Introduction



The story to this overclocking session is in fact a very amusing one as I wasn't intended to break any world record with the card. When the sane review was posted on the front page, MSI called and asked where the LN2 overclocking results were as the Lightning series are designed to be used for extreme overclocking. Although the card was going to be tested on LN2 somewhere this month, having a vendor request LN2 tests is something extraordinary, especially for reviewing websites. So, instead of postponing the session, my colleague Leeghoofd and I decided to do it last Friday.

As agreed, the card landed on my desk a few days early to prepare it for the LN2 session. As we could use software to increase the voltage of the GPU, I only performed the voltage modification for the memory, which you can find on the next page. Afterwards insulating the card was all I had left to do.

The day of the overclocking session, we first did a couple of runs for the HWBOT Country Cup. For those who don't know this competition, it's a country-based challenge hosted by overclocking database HWBOT. Basically, spread over 7 weeks, there are 7 stages, each of which having different hardware limitations (hw you can use to post score) and is using a different benchmark. An average score determines the ranking of the countries and whatever country has the most points in the end, wins the competition. It's a fairly simple concept but quite cool to participate in as it motivates you to ask other countrymen to participate in the competition. This Friday we tried to come up with scores for Stage 7, which is 3DMark01. Afterwards, we were going to do the GTX 275 test session.

After dismantling the Core 2 Duo setup for 3DMark01, my colleague suggested to offer his Core i7 950 to be used in combination with the GTX275 in order to maximize the scores. In fact, he had to insist on using this processor as I was going to settle for a lower performing LGA1156-based platform. In any case, I'm quite happy he did insist, because this is what we scored when completing our second full run of Vantage:



1210MHz on the core and 1400MHz on the memory are already impressive achievements. However, note that in the end, these were my “safety” clocks.

Voltage modifications

As for the Vgpu I didn't even bother looking up the hardware-based modification as there's a simple tool available to increase the voltage through the VID-mode of the voltage regulator. No, it's not the MSI Afterburner software which, in the end, limited me both in terms of voltage options and frequency options. Instead of the Afterburner software, I decided to use the same application we've used when testing the Asus GTX285 Matrix. Let's go back a few months:

As for the application of this I2C feature. I'm not really a very good software programmer, but I was lucky enough that Snot_Aap, a Dutch overclocker, had already programmed this application for the MSI Lightning GTX 260, which was used at MSI's MOA EU 2009 final in Munich, which has been covered by Madshrimps not so long ago. All I had to do was check and partially re-write the code to make it work for the Asus Matrix GTX285!



Link (right-click and save): MadMatrix.rar

How to use:

• Place the 'Madmatrix.bat' file in your rivatuner folder.
  
• Double-click the 'Madmatrix.bat' and enter the HEX value matching the prefered voltage(*)
  
• Make sure you typed the right HEX value and enter 'Y' or 'yes' and press enter.
  
• Check the commands that will be run and press enter.
  
• Check the voltage with a multimeter on the card or with the Asus software utility(*²)
  

  
  (*): The voltages stated in the utility are not the only ones you can use. As said before, there are 256 combinations, so you will have a look in the datasheet (P.15) if you want to apply a different voltage. Note that the voltage has to be indicated by its HEX value, so you will need a BIN to HEX converter to translate VID table. Also note that this is a workaround for the OVP problem: by using the software utility, you will not trigger the OVP. OCP, Over Current Protection, still is an issue though.
  
  (*²): This software has compatibility problems with the Asus software as both re-write the VID settings. Please, do NOT use the Asus utility to change ANYTHING (frequency, voltage or fan speed) if you're using the Madshrimps utility. If you use our utility first and then apply anything in the Asus utility, your voltage selection will have been reset!

