- Joined
- Oct 27, 2003
- Location
- UK (Manchester)
I found this in another forum...thought id share it with you 
So far the best overclocks I've seen using extreme types of cooling are:
LN2 ~ -196°C
P4 3.0C (SL6WK, D1 stepping) --------------- @ 4762MHz w/ 2.08v
XP3000+ (Barton) ------------------------------- @ 3585MHz w/ 2.27v
XP2100+ (T-Bred 'B', AIUHB 0301) ----------- @ 3382MHz w/ 2.15v
Dry ice ~ -78°C
P4 3.06 (SL6S5, C1 Stepping) ----------------- @ 4456MHz w/ 2.04v
XP3200+ (Barton) ------------------------------- @ 3173MHz w/ 2.20v
XP1700+ (T-Bred 'B' JIUHB 0307VPMW) ----- @ 3016MHz w/ 2.10v
Phase-change ~ -30°C to -50°C
P4 3.0C (SL6WK, D1 stepping) ---------------- @ 4474MHz w/ 1.775v using a Prometeia w/ R404a mod
XP2100+ (T-Bred 'B', AUIHB 0302 TP2WA) -- @ 3069MHz w/ 2.30v using a Prometeia w/ R404a mod
XP3200+ (Barton) -------------------------------- @ 3058MHz w/ 2.20v using a Prometeia
links to some of these rigs are...
http://holicho.lib.net/bench/h_clock.htm
http://www.solidhardware.com/macci/nwln2/6s.jpg
http://www.muropaketti.com/artikkelit/cpu/northwood2200/ln2/3.jpg
http://www.muropaketti.com/artikkelit/cpu/nw2400_133/6.jpg
Plus some general info on the substances used
Current electronics are based on a common manufacturing process that use CMOS transistors. Supposedly experiments have shown for every 10oC drop in operating temperature of CMOS transistors, a 1-3% performance gain is achieved. Since going from +60oC to -40oC would be a 100oC difference, that could result in a 10-30% performance gain. Which in term, allows components made from CMOS transistors to run faster than normally spec'd for. Which is why supercooling allows for processors to overclock so much. But as you can see from the figures, it's a case of deminishing returns, so need big drops in temp to achieve reasonable gains over the norm.
Liquid Helium (LH) has a boiling point of -268.93°C (absolute zero in space is -273.15°C), so is way cooler than LN2 and not far off absolute zero either. Think a few people are trying to build as LH system at the moment. But other than the increase cost of LH over LN2 (LH is ~$15/Litre?), there are a few issues that have concerned people about using it. It has the same drawbacks of such extremely low temps as LN2 cooling does, but more so. So might explain why no-one has really tried LH cooling much. Also some claim LN2 has better heat capacity than LH, so you'd need more of LH to cool the same heatload than LN2. Supposedly a LH generator at a physics dept in a US college uses a modified Stirling engine (this?), which is cascade-cooled by LN2 (lower temps, lower pressures), to liquify the helium.
LN2 (liquid nitrogen) cooling is probably the most well known extreme type of liquid cooling with a boiling point of -196°C. However, it's not a practical cooling solution as you have to refill the LN2 every few minutes or so. What it will do is probably get every last MHz out of a processor for a short period of time, just long enough to get benchmarking results and screenshots. Because of the extreme low temps, many LN2 users pre-cool the processor with dry ice and such to gradually lower the temp before adding LN2 and also well insulate the system. So to reduce the risk damaging the processor and motherboard with the sudden drop in temp. Supposedly, with the P4 at least, once the base of the cooler hits -145°C to -135°C, the CPU stops operating properly (works fine again when the temp rises though). I have heard that super-low temps can cause issues with L2 cache.
People have also used LN2 in submersive cooling. Normally you submerge the whole system (motherboard, PSU, etc) into a container of non-conductive liquid (normally a kind of oil), then cool the liquid (via a watterchiller?), which in turn cools the whole system. Done properly and dried out, the system can remain functional afterwards. Some have even used LN2 cooled Fluorinert as the liquid in such submerged cooling systems. However Fluorinert is expensive. The Cray Supercomputers did use this method of cooling though for some of their commercial supercomputers. While the Cray 1 was watercooled, the Cray 2 was 'flouronics' cooled and the whole computer was immersed in the electrically insulated fluid made by 3M. TechTV have done an article or two about such cooling (see here and here). They used used 3M hydrofluoroether (HFE-7100) to submerge cool their system, however, it's expensive at $220 a gallon.
