Overclocking study – Intel Core i7 4790K & Intel Pentium G3258 – air, water, phase-change, DICE & LN2
Goooood morning LAB501 !!! Finally, the article you have all been waiting for is here ! Heck, to be honest, this is also the article we were also waiting for. Even since the dawn of LAB501, we always wanted to do a complete overclocking study, using most of the cooling methods out there. And we never did a full study, because we are very thorough in our work, and each time we would start we would give too much attention to one of the cooling methods and in the end we could not treat all cooling methods properly. Just imagine that one of Ramiro’s air cooling reviews had 96 pages, and you will understand what I am talking about.
But enough about that – now we finally managed to do a full overclocking study, starting from air cooling and going through the most important stages. It doesn’t matter what type of overclocker you are – you might be a casual overclocker who wants to boost his system’s performance in daily-use, or a hard core LN2 user who didn’t try Core i7 4790K or Pentium G3258 yet – this study is for both of you!. From normal ambient air-cooling to low ambient air-cooling, from All-In-One water cooling solutions to extreme water cooling, from Single Stage phase change cooling to Dry Ice, and finally all the way to Liquid Nitrogen… we have tested all!
It’s all about numbers in this study, and that is why we didn’t test one CPU, but two – the overclocker’s dream Intel Core i7 4790K, and the budget overclocker Pentium G3258, or Pentium 20th anniversary edition, as it is also known. If you want to have the most powerful gaming rig, or break world records – Intel Core i7 4790K seems the logical choice. If you want to have fun overclocking, or maybe learn how to do it, Pentium G3258 is a great first step, and it is also very affordable (72 USD).
But we didn’t just tested 2 CPU’s… this review is also a team work result between my colleague matose and me. He is the one who spent hundreds of hours in the lab pushing the CPU’s with all the cooling methods available, and he is going to show you the results and analyze the scaling, while I am going to explain each cooling method we used and which users it is suitable for.
So… are you excited? Are you ready for a great journey from air cooling to liquid nitrogen? Are you ready to see some impressive numbers? Then join us as we guide you trough one of the greatest overclocking adventures ever. Let’s go!!!
This is LAB501, house of world champions overclockers and die-hard overclocking fans, so normally I wouldn’t explain to you guys what overclocking is. However, this time, and only this time, I am thinking that maybe we have new readers interested in overclocking, or guys who only used air so far, and are reading this story to learn about alternative cooling methods. So… the die-hard bunch can skip this page. Newbie overclockers, this page is for you. Pay attention!
You wouldn’t believe this, but actually I often get asked “what is overclocking”. Be it at an event where we do a live show (like Dreamhack or LSE), be it when some of my newer friends ask me exactly what we do for a living, or what was it about in the contest they saw we won… And the answer is always the same – overclocking can be defined as pushing the limits of a PC component, beyond the speed planned by the manufacturers.
Basically, everything in a computer runs at a certain speed (frequency), even if we talk about the CPU, graphic card, memory, PCI devices, S-ATA controller or USB controller. With some of those speeds you don’t want to mess because the system will stop working properly (for instance S-ATA, PCI, etc), but there are also those speeds that you can slightly adjust without causing problems. That is the case with the CPU, the VGA (graphic card) and the memory. You can increase the clock beyond the factory specs and gain performance while you do that.
Of course, it is very easy to overclock nowadays, but this was not the situation many years ago. In fact, manufacturers would try anything possible to stop you from doing it, because you wouldn’t need a more expensive PC if you could overclock. However, the funny part is that some of the first “overclocks” were done in 1983 by none other then… Intel. I’m not kidding – they would sell their Intel 8088 to IBM, who would set the CPU clock in their PC at 4.7MHz, while Intel was offering the same 8088 model clocked at 8MHz.
However, hardware aficionados didn’t have an easy ride overclocking back then, since most of the times you would need some hardware knowledge in order to modify the parts needed for overclocking, because most of the parts in a PC would run at the same clock, mainly the BUS speed. The first big breakthrough came in 1992, with the introduction of 486DX2 (or i486DX2), when a concept we all know and use today was introduced – the multiplier.
This allowed the CPU to run at higher speeds then the bus speed (2x in this case), and many people “upgraded” their 486SX to DX2 speeds (from 25MHz to 33MHz) by moving a jumper on the board. But one other thing also got introduced to later 486 models, and that was the fan mounted on the radiator. Until then, all CPU’s were passively cooled with a small aluminum or copper radiator.
