smaller tube through a venturii effect as there’s little resistance to the water joining the flow and again, the bend multiplies the effect of the 45* angle of the smaller tube. If you blow on the end of a straw at 90* you wont see much action, the air has hardly any time to enter the straw and create either pressure or suction. Blow across the same straw at a 45* angle to the air flow and you’ll either blow bubbles in your Coke or piss off the better half because you’ll have jetted Coke all over her.
Here’s a look at the ramblock in the flesh. Pretty nicely made in my opinion, the fittings are molded in place but to save space that’s the price you pay. That allows for no clearance issues and doesn’t force the fittings to stick out past the edge of the card by a large margin. If the fittings screwed in, the bolsters you see them mounted to would have to nearly triple in size.
As it stands, they’re over the card in one location and right to the edge of it in the other as you’ll see. Looking at the underside reveals the the pads for the ram, note that like everything else Swiftech sells, they’re not mirror polished, just machine lapped. I’m of the school of thought that mirror polishing is bad in that it impedes heat transfer by not allowing the thermal compound to get pulled up into the fissures in the finish thereby causing a thicker layer of compound. Thicker compound acts as a blanket and becomes a thermal barrier.
The card in question is a shiny new 7950GT. Yes, I know that removing the stock cooling solution voids warranties on most cards but not on at least one brand, EVGA. The card in question is a new 7950GT KO, you’ve got to love a company that tells you point blank in person (I spoke to the rep at a recent nVidia event) that as long as you do not physically damage the card in the progress changing cooling solutions won’t void the warranty. Just hang onto all the stock cooling parts as the warranty requires them to be installed properly for RMA.
Anyway, after removing the HSF and ramsinks you’re greeted with this, a bare card. On that bare card you mount first the Ramcool then the GPU block. You could squeak the Ramcool under a GPU block but I prefer to set everything on my card on a straight down motion. This reduces the chance of knocking anything off by sliding something across it. Why risk killing your new toys (or old toys) when it can be avoided?
The Ramcool is happily pretty much universal as long as the GPU block doesn’t project towards the outer edge of the card by any but the smallest margin. I’m not certain about the low-pro Maze 4 blocks though, could go either way depending upon the tubing used. 3/8" with a 1/2 O. D. tubing, it should work, 1/2" fittings on the other hand could be iffy.
If you look at the Ramcool you see four brass standoffs pressed into the bottom of the block, these stock style screws go through the card and screw into them. The screws have a shoulder on them that tightens to the standoffs using them as the stop for the screws, this prevents two possible problems. First is that you can’t over tighten it. Trust me, this bay is secure as can be but it won’t warp or hurt the card. Secondly, you can’t tighten the screw to the point it either punches through the base and into the top of the card or pulls the standoff out of the base. This is great in that you don’t risk either breaking the card or the block, which would cause a leak.
Using the supplied thermal paste which is Arctic Alumina and doesn’t conduct electricity put as light a layer of thermal pastes as possible on the ram chips. Trying to spread thermal paste on ram chips is a challenge and getting it thin enough to get a good even coverage often results on getting a bit on the card. The Arctic Alumina stops it from becoming a chore requiring cleaning, cussing and re-applying.
As a benefit you’re able to spread it as thin as possible and just cover the chip, since thinner is better. The GPU I used AS3 (hey, I misplaced my AS5 and besides, this needs to be used anyways) but since it’s only one chip I’m willing to concentrate on the coat for the time required. When a good coat of AS3 (or 5) requires 5-10 minutes to get thin enough to make a sheet of paper join weight watchers I don’t feel like committing 8 times that on ram, or at least I didn’t. The next time I will go to that length. I was of the school of thought that with ram water is overkill until I discovered differently, but enough of that, I’m prattling.
After lining up the Ramcool with the card you put the screws through the stock mounting locations and tighten them down. No drama. If it weren’t for the tubing sticking out you’d think this was a stock setup. Yes, the SLI bridge connector is blocked. If you don’t have plans to do SLI then this is no issue and if you do there is a version tailored to the 7900-7950 that won’t cramp your style. For the rest of the cards, the SLI bridge connector is in a different location and won’t pose a problem.
From the top you see just how much room is left around the core. You could practically air cool the core but then again, why? ;-)
I pre fitted the tubing in a loop so that I could just mount the "F" fittings and route the tubing to first one then the other plus it gave me an idea of how much added sideways expansion to expect..
After I got the fittings in place and aimed in the path of least hassle it all came together nicely.