Dry Ice (frozen carbon dioxide, CO2) cooling is a little less extreme at only -78.5oC and maybe more accessable, but still extreme never the less. Dry Ice turns to gas as it melts rather than a liquid. Like with LN2, you'll need to keep topping up the dry ice as it basically boils off and evaporates into a gas. Therefore it is also only a short term cooling method. Though you need to combine dry ice with a liquid in the container to achieve proper cooling. Because a layer of CO2 gas forms around the dry ICE and kind of insulates it from the container. So basically the liquid allows for better heat transfer between the dry ice and container. Since the liquid needs to stay liquid at -78oC, people have used 'Ethanol' (or near pure ethanol, like 'Everclear'). Someone also suggested 'Propanol' but not sure. Don't think windshield wiper fluid (66/33% mix of methanol and water) is a good idea due to the fumes 'Methanol' supposedly gives off being nasty.
There are a number of companies that probably can supply Dry Ice, one in the UK is Global ICE in Greenford, Middlesex. Someone said they charge £35.00 inc VAT for a 10 kg bag of dry ice, which includes free delivery within London. Outside London they charge an extra £15. Supposed the bag of dry ice will lose up to 2kg every 24hrs. Someone else mentioned a company in Boston, UK which sells a 10kg of dry ice for £20. You can also get dry ice maker bags, that you fit onto a CO2 gas tank (like used in paintball), to make small amounts of dry ice.
Liquid CO2 (carbon dioxide) has a boiling point of –56.6°C, so is a little less extreme than Dry Ice. One user has taken a tank of Liquid CO2 and pumped it into a modified block (waterblock?) via a copper tube (steal too fragile at such low temps?). Supposedly a small silicon pipe is used to connect the copper tube to the block and might act as a safety valve (expand and come off if pressure is too high). It then boils, cooling the block that is attached to the processor. The block has vents or a vent tube that allow the produced CO2 gas to escape, allowing more Liquid CO2 to flow into the block, and so on. The CO2 in the tank needs to be 80 ATM pressure to achieve liquid at normal room temps. Also the tanks needs a bottom draught, so you get liquid CO2 not gas to begin with. Supposedly, the useful thing is, when you open the tank and let the CO2 flow into the block, the CO2 liquid turns to a solid (dry ice) due to the pressure change from 80 ATM to normal pessure, then it turns to gas. So it's possible this method, might achieve cooling similar to Dry Ice. This might explain why they measured -65°C, even though Liquid CO2 is only meant to be good for –56.6°C.
However, others have had problems with the dry ice blocking the flow of CO2 liquid in the block, when something called the 'triple point' of CO2 is reached (60.428psig @ -69.83oF). It has been suggested that using a pressure regulator valve on the block outlet might solve this by keeping the block pressure from dropping below the triple point pressure (~65psi). But still concerns of dry ice forming at the outlet of the regulating valve, so it might have to be heated somehow (heating cable or element?).
So far the best overclocks I've seen using extreme types of cooling are:
LN2 ~ -196°C
P4 3.0C (SL6WK, D1 stepping) --------------- @ 4762MHz w/ 2.08v
XP3000+ (Barton) ------------------------------- @ 3585MHz w/ 2.27v
XP2100+ (T-Bred 'B', AIUHB 0301) ----------- @ 3382MHz w/ 2.15v
Dry ice ~ -78°C
P4 3.06 (SL6S5, C1 Stepping) ----------------- @ 4456MHz w/ 2.04v
XP3200+ (Barton) ------------------------------- @ 3173MHz w/ 2.20v
XP1700+ (T-Bred 'B' JIUHB 0307VPMW) ----- @ 3016MHz w/ 2.10v
Phase-change ~ -30°C to -50°C
P4 3.0C (SL6WK, D1 stepping) ---------------- @ 4474MHz w/ 1.775v using a Prometeia w/ R404a mod
XP2100+ (T-Bred 'B', AUIHB 0302 TP2WA) -- @ 3069MHz w/ 2.30v using a Prometeia w/ R404a mod
XP3200+ (Barton) -------------------------------- @ 3058MHz w/ 2.20v using a Prometeia
links to some of these rigs are...
http://holicho.lib.net/bench/h_clock.htm
http://www.solidhardware.com/macci/nwln2/6s.jpg
http://www.muropaketti.com/artikkelit/cpu/northwood2200/ln2/3.jpg
http://www.muropaketti.com/artikkelit/cpu/nw2400_133/6.jpg
Plus some general info on the substances used
Current electronics are based on a common manufacturing process that use CMOS transistors. Supposedly experiments have shown for every 10oC drop in operating temperature of CMOS transistors, a 1-3% performance gain is achieved. Since going from +60oC to -40oC would be a 100oC difference, that could result in a 10-30% performance gain. Which in term, allows components made from CMOS transistors to run faster than normally spec'd for. Which is why supercooling allows for processors to overclock so much. But as you can see from the figures, it's a case of deminishing returns, so need big drops in temp to achieve reasonable gains over the norm.