1998 brought another important breakthrough – the birth of Celeron 300A (a 300 MHz Mendocino core CPU, based on a 66MHz FSB and a locked 4.5 multiplier) and Abit BH-6. Celeron 300A was one of the first real overclockable CPU’s, and motherboards like Abit BH6 made that possible. You could ran Celeron 300A at 450MHz simply by using the 100MHz FSB, and most of them would actually be stable at those clocks.
But the introduction of the Abit BH6 board meant something else – it was the first motherboard that would allow you to make all the settings in BIOS, without any need to move jumpers or dip switches on the motherboards. And I think that with this combo, 15 years ago overclocking started to become something mainstream, something that most of the PC enthusiasts could enjoy, and not only the hard core engineer bunch.
Many things changed in the last 15 years and we are not going to go into too much detail, because this is not a history article but a scaling review with the latest CPU’s from Intel. However, we have a few more issue to address, in order to better understand the current situation with overclocking.
Since 15 years ago all motherboard manufacturers started to pay a lot of attention to overclocking and in the last decade we could see motherboards specially tailored for the needs of overclockers, after market cooling solutions becoming more and more popular, liquid nitrogen pots mass produced for retail sale and world wide overclocking championships with cash prizes.
Overclocking became very popular but that also became a problem for CPU manufacturers like Intel and AMD. Because you couldn’t simply block features on motherboards anymore but you also didn’t want anybody to be able to take the cheapest CPU and overclock it until it would get to the performance of the most expensive CPU.
CPU manufacturers used to block some features on the cheaper CPU’s since the 486 days, but now it was time that this was done in a different way. And that also meant that the PC enthusiasts got some attention and recognition from the manufacturers. That was the time when both manufacturers created top of the line series for those who do not care how much they spend for the top performance – the hard core overclockers, the PC enthusiasts, the heavy gamers. That is how AMD’s FX line was born and that is how the Extreme Edition was born, a series of unlocked CPU’s that stretches from the Pentium 4 days up until today, with CPU’s like Intel Core i7 4960X Extreme Edition available on the market and models like Intel Core i7 5960X (8 core Haswell-E CPU) launching sometimes this year.
The guys who wanted everything, the most powerful CPU’s with unlocked multipliers, got what they wanted and everybody else could still get a healthy overclocking on locked CPU’s by raising the FSB. However, the situation changed dramatically when the new Sandy Bridge architecture was introduced, and BUS overclocking wouldn’t bring you too much extra MHz at all. But unlike Pentium days though, overclockers and enthusiasts have become a very important part of the market in the meanwhile so this issue had to be addressed somehow.
In the year before the Sandy Bridge launch, Intel was very well aware of this, so they made a move that not many people understood at the time – the launch of a second special SKU, designed for overclockers and gamers. The K SKU first came to life with Intel Core i5 655K and Intel Core i7 875K. The two models had fully unlocked multiplier, but unlike their Extreme Edition counterparts at the time (Intel Core i7 980X) they wouldn’t cost 999 USD. The Core i5 655K had a price tag of 216 USD while the Core i7 875K had a price tag of 342 USD. Much closer to the needs of overclockers and enthusiasts, I might add.
After that all Intel families would have an unlocked Core i5 and Core i7 version, designed for overclockers. We are talking about Sandy Bridge, Ivy Bridge, Haswell and Haswell refresh. And even the Sandy-E and Ivy-E series have their own K model. And since Sandy Bridge these K models have roughly the same price – ~300-350 USD for the i7 version and 216 USD for the i5 version. In fact, Intel’s price scheme didn’t change much since Conroe – the high-end would have a 300 / 560 / 999 USD price tag and this is what we can see today with the SB-E or Ivy-E families.
Sandy / Ivy Bridge and Haswell topped out < 350 USD for the Core i7 K and < 250 USD for the Core i5 K and everything bellow that price is locked and you cannot achieve much through overclocking. If you compare this situation with 2006 for instance you would see that it’s even better for overclockers, since the non EE CPU’s didn’t have unlocked multiplier at all (think E6300, E6400, etc). So in the end things did improve over the years even from a price point of view, if we think about CPU’s, because now you get an unlocked multiplier at the same price you got a locked CPU (that you would overclock using FSB) a few years ago.
But there was only one little thing missing – an affordable CPU for those who want to try overclocking but don’t have the budget for a Haswell K. And this year, besides the 350 USD Intel Core i7 4790K, we all got another very nice present from Intel. It celebrates 20 years of Pentium, it has an unlocked multiplier and two Haswell cores, and it only costs 72 USD. It is called Pentium G3258, and this is what we are going to test today, together with the Intel Core i7 4790K beast…
Architecturally speaking the Core i7 4790K is the same as the Core i7 4770K, released more than a year ago. One of the problems regarding Haswell generation processors is the exponential rising of the temperature with the increase of the voltage supplied. This is due to reduced size of the die combined with the use of not so good TIM between the die and IHS. Even if you have a very good CPU cooler you will see the temperatures rising very fast once you stress the CPU, because the heat couldn’t be transferred fast enough from the die to the cooler base (bear in mind that you have the following structure: die -> Intel TIM -> IHS -> your TIM -> cooler).
