Corsair takes memory cooling to a new level when it pairs up its best DDR3 memory with a Peltier element and water cooling. We take a closer look at what this extreme solution can do.
By the end of 2008 Intel had its new architecture, Nehalem, out on the market creating new challenges for the memory manufacturers. The new architecture brings triple channels, unlike the older architectures that only have two channels, which means that now you need three modules to get the full bandwidth. Not only do you want to match three modules, but you also have to recognize the properties of the integrated memory controller.
As for the memories, Corsair is of course updated with the latest specifications. With the launch of Core i7 it introduced a number of triple channel memory kits in two different series, XMS3 Classic and Dominator. The Dominator series brings bigger heatspreaders and chips made for higher frequencies and there is also an optional fan for some of them. We’re not testing any of these kits today, but a very exclusive memory kits that ships with Peltier cooling.
Let us present Corsair Dominator 1866C8 TEC.
Corsair is definitely beating the big drums with this kit. Not only does it promise aggressive specifications but also include a water-cooled Peltier block that will keep temperatures at bay. The water block is cooled by Corsair’s own Nautilus cooling system that is normally delivered with a water block for processors. Off the top of my head one might wonder why cooling the memories is even necessary. Normally you don’t even consider the heat coming from the memories. Some have more exclusive heatspreaders than others, some even come with a fan setup, but overall they tend to just work. But when you start packing high-density modules, raise the voltage and start overclocking the effect becomes noticeable. Like with most overclocking lower temperatures are preferred, thus Corsair’s Peltier cooling solution.
A Peltier element is an electronic component that creates a difference in temperature when connecting a source of voltage to it. The size of the component is in this case 40x40mm and is fed with 12V through a regular 4-pin Molex connector. Using the voltage the heat is transferred from one side to the other and the cooler you manage to keep the hot side, the colder the cool side gets. The cold side is connected to the heatsinks of the memory and the hot side to the water block, which cools it. When it comes to the Peltier Corsair hasn’t held back. It has a specified effect of about 120W, which should cool the modules plenty. Our sample was delivered without a temperature controller that normally makes sure the modules don’t get too cold and cause condensation. We assume that this will regulate the Peltier element on a convenient manner.
The water block has 3/8″ nozzles, which is the same as the Nautilus kit. We would have liked to see a threaded connection so that you could use your own nozzles. The Nautilus system consists of an external box that houses pump, radiator and fan, all to minimize amount of work needed inside the case and such.
Let’s look at the test system.
|Motherboard||ASUS P6T Deluxe|
|Processor||Intel Core i7 Extreme Edition Q965|
|Memory||Corsair Dominator TR3X3G1866C8D (3x1048MB)|
|Graphics card||ASUS 9800GT Matrix|
|Power supply||Corsair HX1000|
|Operating system||Windows XP SP3|
|Drivers||Intel Chipset Driver 126.96.36.1995|
nVidia Forceware 180.48
For testing the memories at various configurations we have chosen to use MemTest86+. MemTest is not run in the operating system and communicates directly with the memory. This makes the test a lot faster and that it can test the memory very precisely. We ran test #5 for at least 5 loops at stable conditions. The memories had reached their equilibrated temperature with the open system. The lowest voltage we used was 1.66V, which is really above the maximal 1.65V recommended by Intel. The reason it was 1.66 was that ASUS P6T Deluxe doesn’t have a setting for 1.65V, just 1.64V and 1.66V.
We start by testing how the memories overclocks at 7-7-7-21.
The 7-7-7-21 timings doesn’t seem to be a good combination with these particular memories. Even if we feed them almost unhealthy 2.0V we barely get above 1800DDR. We can only conclude that there is very little reason to use these timings with this platform.
Next up is 8-8-8-24.
With 8-8-8-24 things start to happen. The promised frequencies are surpassed by about 60MHz DDR. with 1.80V we passed DDR2000 and there is still more to get. Over DDR2000 requires that you turn up the QPI voltages so that the memory controller can keep up and to get to DDR2100 we had to feed as much as 1.65V to get the system stable. Over DDR2100 at CAS8 without errors is more than OK for an early engineering sample.
