Extreme overclocking and 'benching'



People who regularly read this website will probably have noticed that the crew of Madshrimps basically consists of less normal reviewers; people who act a little off when around the newest released hardware, hence the name of this website. With permission of the manufacturers and hardware shops, we take hardware to the next level by overclocking on air. Well no, not really ... on liquid nitrogen that is. Using both dry ice and liquid nitrogen, subzero temperatures gives us the possibility to clock our hardware to frequencies unseen on regular air cooling and let us see the real power of the technology on our test bench. In some way, we see the future, as we can predict what will happen when technology is run at higher frequencies. Next to that, we provide manufacturers with information on what components should be improved and which are already decent. For instance, if a certain motherboard blows up in one of our tests, we let the manufacturer know that the power management system of the motherboard is not capable of running at elevated frequencies and the manufacturer can keep that in mind when refreshing the board and release a new revision.

Now, many of you probably think that manufacturers are not interested in hardware running at subzero temperatures, because, let's face it, no one is running their system 24/7 cooled by liquid nitrogen. However, by overclocking we don't instantly kill our components, no; we speed up the cycle of life to see what will eventually kill your system. It's like injecting steroids in a human being: the person will be able to perform much better, the chances of getting a fatal injury are elevated but when it eventually kills, scientists will know what is making the steroids so dangerous. Instead of waiting years and years to find out how long our motherboard lasts, we wait weeks, days or sometimes just hours.

Unlike how it may have come across in the previous paragraph, overclocking isn't that dangerous, because most motherboards are very well built. More and more people overclock extreme and less and less people actually break hardware. Well, that last sentence may not be entirely true ... people tend to not kiss and tell about their latest hardware deaths.



Anyway, after the overclocking part of story comes the benchmarking, or in short 'benching'. Which is in fact the most interesting part of overclocking as what we do is basically testing different technologies in different test environments. Certain applications can tell exactly how fast your computer is in comparison to other systems of other people around the globe. HWBot.org (developed by a Madshrimps reviewer and supported by our site actively) is at the moment the only website that gives these benchers the opportunity to really compete against each other in a global overclocking league for both individuals, teams and countries. A person needs to run a series of benchmarks, upload the scores and see how his (or her) results stack up against the other results in that category. If it's better than most results, it'll probably be awarded points, which are added to the personal total and to the team's total. Obviously, the more points, the higher you are ranked and the better you are overall. On a sidenote, Madshrimps was able to reach the 12th spot in the ranking of best overclocking teams in the world, not too shabby for a small Belgium OC team, eh!


(competition is stiff though, we moved down a few place since this screenshot was taken^)

Cooling

Now, the main problem of this extreme overclocking is the actual cooling part. We work with temperatures around -78°C (dry ice) and -196°C (liquid nitrogen) and we need to find a way to use these small pieces of ice and the fluid nitrogen to cool down the hardware. How? With containers or pots of course. Basically, you drill a hole in a piece of metal, attach that piece of metal to on your CPU or GPU and throw in the dry ice or liquid nitrogen. It's quite hard to explain using words only, but you'll understand it much better when I'm actually showing you the pictures of today's extreme cooling units.

On with the show ->

Dragon F1 Extreme



"Once upon a time" ... sorry for that, but I really need to tell the background story of this container. The creator of this cooling unit is Vince Lucido, better known as K|ngp|n, who has been in the overclocking community for a long time. Since a few years, he spends time designing LN2 cooling units to make the overclocking process more pleasant, temperature-wise. In the past, people used brazed containers, which is a metal construction where you attach a base block to a tube of metal. In the early days, these type of containers did in fact work quite well, as most of the extreme overclockers were spending their time with Athlon64 processors. Those processors did not have such a big heat load and were easy to clock to their limit using dry ice cooling. However, as technology moved on, the hardware generated a higher heat load and in the end, our brazed cooling units were not capable of keeping the processors at low temperatures without losing too much of our cooling substance.

