Slashdot Mirror
Ultimate Cooling System
OCGeek writes "This should be interesting for the overclockers as
VR-Zone has an article up on building a
cascade cooling system that
cools chips down to -110C. The guide shows you the components that are required
for the cascade cooling system such as the compressors, condensers,
refrigerants, evaporators, heat exchangers, oil separators etc. and the tools
you would need. It allows hot chip like Prescott to reach over 5.1Ghz and ATi
Radeon 9800 XT card to reach over 660Mhz core."
1. Because it's possible
2. It's kinda cool (literally0
3. It keeps overclockers off the streets
4. It gives us something to do
5. It's just interesting
6. Performance!
Wouldn't it be cheaper and easier to just buy a multi-processor box, rather than invest in all this gear to make one cpu run even twice as fast?
I mean, there's only so far Hyperthreading will take you...
The party of stupid and the party of evil get together and do something both stupid and evil, then call it bipartisan.
and cost less to boot...at least in time and energy consumption.
01100111 01100101 01110100 00100000 01101111 01110101 01110100 00100000 01101101 01101111 01110010 01100101 00101110
In a controlled situation, you wouldn't have any problems with condensation. I imagine when they turn the coolers off, they would want to bring the temperature back up to room temperature via a controlled sequence. You will get condensation if you go from that cold to warm rather quickly.
--
Real-time deal updates from all the major deal sites.
This seems a little complex and extreme for the home builder. Maybe a specialty co-lo opportunity, though? "Icebox netbox"? No good for gamers, of course. But for others who need MIPS for problems that can't be parallelized...
Everybody's a libertarian 'till their neighbour's becomes a crack house.
I've always wondered this but nobody's ever given a satisfactory answer to "why not".
Why not overclock network cards as well as CPU and graphics cards?
think about it
If I can get 10mg from a normal network card and overclock it for say 15 even if I need shorter cables, that's only shorter than maximum isn't it? So instead of 30ft cables I might be limited to 20ft. Big deal in a home network, NOT. I could overclock some more of my machines and have them all going at 15mg, and get better network speeds. I'm surprised there have not been articles about this. Or what about modems? or monitors? or even sound cards to get higher frequencies and better bass from them?
Why stop with just a cpu when it's not even a slowest part of a system but hard drives might be more difficult
I remember reading about something similar to this using florinert and liquid nitrogen. It must be nice to have more money than sense, because for the amount of money they spend on the cooling equipment, they could have bought "bleeding edge" hardware. But I am not an overclocker either.
http://www.google.com/search?q=cache:M3MveYmm8lQJ: www.vr-zone.com/%3Fi%3D618%26p%3D1++site:www.vr-zo ne.com+cascade&hl=de&ie=UTF-8
Zero degrees isnt enough. The lower the temprature, the easier electrons move and the faster gates switch. If you were to try to get a prescott to run at 5.5ghz normaly, it would result in errors as the gates wouldnt switch fast enough to keep up with the clock. With this level of cooling, it's no longer about heat concerns, but the speed of the logic gates.
When life gives you crap, Make Crapade.
Sluggy Freelance.
Sounds suspiciously like they stole the technology from Michael's computers...
11 comments and the site is down. guess that cooler failed the test!
I have done something similar but I used a fridge cooling coil on my cooling mount on the processor, I gained up to 4ghz on my rig... was pretty fun
although, the computer isn't really mobile having a mini-fridge's heating coils/compressor outside of the case!
Too much cold can be just as bad as too much heat..
Mods: please mod as +1, Insightful, or +1, Funny.
Resistance is fu^H^H inversely proportional to temperature. Metal on silicon is just the same as any conductor in this respect. Perhaps a real physicist can fill in the gaps as to the relationship between frequency (and consequent electron flux density) and resistance. So there is an absolute corelation between temperature and frequency as far as I know. This is a separate issue from the thermodynamics of the system, how heat is conducted away. Can't read the article, but I assume its old news since Cryotek were making 1GHz systems almost 6 months before AMD and Intel officially released 1G uPs using referigerant tech. btw, I have always wondered why systems don't use peltier effect heatsinks more for this.
A system that cools to -110C does really seem "ultimate" to me since absolute zero is -273C.
Because that system is toast. And smoking. And there's a puddle of liquified metal under the rack.
Anyone got another link?
should have started!