Liquid Helium (LH) has a boiling point of -268.93°C (absolute zero in space is -273.15°C), so is way cooler than LN2 and not far off absolute zero either. Think a few people are trying to build as LH system at the moment. But other than the increase cost of LH over LN2 (LH is ~$15/Litre?), there are a few issues that have concerned people about using it. It has the same drawbacks of such extremely low temps as LN2 cooling does, but more so. So might explain why no-one has really tried LH cooling much. Also some claim LN2 has better heat capacity than LH, so you'd need more of LH to cool the same heatload than LN2. Supposedly a LH generator at a physics dept in a US college uses a modified Stirling engine (this?), which is cascade-cooled by LN2 (lower temps, lower pressures), to liquify the helium.
LN2 (liquid nitrogen) cooling is probably the most well known extreme type of liquid cooling with a boiling point of -196°C. However, it's not a practical cooling solution as you have to refill the LN2 every few minutes or so. What it will do is probably get every last MHz out of a processor for a short period of time, just long enough to get benchmarking results and screenshots. Because of the extreme low temps, many LN2 users pre-cool the processor with dry ice and such to gradually lower the temp before adding LN2 and also well insulate the system. So to reduce the risk damaging the processor and motherboard with the sudden drop in temp. Supposedly, with the P4 at least, once the base of the cooler hits -145°C to -135°C, the CPU stops operating properly (works fine again when the temp rises though). I have heard that super-low temps can cause issues with L2 cache.
People have also used LN2 in submersive cooling. Normally you submerge the whole system (motherboard, PSU, etc) into a container of non-conductive liquid (normally a kind of oil), then cool the liquid (via a watterchiller?), which in turn cools the whole system. Done properly and dried out, the system can remain functional afterwards. Some have even used LN2 cooled Fluorinert as the liquid in such submerged cooling systems. However Fluorinert is expensive. The Cray Supercomputers did use this method of cooling though for some of their commercial supercomputers. While the Cray 1 was watercooled, the Cray 2 was 'flouronics' cooled and the whole computer was immersed in the electrically insulated fluid made by 3M. TechTV have done an article or two about such cooling (see here and here). They used used 3M hydrofluoroether (HFE-7100) to submerge cool their system, however, it's expensive at $220 a gallon.
Dry Ice (frozen carbon dioxide, CO2) cooling is a little less extreme at only -78.5oC and maybe more accessable, but still extreme never the less. Dry Ice turns to gas as it melts rather than a liquid. Like with LN2, you'll need to keep topping up the dry ice as it basically boils off and evaporates into a gas. Therefore it is also only a short term cooling method. Though you need to combine dry ice with a liquid in the container to achieve proper cooling. Because a layer of CO2 gas forms around the dry ICE and kind of insulates it from the container. So basically the liquid allows for better heat transfer between the dry ice and container. Since the liquid needs to stay liquid at -78oC, people have used 'Ethanol' (or near pure ethanol, like 'Everclear'). Someone also suggested 'Propanol' but not sure. Don't think windshield wiper fluid (66/33% mix of methanol and water) is a good idea due to the fumes 'Methanol' supposedly gives off being nasty.
There are a number of companies that probably can supply Dry Ice, one in the UK is Global ICE in Greenford, Middlesex. Someone said they charge £35.00 inc VAT for a 10 kg bag of dry ice, which includes free delivery within London. Outside London they charge an extra £15. Supposed the bag of dry ice will lose up to 2kg every 24hrs. Someone else mentioned a company in Boston, UK which sells a 10kg of dry ice for £20. You can also get dry ice maker bags, that you fit onto a CO2 gas tank (like used in paintball), to make small amounts of dry ice.
Liquid CO2 (carbon dioxide) has a boiling point of –56.6°C, so is a little less extreme than Dry Ice. One user has taken a tank of Liquid CO2 and pumped it into a modified block (waterblock?) via a copper tube (steal too fragile at such low temps?). Supposedly a small silicon pipe is used to connect the copper tube to the block and might act as a safety valve (expand and come off if pressure is too high). It then boils, cooling the block that is attached to the processor. The block has vents or a vent tube that allow the produced CO2 gas to escape, allowing more Liquid CO2 to flow into the block, and so on. The CO2 in the tank needs to be 80 ATM pressure to achieve liquid at normal room temps. Also the tanks needs a bottom draught, so you get liquid CO2 not gas to begin with. Supposedly, the useful thing is, when you open the tank and let the CO2 flow into the block, the CO2 liquid turns to a solid (dry ice) due to the pressure change from 80 ATM to normal pessure, then it turns to gas. So it's possible this method, might achieve cooling similar to Dry Ice. This might explain why they measured -65°C, even though Liquid CO2 is only meant to be good for –56.6°C.
However, others have had problems with the dry ice blocking the flow of CO2 liquid in the block, when something called the 'triple point' of CO2 is reached (60.428psig @ -69.83oF). It has been suggested that using a pressure regulator valve on the block outlet might solve this by keeping the block pressure from dropping below the triple point pressure (~65psi). But still concerns of dry ice forming at the outlet of the regulating valve, so it might have to be heated somehow (heating cable or element?).