The Core i7 4790K is running at a base frequency of 4GHz with the Turbo Boost allowing it to reach 4.4GHz for light single-threaded applications. This means the 4790K will be a much faster CPU out-of-the-box than the existing 4770K which is running 500MHz lower. For this frequency to be possible, Intel had to raise the VID for the CPU’s and do something about the heat problems and. The solution came through a new and improved TIM between the die and the IHS, although it will have been better if the IHS was soldered to the die like in the Sandy Bridge era.
Although Core i7 4790K is a performance monster, the real star of the show is the dual-core, non Turbo Boost Pentium Anniversary Edition. With a price tag of only 72$, the Pentium G3258 is a fully unlocked dual-core Haswell based CPU, having the operating frequency set at 3.2GHz for a TDP of only 53W. The frequency doesn’t really matter because having only 2 cores it will probably overclock like crazy and nobody will keep it at default clocks.
|CPU||Core i7 4790K & Pentium G3258|
|MB||GIGABYTE Z97X-SOC Force|
|RAM||Team Zeus 2 x 4GB DDR3 1600 / Corsair Dominator Platinum 2400 CL9|
|Air Cooling||Noctua NH-D14 + 2 x Coolink SwiF 120p @ 2000 rpm + Arctic Cooling MX-4|
|Air Cooling - low ambient||Noctua NH-D14 + 2 x Scythe Gentle Typhoon @ 5400 rpm + Arctic Cooling MX-4|
|AIO||Enermax Liqtech 120X + Arctic Cooling MX-4|
|Extreme Water Cooling||EK-Supremacy Clean CSQ + EK-DCP 4.0 PWM + Gelid GC Extreme|
|Single Stage||DimasTech Single Stage+ Gelid GC Extreme|
|Dry Ice||SF3D Inflection Point + Gelid GC Extreme|
|LN2||EK-SF3D Inflection Point Evo + Gelid GC Extreme|
|SSD||Intel SSD 730 240GB|
|Case||Dimastech Benchtable Easy v2.5|
|Ambient temp||25 oC (17-18 oC for low ambient air cooling)|
|OS||Windows 7 Ultimate x64 SP1|
For this test we used a motherboard designed for overclocking, GIGABYTE Z97X-SOC Force, to allow our two candidates to scale properly without any artificial limits. For LN2 tests we used the Corsair Dominator 2x4GB 2400MHz CL9, while for the other scenarios we used the Team Zeus 2x4GB 1600MHz CL9 (to fit nicely under the monster Noctua NH-D14).
While the use of a skinned Windows XP 32bit would have been nice talking strictly about benching, this article is not just about that. We are talking scaling and stability starting with plain aircooling and finishing with LN2, so to make the article relevant also for the average home user we used Windows 7 64 bit for all tests. For each scenario (except LN2) we tested the stability with at least 10 minutes of Prime95 64bit using all cores, a light single threaded stability with SuperPi 1M and the absolute maximum frequency with a CPU-Z validation.
Now there is one more thing – as I said, this review is read by die hard overclockers but also beginners. So before we show you the results for each cooling stage we will also briefly explain it with a simple language, so everybody can understand. Forgive us, die hard fans, but everybody started from somewhere and needed to learn…even you…even us. So be patient and bear with us.
Ok, in order to achieve a nice looking clock speed you need an unlocked CPU, an overclocker friendly motherboard and a good quality power supply, among other things. But one of the most important aspect of overclocking is cooling the components we want to overclock, in this case the CPU. Because higher frequency and voltage will always mean higher temperatures. Don’t get me wrong, the CPU’s have their internal protection mechanism that will always lower the voltage or even shut the system down before anything happens to you CPU.
So it’s not a “oh my gosh, I am afraid I will burn my CPU” situation, but if we want to reach system stability, or the highest clock possible, the temperature will eventually stop us. When we are looking for stability, it will stop us because eventually the CPU will go into throttle at one point, that means that not all cores will be active and the speed we set will no longer be applied, so the performance will drop. When we are looking for the highest clock speed temperature will also stop us after a point, and we will need a different type of cooling in order to apply higher voltages and get higher clocks.