Write it out 100 times, on my desk in the morning.
"Why not overclock network cards as well as CPU and graphics cards?"'
Hew! Why not sound cards, so that only our dogs can hear them?
Or our TV cards so that we can watch TV faster.
Or our mice so that darn cat's unable to catch it.
There are some things that overclocking will really do nothing for, and just increase cost and complexity.
This reminded me of another extreme hobby, BBQ lighting by George Goble, who also happens to be a systems engineer (go figure).
To do something just to show that it can be done is one thing, but I don't think anyone should seriously consider doing this for any other reason than to merely say "I did it".
And for those asking "why so cold", I can answer that one, it has to do with total thermal inertia, and thermal gradients. Basically, the larger the temprature difference, the faster heat will attempt to move from hot to cold. Just because the outside of the chip may be 0 degrees, doesn't mean the inside of the chip is.
The cesspool just got a check and balance.
This is the ultimate cooling system:
1. Walk up to Tyra Banks.
2. Drop "How would you like to be a model?" on her.
3. Seek shelter.
These techniques seem like brute force schemes to deal with the thermal resistance of chip packages -- you have to cool the heatsink to -110C in order to keep the "intel inside" at less than +60C). Why not use backside thinning. to bring the hot circuits of the processor within microns of a high coolant flux chamber. Backside thinning could get the coolant to within 10 microns of the junctions. If the CCD people can thin a massive 2k x 2k CCDs (the die is bigger than 1" square), I'm sure an enterprising overclocker could thin a Pentium.
Two wrongs don't make a right, but three lefts do.
Hey, that sounds like more fun than reading the crap here! Wait for me!
Won't work in the summer, but you'll be too busy trying to scrape mosquitos out of your cooling fan to care.
it's -273.16 C
I want to delete my account but Slashdot doesn't allow it.
Here's my 2 cents worth..
I am pretty sure that silicon becomes more like a metal at higher tempretures (conductivity increases) and becomes more like an insulator at lower tempretures.
So... Lowering the tempreture too much turns the silicon into a brick.
What cooling would help is to dissapate the waste heat caused by the circuits shorting the power to ground - in an ideal world, the switching of the gates would be perfect and the processors would consume very little power but it is currently the state that every time a logic gate switches state, there is a moment where the supply power is shorted to the ground.
When the gates get hotter, they get more conductive and more power gets shorted
So these cooling systems are useful at getting rid of the waste heat (which are incidentially a side effect) The cooling of the chips probably slows down the switching time of the gates as the semiconductor becomes more of an insulator.
Solutions
If it is possible to fabricate more perfect switching pair of FETs, the power consumption and heat output would go down. Of course, this would be the Holy Grail but probably won't be realisable until we can construct each gate atom-by-atom...
Another idea is to fabricate processors from a different semiconductor whose operating range is at a higher tempreture so the chip is harder to burn-out and possibly reach higher switching speeds... but the problem is so much time/money is invested on Silicon, that there is little interest in researching the use of other semiconductors.
But for the overclockers... if they can regulate the tempreture of the silicon to be nice and warm (so the switching speeds are good) but are able to move enough watts to keep the temp from running away...
-- The universe began. Life started on a billion worlds...
-- Except on one where stupidity was there first.
A 4 inches hard dick are'nt good. Push your errection potential to limit OverCock it to 12 inches then.
Isn't it a bit dangerous to cool a chip to -120?
that's why i like heat sinks. they can only fail if you fuck up their installation. or if the fan fails. or if the power to the fan fails... hmm...
in this age of communication i'm just not getting through
Probably because Peltier Effect devices are very expensive. Though I wonder how well you could do masking on the peltier devices in the chip itself, to pull heat to the chip's upper surface.
hotter, just look to the cool relief of Preparation-H
to get you on your way.
Preparation-H, the ultimate cooling system.
Apart from the Quake series and the likes of Photoshop or 3D apps I doubt that an extra CPU would help much. Unless you play games on a machine that runs large background processes - the accounts dept's server or such...
I'm sorry if I haven't offended anyone
Their server seems to encountered a resonance cascade.
www.akiba-pc.com
its working
It's not that cooling makes the silicon faster. In fact, some circuits (analog bits like PLLs in particular) perform worse at temperatures outside their design target. The reason that cooling works is that hot silicon stops working. Inter-junction thermal voltages (same as in a thermocouple) increase as the temperature goes up. This adds noise to signals and screws the threshhold voltage of the gates on the silicon. This is also why overvolting a CPU can sometimes help: Increasing the supply voltage increases the S/N ratio of driven signals, up to a point.
omg terrorists have infiltrated philly city government.
out with yer well organised militia, people!
