Ram cooling analysis

[pro]

With the rise of P55 and the return to dual channel we have seen many members of the community looking to subzero ram mainly to set high ram frequencys but in some circumstances gaining better clocks from their ram in benchmarks like PI.

Products like corsairs T30 water/TEC based cooling solutions have been on the market for a while now, they definatly provide some advantage in terms of squeezing extra clocks out of your best set of hypers, but for the overclocker looking for a top 32m score, this might not be enough.

Vince has been kind enough to put some time into developing his new Dominance ram cooler, at this stage only fitting the CORSAIR modules, but in the future you will be able to clip this cooler to a number of DIMMs on the market. For my own use I would be looking to find an optimum temperature where I can get good gains out of my modules but not really on the edge.. Ram modules are delicate and it is hard to properly insulate the gold pins on the modules. As a result I would be trying to only go as cold as I had to, with superpi 32m sessions often lasting a long time to get that perfect run, you need to approach ram cooling with a sense of caution as to not destroy your session before it has started.

I have setup a test platform to test the performance gains going from air cooling with a single low RPM fan, to corsairs T30 TEC based solution, and finally Vinces Dominance pot. I am using a pass of memtest, test 5 as a quick test of stability, I am also working up in 25mhz increments (50mhz DDR).

Test Setup:
Gigabyte X58 UD4P
Corsair Dominator 2250 8-8-8-24 triple channel kit
Intel 960 Bloomfield @ single stage -40
Corsair TEC kit and Vinces dominance cooler.

Starting with the preperation process, i just chose to cover the outside sides of the container, to try to stop some of that ice/condensation dropping down into and around the ram slots.





I also setup the sticks with the Corsair TEC kit to see what sort of scaling we would have from this unit.





8-8-8-24
First lets start with the 8-8-8-24 timings. The performance on this kit on air is very impressive reaching 1200, and unfortunatly for me thats where things stopped. As my CPU has a benching uncore wall about 4800, I was not able to flex this kit any higher than 1200 due to the CPU limitation. To test 8-8-8-24 we would need a better CPU or a P55 setup. You can see from this table though that the LN2 is helping straight away with less volts required to hit the 1200.



7-8-7-21
Next to 7-8-7-21, this is where we can start to see the effects of subzero cooling our kit. Able to hit 1100 on air, but we are seeing a 50MHZ (DDR 100mhz) gain when moving from air to LN2. We can also see the benefits of using the TEC cooler here gaining 25mhz while using low voltages on the kit.



7-7-7-20
I took a look at 7-7-7-20 next, we can see a benefit of 25mhz from cooling the ram, in this case I got the scale sort of benefits at 0 degrees as I did at -30 degrees. You can see from the table that there is no scaling at 7-7 when I increase voltages with this kit.




6-7-6-21
What has been very popular with those benching PI is sometimes running a lower ram frequency like 1000 and 6-7-6 timings, here I tested 6-7-6-21. We again can see some significant gains when looking at LN2 over air/tec of around 50mhz. The trend of low voltage gains on the TEC again happens at these timings and would suggest the TEC would be great for the home user wanting to push to the absolute limits of their kit without taking any risks with their hardware.




6-6-6-20
Lastly I used the timings of 6-6-6-20, with a real tight TRCD of 6 we always drop our clocks a fair way and this is probably the most shining example of the gains on LN2. We moved from 925mhz limit on air/tec to 1025mhz on LN2 which is a massive gain for this timings. With 1000mhz+, TRCD 6, tight subs and a reasonable uncore the bandwidth starts to look really good and this could certainly be a very damaging PI setup, especially for those still benching clarkdale with low ram mhz issues.



In summary it is clear the advantages as we add cooling to our ram. For those wanting to push the best scores in PI, AM3, 05 then this might be mandatory in the future. Users really wanting to push the limits might look at more voltage and colder temperatures and we might start seeing 1200 7-8-7-21 and similar with great kits.

Here are some additional photos of my setup and testing process..

[k|ngp|n]

VERY NICE testing pro THANK YOU!

[chispy]

Well done analysis James , a great ammount of work involved in your testing , thank you for the hard work and information , it comes handy and very useful.

[Oliver]

neat write up Pro and sweet unit Vince

[diezel]

i have orderd one

[Gunslinger]

Very nice write up Pro, hope to see you revisit it with a better CPU.

[nickshih]

wow , great testing any vasline on it ?

[VasGTO]

Cool... Thanks for the info. To bad you didn't use Andre voltages, he would have zapped them with 2.0v I wonder what kind of frequencies/timings you would have reached with that...

[pro]

thanks guys,

yeah nick, plenty of dielectric grease (or you could use vaseline) on the pins of the ram, and also in all of the ram slots on the board, it might be worth for prolonged sessions getting plenty of paper between the sticks and even some eraser in between the little gaps between the ram slots..

all of that type of thing will make your session last alot longer

[pauld5999]

thanks for all the info and testing, awesome stuff