Well, the first type of cooling we are going to use when we overclock is obviously air-cooling. Simply by switching the flimsy stock cooler with an aftermarket beast, like Noctua NH-D14, will get you more MHz. Using a good quality thermal interface material and some strong fans will also get us extra MHz.
Any type of PC component cooling is based on a simple principle – heat exchange between the thing we are trying to cool (CPU) and what we use to cool it (cooler). Air cooling principles haven’t changed in a while – you apply TIM on the CPU in order to maximize the heat transfer then you use a cooler with a good mounting system and a good quality base to take the heat from the CPU and use the cooling agent to keep the CPU temperature down. In the case of air cooling, the cooling agent is air. And the performance of our cooler depends on the temperature in the room – the ambient temperature. Basically you can only cool your CPU down to the temperature of your room (at most).
The first CPU’s didn’t even need a cooler, then a small aluminum or copper radiator was placed on the CPU to dissipate the heat into the air, then a fan was added to help with this process. With time coolers got bigger and bigger, in order to keep up with the power consumption of the newer CPU’s, and after a while the heatpipe was invented. Now the base of the cooler would take away the heat from the CPU, the heatpipes would transport it to the fins, and the fans would help with the heat exchange between the surrounding air and the fins. The cooling agent in the heatpipes would get cooler, it would go down to the base, realize the thermal exchange with the CPU and then the whole process would be repeated.
As we already mentioned, all our tests (except the low ambient one) were done in the same conditions – 25 oC ambient temperature. We used one of the best air coolers out-there (Noctua NH-D14), together with a good quality TIM (Arctic Cooling MX-4) and two fans (Coolink SwiF2 120P) that can get up to 2000 rpm. This is the situation most relevant for a person that wants to find the best clocks for daily use, with a retail cooler and some fans that don’t blow your ears away.
The next step was the low ambient air-cooling – we lowered the ambient temperature to 17-18 oC, we switched the fans for some 5400 rpm monsters and we gave it another try.
Our Pentium G3258 sample had a VID of 1.06v, but bear in mind that each an every CPU has a different VID. For the first step of the stability test, we set a voltage of 1.15v and looked for the highest frequency at which we could run Prime 95 for at least 10 minutes. The little Pentium reached with ease 4435MHz and the temperature was only 60 degrees C on the hottest core.
The next step was setting the voltage up to 1.3v and trying again for the higher full stable frequency, which in this case was 4766MHz at a temperature of maximum 75 degrees C. We can see a nice scaling here, but let’s see if it scales further.
We are aproaching the thermal limits of the CPU, so we are upping the voltage one more time to 1.425v. With a maximum temperature of 92 degrees C, the G3258 manages an excellent 4847MHz Prime95 stable. We can see that the scaling is not so spectacular now, mostly due to the high temperature.
For the light single-threaded SuperPI 1.5 benchmark, the little Pentium flies over the 5GHz mark and manages a complete run at 5040MHz with a voltage of 1.525v. Using the same voltage we can get a maximum frequency validation of 5140.8MHz. In case you are wondering, the Pentium G3258 has fully unlocked memory and NB multiplier that work without any issues as you can see in the screenshots.
Now we are on to serious business, from 2 cores / 2 threads we jump to 4 cores / 8 threads. The stock VID measured with Fluke 77 III multimeter is 1.3v, but other CPUs can have different VID. First we are testing for stability, so we are running Prime95 searching for the highest stable frequency. Normal users cannot induce that much stress for the CPU as we are doing using Prime95, so the fact that the CPU is close to 100 oC should not surprise us. Take notice that you will not experience any issues under normal operation.
For the first step, we set the voltage to 1.15v and managed to run stable all cores at 4369MHz with the temperature reaching 91 degrees C. The maximum possible voltage without entering throttling is 1.22v, stabilizing the CPU at 4502MHz with a sky high temperature of 102 degrees C (throttle starts at 105 degrees C so we are at the limit).
Deactivating HyperThreading and 2 cores we can push the Core i7 4790K to 5090MHz at a voltage of 1.512v in the light single-threaded SuperPi 1.5. Using the same voltage we could make a validation at a frequency of 5202.51MHz. It’s not the greatest Core i7 4700 series CPU you can get, but as you can see the heat is its main problem so I’m very curious to see how it scales on more powerful cooling systems.
Lowering ambient temperature and changing the fans to serious screamers will allow us to keep the CPU cooler and we could push more voltage. Using 1.5v we could get the Pentium G3258 Prime95 stable at 4947MHz with a maximum temperature of 87 degrees C.
With a voltage set at 1.55v we saw some improvements also for SuperPi 1M (5126MHz) and maximum frequency validation (5247.18MHz).