The memories clocked around 250MHz higher by moving from CAS7 to CAS8, but will we see the same jump when we move to 9-9-9-27?
Yes, well almost! At 1.66V there is a 200MHz DDR difference versus CAS8. After that the trend stagnates and the reason is simply because we didn’t manage to get any higher frequencies at these voltages and temperatures. 1.65V QPI is a lot for a 45nm process and not possible in the long run with just air cooling. It definitely feels like there is more to get from the memories.
So far we haven’t even used the Peltier element and the memories have stayed at a temperature around 40°C. Let’s see what it can add to the equation.
Mounting the Peltier cooling is done by unscrewing the top of the heatspreaders, followed by attaching a metal plate on top of all three memory modules. You then mount the actual Peltier between the metal plate and the water block. Eventually it will like the pictures above.
To investigate to what extent the Peltier has an effect or not we started running MemTest before the Peltier was connected, at a frequency that was too high to remain free of errors. At 39°C, which was the higher temperature we registered without the Peltier, MemTest reported about 150 errors per round of test #5. When the Peltier was connected the temperature slowly dropped and at around 20C the number of errors dropped to around 40. At around 10°C MemTest didn’t report any errors at all.
To avoid as many outer factors as possible we chose to use 8-8-8-24 timings to compare the difference in overclocking with and without the Peltier. The difference isn’t substantial, and if we compare them to the results we achieved earlier in the review we get the following table.
Difference in overclocking
8-8-8-24, [DDR MHz]
Even if we can see a clear positive effect on the overclocking potential at lower temperatures the difference isn’t exactly breathtaking. Our test kit lacks a module for keeping the condensation at bay, which is required so that the memories or any other components are damaged. On the other hand it is possible to get really low temperatures with this kind of cooling since it will most likely be considerably hotter inside a case. But on the other hand you need a water cooling system to cool off the heat emitted by the Peltier.
Before we conclude the article we wanted to run some benchmarks.
Even though MemTest shows no errors running at more than 1050MHz/2100DDR we were hoping to run some of our benchmarks at even higher frequencies by nudging the stability a bit. This turned out to be harder than we thought and among others WinRAR refused to run at higher settings than seen above. When WinRAR was satisfied with the stability we got some evident proof of how extreme the bandwidth gets with the triple-channel memory controller and a nice setup of overclocked memories.
We conclude our experiences on the next page.
The kit we have had the chance of evaluating is an engineering sample, as shown by the pictures. Corsair has informed us that the design is not final and we assume that when this product becomes available it will look more like Dominator GT, which is a new line of products. A circuit board for keeping track of memory circuit temperatures is also something that will complete the product. The solution with the water block is more than OK and you can cool everything from 2 to 6 memory modules with the same water block. You don’t have to break the water loop to install more memories, which is of course an advantage when you have a system connected and leak tested. If you think the Peltier is too much you can simply remove and just use the water block.
The performance is highly satisfactory and the modules lives up to the specifications and more, and if you increase the memory voltage you can push them quite a lot higher. The memories are specified to 1866C8 and our overclocking tests shows that it is possible to reach as high as DDR2100 at the same settings and even higher with looser timings. The water block is proportional for handling both Peltier and the modules. The additional performance the Peltier enables isn’t great, but the lower temperatures could be more useful with a closed system.
Corsair has presented an innovative solution for cooling memories way below room temperature using water cooling and Peltier. The solution is flexible and can be used with a number of Corsair’s new memory circuits and with everything from 2 to 6 modules. During our tests we haven’t been able to register any kind of overwhelming improvements with the Peltier element, but then again it’s not necessary to activate it, which is just another sign for flexibility. The memories have good margins for overclocking and responds well to higher voltages. From a larger perspective you realize that you only need water block for the processor and possibly graphics card for a water-cooled system. This system is also very easy to install without any major modifications needed. We have no information regarding the price, but we can’t imagine it will be cheap.
+ Overclocking potential
+ Flexible water block
+ Water cooling easy to use
We want to thank Corsair for handing us the sample.