The only modification left was the Vmem modification which is also quite important as you will find out later on in this article. For creating the picture, we've used the extremely large picture hosted on TechpowerUP.com

Test setup

Massman's 3D on LN2 setup
CPU	Intel Core i7 950
Cooling	KingpinCooling Dragon F1 Extreme KingpinCooling Tek9 3.0
Mainboard	Asus Rampage II Extreme
Videocard	MSI GTX275 Lightning
Memory	3 x 2GB OCZ Gold PC3 12800 8-8-8 1.65V
Storage	Western Digital 250GB
PSU	Enermax Revolution 85+ 1050W

Pictures



The front of the card was insulated using rubber gum. We used the stock cooling for the memory and pwm installed because it makes the whole card more stable (physically) and lowers the heat transfer from GPU to the memory.



The back was insulated using the same rubber gum.



Setup after an hour of testing.



The longer the session lasts, the more ice is formed on the GPU cooling. On the background, you can read-out the voltage during 3D load.



The ice is spreading!



Session finished ... the ice spread all the way to the left of the card, but never formed an issue during the overclocking session.



PWM has no ice on it. The insulation was perfect.

Results

Ordinarily, I consider 3DMark Vantage more of a task than a benchmark pleasure, however with this kind of high-end video card the Vantage benchmark can be more fun, because the CPU bottleneck, which is present in benchmarks such as 3DMark05 and 3DMark06, is almost non-existent. That's why I've mainly ran 3DMark Vantage and why the overclocking scaling table underneath is based on this benchmark. Do note that I didn't spend time on trying to see the exact scaling of the graphics card; this table is a combination of all the screenshots I've taken during the overclocking session.



The most impressive result is, of course, the 1375MHz core frequency which was perfectly stable through 3DMark Vantage. This is the highest clock frequency we've seen so far on an Nvidia-based graphics card, so needless to say we're quite proud of this result. In any case, given that the voltage modifications performed by the end-user, me, are limited to Vmem only, I'd see a large part of this accomplishment must be credited to the VGA-team at MSI. No coldbug, no OVP, no OCP and perfect scaling ... it are every overclocker's wet dream.



Combining this massive core frequency with an also very impressive 1435MHz memory clock frequency obviously leads to the Nvidia single GPU world record in 3DMark Vantage: P20460!


(Clickable)

A few comparisons to put things into perspective, a Geforce 285 costs about €270 in shops, a MSI Lightening 275 about €255:

fastest Geforce GTX 285 (885/1425) scores 16956

fastest Geforce GTX 280 (945/1250) scores 17665

fastest Geforce GTX 260 (1160/1400) scores 18822

the previous record with GTX 275 (1200/1340) was 18833

Conclusive thoughts

In the past I've had my share of bad luck when putting graphics cards under LN2, but after some practice rounds with the low-end cards, I found the way to insulate the motherboard and graphics card to minimize the damage done by condensation and other cold-related problems. Combining that experience with the overly-impressive GTX275 Lightning and you have a winning. The card has impressed me beyond what I can describe with words ... it says enough that I've created a thread at XtremeSystems when I was running 1210MHz on the core. I was already SO impressive by that frequency that I decided to start a new discussion topic (which is rare for me) ... and the session ended an insane 165MHz higher! All superlatives aside, I'm convinced that this is the strongest card I've ever overclocked.

As a more general note, I'm happy to see that after a couple of months of less activity due to the works at HWBOT.org, the overclocking results posted at Madshrimps are still pretty decent. A few months ago, we had the article of the Asus GTX285 Matrix running at a world record breaking memory frequency (1584MHz), about a month ago we saw the highest shader frequency on the Leadtek GT220 running at 3400MHz and today we've seen the highest Nvidia core frequency. It's so much fun when everything works out so nicely!

A big thanks goes out to several people:

Leeghoofd for insisting on using his Core i7 950 for the tests and the support during the session

Janneke from MSI for sending the card for review and insisting on more LN2 tests

Milan from Asus for sending the Asus Rampage II Extreme

Blind for letting me use the Tek9 3.0 GPU cooling

Tobias from OCZ for sending over the OCZ Freeze Cooling paste

---

Addendum: a video on youtube



This is pretty much the complete Vantage attempt at 1400/2800/1435; it fails just a few seconds before the result comes up ... which you'll notice as I sound pretty disappointed. If the benchmark would’ve finished I expect it to have been 20700-ish