The low ambient allowed us to push 1.3v through our Core i7 4790K, which resulted in a Prime95 stable speed of 4653MHz and a temperature of maximum 95 degrees C.
The lower temperature allowed higher frequencies also in SuperPi (5146MHz) and CPU-Z validation (5236MHz). Higher voltage didn’t help so I used the same 1.512v running through the CPU.
Water cooling has been around for ages now and it got popular because of a very simple reason – water has a much better thermal conductivity compared to air. A water cooler has more components – the water block is the part that takes the heat away from the CPU, the radiator is where the water gets cooled and the pump moves the liquid in the loop. Ironically, in the end our cooling agent (water) is cooled using the same cooling agent as in the case of air cooling (fans blowing good old air over the radiator). But the heat exchange between the ambient air and the CPU is better because of water’s thermal conductivity and because of the bigger surface of the radiators.
However, serious water cooling loops do have some downsides – like the fact that you have to buy all the expensive components and carefully assemble your loop. If you don’t want to go that far, there is another solution called All-in-one water cooling, or more simply AIO. That means that you are buying a closed loop already assembled and you just have to mount it in your case. The main advantages are the fact that you don’t have to worry about leaks and stuff like that and also the more friendly price. The downside is that you are limited to the components the manufacturer chooses to use (pump, radiator, etc) and many times the pumps included in an AIO system are not as powerful when compared to a DIY kit, for instance.
In this case we used a pretty strong AIO system from Enermax, called Enermax Liqtech 120X, comprised of a waterblock + pump + a 120mm radiator with two 120mm fans.
Using AIO on the little Pentium G3258 we got the exactly same results as those obtained on air cooling. The only difference is the temperature which now reached 99 degrees C in Prime95 at 4846MHz and a voltage of 1.425v. Seems that the IHS planarity of the Pentium is better matched with the base of the NH-D14 cooler than with the base of the AIO.
With the Core i7 4790K on the other side we get better temps with the AIO than with regular aircooling. It’s not a problem of cooling capability, but more a compatibility between die size, IHS planarity, cooler base and TIM application method.
At the same voltage of 1.22v we get 101 degrees C and a maximum of 4507MHz Prime95 stable frequency. If we up the voltage to 1.24v we can push the CPU all the way to 4544MHz at a temperature of 103 degrees C. Slightly better than air in this case, but far from spectacular improvements.
With a voltage of 1.517v we get single-threaded stability in SuperPi at 5102.46MHz and a CPU-Z validation frequency of 5202.51MHz at 1.512v.
What do I mean by extreme water cooling? Well, we could consider anything that goes bellow 0 oC as being extreme cooling. But how do you do that with water? Good question. First of all, you have to eliminate the limitations that come from using air as the final cooling agent. You cannot do that with an AIO, but you can certainly do that with a DIY water cooling kit.
We choose some components from our friends from EK Water Blocks to demonstrate this theory. We used an EK-Supremacy Clean CSQ waterblock, an EK-DCP 4.0 PWM pump and some EK tube and fittings. You might wonder why I didn’t mention any radiators or why the pump we chose is not EK’s top of the line.
Well, it is very simple – we don’t need radiators, as we will use a different cooling agent, and we use this pump because it can handle temperatures well under 0 oC. How do we get under 0 oC with a watercooling rig, you might ask? Good question. Instead of using a radiator, we will be using a regular solid thermal bag (the ones you use when you go to a picnic to keep your beer cold) and in that we will put our thermal agent, which is a mix of water, ice and antifreeze.
The mix of water with ice and antifreeze can get down to – 10 / – 20 oC (depending on how much ice do you use) so we can talk about a first step into extreme cooling. That also means that we have to start thinking about insulation, because condensation will occur when we go down bellow 0 oC. You will need to insulate the area around the waterblock using K-Flex or art eraser (the grey Faber-Castell will do just fine) and also wrap the tubes in something that will keep the moisture away (you can use paper towels).
This is not in the comfort zone anymore, so pay attention because from this stage you can bring serious damage to your components. Also, be careful with antifreeze, that stuff is usually poisonous if ingested. On the other hand, if you are careful, the results are worth it. Take a look on the next page!
This is the moment when things start to get serious and we start with a nice 5126MHz fully stable at a voltage of 1.55v and a maximum temperature of 73 degrees C. It seems that the little Pentium scales nicely with lower temperatures.
SuperPi was stable at 5250MHz with a voltage of 1.56v, while the maximum CPU-Z frequency validation was done at 5352MHz with 1.6v.