Being an overclocker himself, K|ngp|n decided to design his own cooling units to get the most out of his hardware and not be bothered again by temperatures swinging from -100°C to -50°C in just a few seconds. After a series of high-quality test units, the first 'Dragon F1' units saw the daylight in September '07, more than a year ago. These first units were called thé best in the world by many top overclockers and not much later a second design, the Extreme version, was released.

Did you know, by the way, that K|ngp|n uses the term 'Dragon' for his CPU cooling units and 'Tek9' for his GPU cooling units?

Package




The package is, it has to be said, quite original. I can only remember seeing this kind of boxing method once in the hardware business as a manufacturer of video cards, of which I forgot the name, used the 'tube' for one of it's special release cards.

Features:

Solid piece copper design and construction

Universal mounting for sockets 754, 939, AM2 and LGA 775 based motherboards

All mounting hardware and insulation included

Dynamic LN2 temperature control

Built for Xtreme processor overclocking and performance

Inside the tube

Dragon F1 Extreme container

Insulation material for cooling unit

Mounting for cooling unit

Normally, we would be spending a few lines on the base of the container, which is one of the more important aspects of an extreme cooling unit. However, we want to respect the wishes of the creator of this product, K|ngp|n, and decided not to publish any pictures of the inside of this container. Feel free to browse the internet and look around for inside pics; I know there are already a few around that allow you to look inside the container.

Mounting procedure

The mounting procedure is really straight-forward and is quite simple even if you haven't seen the unit being put together before.






This is a very good way to balance the pressure on the container. If the pressure is not balanced properly, your cooling unit isn't touching the processor completely, making temperature readouts very difficult and will make your processor unstable due to temperature fluctuations.

Test setup

Madshrimps' Ph-II LN2 Test Setup
CPU	AMD Phenom II X4 940
Cooling	Dragon F1 EE
Mainboard	DFI Lanparty DK 790GX-M2RS
Memory	2 * 1GB G.Skill PC8500
Other	Sapphire HD 4870X2 Antec 1000W PSU Western Digital 320Gb SATA HDD Windows XP SP3 OCZ Freeze thermal paste

Methodology: Cooling competitors

We'll be comparing this container to two other own-made cooling units by two Belgian overclockers, Troman and Jort.



The first container is made by Troman, who was inspired by the first Dragon Evo container of K|ngp|n. The base design consists of a series of steppers and a central cylinder, both to increase the surface area. The more surface area, the more room the cooling substance has to cool down the container and thus the processor. The container is a solid copper design, which means cut from one piece.



This container is a very old one, made by Jort many years ago. It was one of the first extreme cooling containers in Belgium and served me quite well when I was playing with the Athlon64 as you can see in this Athlon64 3000+ overclocking article. The base design is very simple: just a place of copper with a few holes in it to increase the surface area. As you can see, this is not a solid container, but a brazed one, a type which I already introduced on the first page.

NOTE: We decided to withdraw the “Jort” container from the tests due to mounting problems. To show you what old-skool containers look like, we'll keep the pictures up, though.

Methodology: Temperature

Finding a decent methodology to test these containers was in fact the biggest challenge of this review. Not only do we have to measure temperatures, but we need to find a repeatable test procedure, which means that the obtained results should be independent of the reviewer and the results must be significant enough to draw conclusions from.

First of all, we chose the Phenom II platform because the lack of any form of coldbug, which makes it much easier to work with in subzero environments. The DFI 790GX motherboard has already proven that it's capable of providing the necessary juice for high-demanding Phenom II quad cores. The insulation was kept the same over the different test sessions.

We will explain the precise test methodology on each page containing test results.

Superzero behavior

In the contrary to what you probably expected to see, we start off with a test with temperatures above 0°C. We booted the system at 3GHz and 1.5v and checked the temperatures going up from 25°C each minute.