What you couldn’t really see from the top down was that I’d angled the rear fitting up and the front down. This was the easiest way to work with my loop, your results will vary. I’ve got to say, I was surprised how much velocity that the Ramcool has going through it, I figured that with it parallel to the GPU block in the loop that the flow through it would e sub-par. In fact I had expected to see a bubble that took a bit to bleed out at first. Imagine my amazement when I looked at the tiny bubbles flowing through my tubing during the fill & bleed process and saw that they were whizzing through the Ramcool at breakneck speed.
As you can see the block only adds minimal complexity to the loop. The nice thing is that with the angle on the "F" fittings you won’t have to modify the hoses going to the GPU block. The angle allows you to aim right towards an easy departure angle on the existing tubing rather than if hey were straight then it could become problematic in that you might have to add tubing to avoid kinks or undue torque on the fittings.
That’s it for the pics, sorry there was no inside pic but I did take it apart. My cam was unavailable at the time and after the issues I had with the screws (I’ll get to that) I decided to forego trying it again. When I disassembled it I found out that the screws are rather soft and slightly wallowed out, putting it together again was a case of beat the clock against stripping them out, that and the odd shape the "O" ring has to follow made reassemble a study in invented curse words, I’ll just say that I was rather inventive. ;-)
There’s not much to say about the inside, the base is flat but with the total absorbed wattage from the ram it doesn’t really need anything fancy and besides, this helps to keep the profile very slim and lowers restriction, very important to this block. If you’ve seen the inside of the MCW30 N/B cooler you’ll know what the Ramcool is like as far as the base is concerned.
On to the numbers!
I tested the Ramcool combined with the MCW60 GPU block versus the stock air cooling, no I didn’t bother with ramsinks as the stock cooling is pretty healthy. The GPU temp never got above 56*C and the cooler is solid copper with large thin fins all the way over the ram. It wasn’t GPU only with no cooling on the ram.
Note that my ram Mhz is in actual Mhz rather than in double data rate as that’s only effective clock speed.
I tested the stability of OC’s with 3DMark ’06 and by several sessions of gaming with Call of Duty 2. I played a bit of Oblivion as well. This card is simply amazing as I am playing both games at 1600X1200 with max details and no lag. The trouble is that at the same res the stock cooling kept things going long also with no lag. Sucks for a real comparison but fortunately I have the 3DMark ’06 scores to compare, I won’t rely on the CPU score and will instead give the shadermark score.
There’s a Core 2 Duo in the database running 7950GT’s that are over my system in terms of overall score due to the CPU score but were considerably lower in SM scores. As a result there are two SM scores for single 7950GT’s in my CPU clock speed range (2000Mhz to 2900Mhz) that are higher than this setup but one is volt modded to 1.7V and is OC’ed to 704 core and 823 mem. Only one that’s higher looks to be un-modified and it’s a horse race.
Nothing drastic to be honest, the results in games is likely more drastic but as this isn’t a videocard review I’m not going into great depth, the point of the exercise was to achieve a higher stable OC. I tried the driver’s "Suggested OC" of 665 core 750 mem and I discovered to my horror that 3DMark would drop to frame rates in line with my 6800nu when run after a few seconds.
No artifacts but the FPS dropped from an average of 16 on Return to Proxycon on the highest stable OC I was able to run on air to 6-7. This is at the point in the test where the scene moves to the defenders. I backed both the core and mem down to stock and it was fine. After a while I found that any speeds higher than I listed in the table caused the same slowdown. I know it wasn’t over-thermal due to logging with Rivatuner 2.0 RC16.2 which is how I knew that the core was at 56*C.
I don’t know what caused that but I know that moving to water cooling removed that obstacle and I’m fairly convinced that had I gone with ramsinks I’d have seen the same ram clocks but the core would’ve still improved. The slowdown happened on the ram above 735Mhz no matter the GPU clock, it was just freaky. I do know that I haven’t hit the wall on the OC yet, I’m still working on it, I’ve just hit the point where I want to live with the system a while to get used to any quirks it may have.
At nearly $60 MSRP the MCW-Ramcool is an easy to stomach alternative to the full card blocks as an upgrade path if you have a GPU block already that is. If not the cost is a bit high and could be a stumbling block to its success. No it doesn’t cool the PWM circuitry on the uber cards but I think that with good air flow around your cards that it’s not an issue. Most of the heat has been removed from the air in that area so the PWM section of those cards will be getting all cool air.
Overall though I’m very impressed, I never dreamed that watercooling my Vram would make such a difference to the tweakability of my vidcard but I have to say, I’m sorry I didn’t try this sooner. I’m awarding the MCW Ramcool an 9/10 along with our Editors Choice award!
If you have a comment you wish to make on this review, feel free to head on into our forums! There is no need to register in order to reply to such threads.
Copyright © 2005-2020 Techgage Networks Inc. - All Rights Reserved.