As heat is the main problem of the Core i7 4790K, cooling it down to subzero temperatures should help a lot more than in the Pentium’s case. And it did up to a Prime stable frequency of 4847MHz at 1.4v with a maximum temperature of 83 degrees C. A very nice boost from 4653MHz obtained on low ambient air cooling.
Applying 1.517v we got a nice boost also in the SuperPi test (5271MHz) and also in the CPU-Z validation frequency (5342.4MHz).
Ok, now we get into the interesting zone – the phase change cooling area. With air or AIO you can get as low as the ambient temperature allows you, but usually now lower then 0 oC because heat pipe coolers and AIO’s will usually stop working properly under 0 degrees. With extreme water cooling (water + ice + antifreeze) you can get down to -20 oC, but the temperatures on your CPU will most likely still be in the positive area. If you want to get sub zero temperatures on your CPU you will need to use at least phase change cooling.
And luckily, we are already using this type of cooling all around us. Your fridge and your air conditioner are two of the best examples. And the principles are the same – you need a compressor charged with a liquified gas (freon in most cases), a radiator where the gas gets cool and condensates back to liquified form and a chamber / pipe / etc where the liquified gas exchanges heat with what you want to cool (air in the case of fridges and A/C units, a CPU in our case) and evaporates. The main difference in our case is that we need to concentrate all that cooling power in a small surface (a copper evaporator) that we can place on a CPU.
The phase change cooling system can have one or more stages. If we have one stage it is called a Single Stage (dooh), if it has more stages (2 or 3, where each stage is used to cool the other one) it is called a cascade. We will not go into cascade cooling today, mostly because it is very expensive, very hard to get and the only advantage compared to a Single Stage unit is that it can provide lower temperatures.
There are retail Single Stage systems on the market (DimasTech for instance) and they usually cost around 500-800 USD (for comparison a cascade goes beyond 1000 USD). Using a Single Stage we can get temperatures down to -50 / -60 oC on the evaporator’s head, so we can finally take the CPU’s temperature sub zero.
Just as with extreme watercooling, remember that we need to insulate the motherboard around the socket area so no condensation occurs while using the system. Check the next page to see what results we achieved using a DimasTech Single Stage unit.
The phase-change cooling is another step closer to the ultimate cooling that we have at our disposal today, that is liquid nitrogen. At a load temperature of 32 degrees C we could run Prime95 at 5380MHz with 1.55v.
Applying 1.65v we managed to get single threaded stability at 5593MHz for SuperPi and 5698MHz for CPU-Z validation. Getting closer and closer to the 6GHz milestone, let’s see if we get there!
We can see that this 4790K likes cold very much, because using the same 1.4v voltage as in the extreme water scenario we can run now Prime95 at 4953MHz with a temperature of maximum 63 degrees C (20 degrees C less than extreme water).
The light load scenarios are also very impressive, 5638MHz for SuperPi and 5729MHz for CPU-Z validation. We are slowly getting there…
Well it’s been all fine and dandy so far but as extreme overclockers we want more… we always want more. Enter the second step into real extreme cooling, the dry ice (or dice, as we like to call it). Dry ice is actually frozen carbon dioxide. This beauty has a very nice property called sublimation – it does not melt, it transforms direct into gas. When that happens, the temperature is around – 78.5 oC. How do you use dice to cool a CPU?
Well, first you need a container (tube, pot, evaporator, etc). It may be made from aluminum or copper, it might be taller or shorter, and there are actually a couple of retail solutions available on the market now. Imagine that back when we started doing this you had to build your own container or talk to somebody had had connections to a CNC machining place in order to get one done. Now, you lucky guys can simply search the web for an online shop, order one, pay with a credit card and just wait as the UPS guy delivers it to your door…
Benching dry ice is relatively simple – you need to insulate your motherboard, mount the CPU pot, pour in 100 ml of pure alcohol and then start adding dry ice. You need alcohol in order to facilitate the heat transfer between the dry ice and the bottom of the pot you are using and you need pure (at lest 95%) alcohol because water would freeze at that temperature. Be careful while handling dry ice and especially the first time you put dry ice in the alcohol (the first reaction will be pretty brutal). While you bench, the dry ice will sublimate so once in a while you will need to add more pellets into the pot and mix them with the remaining dry ice and alcohol from the pot.
I hope I don’t have to tell you this but from the extreme water cooling until so far we are not talking about daily use 24/7 cooling methods anymore. We are talking about overclocking sessions that can last many hours but we don’t support using extreme cooling methods for 24/7 PC usage. Also, your motherboard has to be very well insulated and… I think you realized by now, this has a certain degree of risk so if you hurt any component (or god forbid, yourself) this is totally your responsibility.