• 200x15 = 3GHz
  
• +200mV = 1.5v
  
• System idles on Windows desktop
  
• CPU temperature read-out with the DFI Smart ITE Smart Guardian (verified with Speedfan)
  
• Base temperature read-out with Fluke 51 II
  
• We start the measurement when the CPU is at 25°C

The Dragon F1 seems to be able to hold the 'load' much, much better than the custom made container of Troman, most likely to the copper mass of the Dragon F1. Since the base of the custom container doesn't consist of a lot of copper, the container heats up faster than the F1, which makes it harder to keep the temperatures of the CPU down. The heat-up works in fact both ways: the processor heats up, so the container becomes warmer. But, the warmer the container is, the more difficult it is to absorb heat coming from the processor and cool it down.

Note that the custom container only has four data points whereas the Dragon F1 has five. The reason is very simple: the system has shut itself down to prevent the processor from getting damaged. After all, the temperature of the processor was beyond 80°C!

From super to subzero

When you're setup is running in windows and everything seems okay, the next step would be to take the temperatures down to a more overclock-friendly levels. We tried to simulate the speed with which the container is able to decrease in temperature by measuring the decrease in temperature in 5 minutes time, constantly having the container filled with LN2, of course.

• 200x15 = 3GHz
  
• +200mV = 1.5v
  
• System idles on Windows desktop
  
• CPU temperature read-out with the DFI Smart ITE Smart Guardian (verified with Speedfan)
  
• Base temperature read-out with Fluke 51 II
  
• We start the measurement when the CPU is at 25°C
  
• Container constantly filled

As you can see, the custom container is capable of a very rapid decrease in temperatures, much faster than the F1. This can again be related to the amount of copper in the base of the container: as the custom made container has very little mass in the base, the liquid nitrogen cools down the container faster, which allows the processor to be cooled down faster. The F1 has much more copper mass in its base, so initially, the liquid nitrogen cools down the copper.

Note: the fifth data point of the Custom - CPU chart is -90°C, but should be around -100°C. Both ITE Smart Guardian, the DFI utility, and Speedfan were incapable of reporting temperatures below -90°C.

Subzero behavior

Onto the subzero test results, which are after all, the most important ones. For these tests, we increase the load of the processor by increasing the frequency and voltage to 4.5GHz and 1.7v. Next loaded up Wprime 1024M on all 4 cores to generate a heat load.

• 200x22,5 = 4,5GHz
  
• +400mV = 1.7v
  
• Idle = Windows desktop
  
• Load = Wprime 1024M
  
• CPU temperature read-out with the DFI Smart ITE Smart Guardian (verified with Speedfan)
  
• Base temperature read-out with Fluke 51 II

This is pretty much what we expect from the containers in a situation where there's no cooling liquid and the setup idles: the Dragon F1 is able to keep the increase of the temperature within a closer range than the custom container does.



Running Wprime on all four cores having no liquid nitrogen being poured into the container is a pretty difficult situation for the containers. The custom container is not at all capable of keeping the system running: after about 60 seconds, the setup shuts down because of heat related stability issues. The Dragon F1 is capable of running almost 5 minutes, but eventually shuts down as well before the 5 minutes mark.

Now time for the real deal, let’s pour some LN2 in the containers:



This test is pretty interesting. In a situation where the processor is heavily stressed, but we keep the container filled with liquid nitrogen at all times, both containers perform pretty much the same. This indicates that this custom container is quite capable of handling a Phenom II, or other high-end processors.

Container efficiency

The last test is all about efficiency ; no not power efficiency, but how much LN2 has to be used to keep temperatures stable, or how does the container react with a limited amount of liquid nitrogen available. Most interesting to find out, as the LN2 supply is never endless.

• 200x22,5 = 4,5GHz
  
• +400mV = 1.7v
  
• Idle = Windows desktop
  
• Load = Wprime 1024M
  
• CPU temperature read-out with the DFI Smart ITE Smart Guardian (verified with Speedfan)
  
• Base temperature read-out with Fluke 51 II
  
• 100ml of LN2

This chart shows you the difference in temperature between the starting point and the lowest temperature achieved with this 100ml of LN2. The higher the value, the more the temperature decreases after the LN2 was poured into the container.