With 1.65v we can run the Pentium G3258 at 5452MHz at a load temperature of just 10 degrees C. Not a lot of scaling present when going from SS to DICE in terms of Prime stable frequencies.
With 1.75v we got SuperPi frequency of 5802MHz and a CPU-Z validation of 5957MHz, just 43MHz shy of 6GHz.
The dry ice made our Core i7 4790K to exceed 5GHz in terms of stable frequency using all threads, ending up at 5102MHz wih 1.45v and a maximum temperature of 43 degrees C (20 degrees lower than SS).
SuperPi calculated 1 million decimals at 5761MHz with 1.7v, while the maximum frequency obtained was 5923MHz at 1.72v. We can see that both CPU’s are very close in terms of SuperPi and validation frequencies, let’s see if this is also valid when using LN2.
Ok, if you got this far I am assuming that overclocking doesn’t have much more secrets to you, you are a die hard enthusiast (a responsible one of course) and you acknowledge the risks you are taking in using extreme cooling. If that is the case, welcome to the big boys world. This is what separates boys from man and casual overclockers from hard core enthusiasts. This will put hair on your chest, this will get you those top results and maybe even those world records. I am talking about liquid nitrogen (or LN2, as we like to call it) the nectar of gods when it comes to overclocking.
Liquid nitrogen is … well, liquefied nitrogen. It is used in all sorts of industries (medicine, geology, agriculture and many others) and it has a boiling point of – 196 oC. Yeah, you got that right, it boils and evaporates at – 196 oC. Of course there are many other liquefied gases that can get under – 100 oC but we use LN2 because it has the best price / performance / safety ratio from all of those. With many other liquid gases (like ammonia or oxygen) you would end up dead in a matter of seconds one way or another (poisoning, explosion, etc) while LHe (liquid helium) is much more expensive, complicated to work with and it only has a point to use it if you have a CPU without any cold bug, that can scale down to – 269 oC. And in 99.99% of the situations it is not the case.
However, remember that you should also be careful transporting, using and storing LN2, as it can also cause serious health issues like cryogenic burns if it is not used properly. This is a very serious business so you have to be very careful and responsible – if your pinky gets frozen and falls of it’s not our fault, it’s not the LN2 suppliers fault, it’s all on you. So pay attention when using it and especially keep it away from the eyes at all time.
Unlike Single Stage or Dry Ice, LN2 allows much more control over temperature (you can work at whatever temperature you like simply by dosing the amount of LN2 you pour in the pot) but it also requires your constant attention, so it is not a walk in the park like SS or Dice. Also, here is where you get into troubles like cold boot and cold bug, so it’s not as easy as it looks. But as I said, if you made it this far, you probably know what I am talking about.
Bellow you can see a retail extreme cooling pot, produced by EK Water Blocks, the EK-SF3D Inflection Point Evo, a CPU pot that has it’s own insulation system included in the mounting system. This type of CPU pot goes around for ~ 250 EUR / 300 USD.
In the case of LN2 we considered that Prime95 stability is not important, but SuperPi and maximum validation frequencies are. Please don’t mind the score, it was on Windows 7 64bit and completely not optimized as that was not the point.
The optimum voltage for the Pentium G3258 running on LN2 was 1.88v set in BIOS which translates in 1.89-1.9v measured with the multimeter. Regarding temperature, we run the CPU all the time at the limit of the coldbug which was -120 degrees C.
We managed to run SuperPi at a very nice 6192MHz, while the validation frequency stopped at 6309MHz. It is good for a randomly chosen sample, but bear in mind that you can find better or worse specimens in the wild.
The optimum voltage for the Core i7 4790K running on LN2 was 1.82v for SuperPi where we got 6159MHz and 1.95v for CPU-Z validation where we reached 6346MHz. Regarding temperature, we run the CPU all the time at the limit of the coldbug which was -120 degrees C.
We couldn’t wrap up this article without some geekish graphs and some scaling conclusions, just wouldn’t feel right. So we start with the Prime95 stability test where both CPU’s had a pretty liniar scaling with no unusual behavior. The little Pentium G3258 scaled with 12.5% from plain air to dry ice, while the “devilish” Core i7 4790K had gained 13.3%. Keep in mind that we are discussing overclocking percentage compared to the maximum clocks you can get on air, not to the stock clocks (in which case we almost overclocked the little Pentium 100% on LN2).
The frequency stable for SuperPi for both CPU were pretty similar in the whole range of scenarios and it worth mentioning that the G3258 was better than 4790K both in DICE and LN2. The scaling factor for G3258 is 22.9%, while the 4790K settled with only 21%.
In the maximum validation test the situation is opposite, with Core i7 4790K besting the anniversary Pentium in most scenarios. The scaling factor for G3258 was 22.7% while the 4790K has 22%.
Well first of all, I know – we didn’t test all cooling methods. Indeed, we only tested the most convenient ones and the best performing ones. We didn’t talk about TEC (thermoelectric cooling using the Peltier effect) because it is a thing of the past and it cannot be used very successfully on today’s CPU’s. We didn’t test a cascade because there is no actual difference between a cascade and single stage units except the temperature and loads they can keep. And we didn’t test liquid helium because it is pointless when it comes to most CPU’s (aka the ones that have cold bug at one point). And yes, there are many more ways in which you can chill water but we chose to show you the one that gets the best results.
Other than that I think we pretty much covered the whole array of cooling methods from air to LN2 and you also got to see how Core i7 4790K and Pentium G3258 react and scale with different cooling solutions. As you can see, the frequency reached with both CPU’s are pretty similar if we talk about single-threaded applications or validations. When we talk about full stability using all available cores then things change drastically because Core i7 4790K is limited by the heat. The little Pentium G3258 doesn’t have this problem because it has only 2 cores to begin with and only 3MB of cache.
Note that our samples are random, not hand picked… this means that your experience at home can be completely different. As a side note I can say that the G3258 we have is an above average sample, while the 4790K is just mediocre at best. Anyway, the Pentium G3258 is very fun to play with judging by its excellent overclock potential and also the very friendly price tag of only 72$. Also, the Core i7 4790K comes at the same price with the 4770K but with better TIM and with a healthy 500MHz frequency boost, which is quite a lot for home users.
Well, we hope you enjoyed our test and you find the results useful. If you are a beginner, hopefully we managed to shed some light on the difference between cooling methods. If you are an expert, at lest you know what to expect from some random samples. In the end, this is also a thing of luck, because there are CPU’s out there that get past 7GHz and there are those ones that cannot get past 6GHz. Even so, if you want to give overclocking a shot and want to start using air or something a bit stronger, chances are you are going to see some nice numbers both from the Pentium and the Devil’s Canyon.
Ok, I hope you enjoyed your reading today. If you got here and you liked the article, we have two bonus pages for you. The first bonus is … the beer cooled PC!!! Using the same gear as we did in our extreme water cooling section (EK-Supremacy Clean CSQ + EK-DCP 4.0 PWM) we replaced the water and antifreeze with chilled beer (7.5L of beer and 5KG of ice) and we took the whole system on the romanian national television (ProTV). Live, during the morning news we easily managed to get to 5.2GHz. We could have gone to 5.4-5.5GHz but we didn’t want to get any blue screen on live TV so we didn’t push our beer cooled rig too much.
And you should also know that we used a very “special” kind of beer, which can be purchased for the extraordinary price of 1.66 USD / 2.5L… So….yeah… beer is good for cooling you, but it can also be used to cool your CPU… And instead of a potato, here is a 2.5L beer !!!
Did you like the results we got using the extreme water cooling methodology? Or maybe our beer cooled rig? Do you want to try this at home? Well, you could be one of the lucky winners that do!!! Our friends at EK Water Blocks are offering one EK-Supremacy Clean CSQ – Acetal for one lucky LAB501 reader !!!
Again, this giveaway is opened for international readers, so it does not matter if you are from USA, Russia, Japan, Germany, Indonesia or Romania – if you enter this giveaway you can be the lucky winner of a cool EKWB waterblock, just like the one we used in our review!!! It works with water…or chilled water… and even beer!!!
The request is very simple. Please comment on this article by July 21, 23:59 Romanian time (16:59 New York Time, 21:59 UK time) answering these two very simple questions:
1 – At what frequency did the Pentium G3258 ran fully stable using extreme water cooling?
2 – What maximum frequency did we achieve on Intel Core i7 4790K using LN2?
-The contestants must post a reply for this article with their answers (see the “Leave a reply” box bellow)
-The comments will be approved after the contest ends, so no nobody can take a peak at the right answers from the others participants
-You must fill in a valid email address when you leave your comment (in the “email” box) – this is how we will contact you and send you the prize in case you win
-The contest ends on July 21, 16:59 New York Time – 21:59 UK time – 23:59 Romanian time
-The contest is open to all readers from all over the world, except LAB501 and EKWB employees and relatives
-The winners will be designated by using Random.org
-The winners will be announced on July 23, 2014
-You can only submit one entry for the competition (we will be checking the emails and IP addresses)
-The prize will be shipped by EK Water Blocks directly to the winner
Good luck !!!