We see that the the custom container is in any test scenario the 'better' container: the temperature decreases the most. However, should we really be considering more as better? When you're benching, you need temperatures to be stable, not going up and down all the time. This test shows that our custom made container will does not give you any room for mistakes: if you pour 100ml too much at the end of the benchmark, you're very likely to see the temperatures drop really fast when the benchmark finishes, which increases the risk of having your system shut down due to coldbug issues.

Note: the temperature difference of 'Custom - Idle - CPU' should be much higher; the temperature was definitely below -90°C, but our temperature reading software is incapable of reporting temperatures below this value.



In this chart, you see the time a container needs to return to the starting temperature level. For instance, if we start at -70°C, we measure the time it takes before the container is back at -70°C after pouring the 100ml of LN2.

In idle mode, we see that the custom made container returns to its starting point almost twice as fast as the Dragon F1 does. In load situation, the return point is almost the same.

If we combine above two charts, we can see that the Dragon F1 provides a more stable temperature with certain amount of LN2. With the custom container, the temperature seems to fluctuate more: very rapid decrease in temperature, but also a very rapid increase again.

Evaluation

Time to wrap things up. The Dragon F1 EE is a very decent container, there's no doubt about that: temperature wise, it provides you a very stable platform to overclock with. Pouring a bit too much is not really a problem, so there's room for mistakes, but more important this container gives the user the opportunity to focus on tuning the configuration. As you need to spend less time focusing on the temperature, you can focus more on setting up the system for the highest performance; I'm thinking of setting the correct clock frequencies, applying BIOS/OS tweaks and so on. The container does need more time to decrease the temperature of the processor going from +25°C to subzero temperatures, but that can hardly be called a flaw.

I have to say that the mounting of this container is by far the best I've seen: it's very easy to switch sockets, to mount the container and apply an equal pressure. It has even come to this that I'm using the Dragon F1 mounting system with my Mach2 single stage cooling system and even with basic air cooling, I don’t shy away to use this mounting.

Price-wise, the Dragon F1 is not the cheapest around ($300+). The creator of this container has spend a lot of hours working on the design, so naturally, the container costs a bit more. I do believe that this is the biggest drawback of the container, a lot of people don't want to spend that much on their first LN2 container, so they look for cheaper alternatives. The actual price I cannot provide, but you can always contact K|ngp|n himself and check with him for a price (Vince/KingPin).

+ Temperature stability
+ Mounting
+ Support (new socket = new mounting available)
- Price ($300+)

Dragon F1 EE recommended For


Conclusive thoughts

Having spoken to many other extreme cooling overclockers and having tested more than these two containers, I have to say that most containers are in fact capable of pushing your processor to its limit. The biggest difference between most custom made containers and this Dragon F1 EE is how easy it is to maintain a certain temperature. If you are able to control the temperature on your $25 container, then there's no reason except the mounting why you should choose for the Dragon F1 EE. However, if you're looking for a container that controls the temperature almost on its own, one where you are not obliged to constantly watch'n'pour, the Dragon F1 EE definitely is a very good choice.

Before I close off, I just want to say that this article has been a real adventure for me. Since I have not yet come across another review of LN2 containers, it was quite difficult to find a decent methodology. With the experience I now have, I know that there are still much aspects I can improve. In addition, I should have checked the Dry Ice cooling performance as well: certain containers perform better with dry ice (and worse with LN2) because of the design of the base of the container. That'll be something for another article, though. Note that this article is one of the more expensive ones I've done so far. Not so much hardware-wise, but the liquid nitrogen in Belgium is just not sold at a very cheap price (€2.75 per liter), so please don't expect me to test one container each month.

To end, I'd like to give my appreciation to:

Wendell and Erik from DFI for providing us the DFI LanParty DK 790GX-M2RS

Steeve from AMD for the Phenom II X4 940 and HD4870X2

Tobias from OCZ for the package of OCZ Freeze

Vince for the Dragon F1 EE