Now this is the ultimate cooling system... =)
The last part of the video (the flower thing) is even scary!
The ./ article and the
site
I'm sorry if I haven't offended anyone
Aside from the cost of peltier cooling, my past research (I was interested in this, too) has indicated that they get cold too fast, creating condenstation on the chip before the CPU gets hot enough. Then, of course, you have to devise a way to get the baking heat from the other end of the peltier out of the case....
I have seen a couple instances of people making this work. One involved using a lot of rubber sealing compound and essentially making an airtight seal around the CPU socket and the CPU itself. Another involved a seperate unit with a power regulator to solve this 'instant condense' problem. It worked very poorly, though.
Now... if you could create a better delay-on control for the peltier unit, that might work. I also had an interesting concept of mounting a heatsink to the CPU, putting the peltier square on the cooling fins of that heatsink, then attaching another heatsink to the hot side of the peltier, and a cooling fan for that, with maybe some kind of ducting pipe for direct ventilation of the heated air outside of the case. The peltier would have to cool the CPU's heatsink first before forming condensation on the CPU itself, and I imagine that would be enough of a delay for the CPU to get warmed up. But that'd be quite a tower stacked atop a processor, and I wouldn't want to discover a fallen Tower of Pisa inside of my case....
What's gonna happen when the cooling fails? Oh, believe me, it will happen sooner or later. Next thing you know your house is on fire?
Crack
How much is your data worth? Back it up now.
The only way to cool is not to get hot.
Go C3.
I LOVED THAT GAME.
Had an old Permedia 2 that barley ran it but damn. I loved that game. Stereo effect was AWESOME.
How much is your data worth? Back it up now.
While I don't work at a microprocessor company, I do work on the physical implementation of mixed-signal ASICs and I'm surprised these CPUs can work at -110C. As I recall even military limits only go down to -50C (at the maximum allowable voltage, usually no more than +10% of nominal) for design timing closure; beyond this (higher voltage and/or lower temperatures) the flip-flop to flip-flop paths may get fast enough to result in a "hold-time violation" . This is when the signal from one flip-flop reaches a downstream flip-flop so quickly that it is registered one clock-cycle early (basically, it is captured on the same clock edge as it was launched). This is most critical on timing paths with no combinational logic (occurs often in shift registers and cross-clock domain synchronizers) and is further complicated by clock distribution networks that take advantage of "useful skew" to borrow time from one timing path for use on another. I'd be surprised if even CPUs were designed with enough hold-time margin built-in to handle -110C.
The other variable is the fabrication process corner, so assuming the CPU isn't on the edge of being "fast" there could be some hold-time margin on a given chip to allow this kind of cooling to result in a working processor. Still, I'm kinda surprised it works at that temperature with any reliability.
- Leo
You don't use science to show that you're right, you use science to become right.
Ultimate Cooling System ???
Yeah right. There's no way this thing can top a quality speed-demon from Michael's Computers.
oh baby that's the stuff...
I posted this at "2004-03-20 14:04:14", I usually don't whine about it but as it is the second time in a row in a short time =).
When the cost of cooling a system is more than adding another processor, you have to remember the saying: "cascade cooling systems make your cpu cool, not you!"
This comment does not represent the views or opinions of the user.
Plus -110C would be great for making ice cream!
I was a sysop for a BBS back in the dark ages before Internet, and one of users once asked me if it was possible to overclock a modem to get higher speeds. I promptly answered: "Do you have an external modem? Good, just replace your current transformer with something that gives you more volts for your modem." He thanked for advice and logged off.
He never called back.
Why yes, I do like reading BOFH stories, why do you ask?
Follow your Euro bills at EBT
and it doesn't need to be a large background process either...there's enough stuff running in a modern operating system that a second processor is an enourmous benefit. It won't make a single threaded game go faster...but it will help stop it going slower.
... but got fed up of having to go to sleep and get up every four hours.
> I have always wondered why systems don't use peltier effect heatsinks more for this.
Peltier's are about 5% efficient... Compressors are about 50% efficient... You get a lot more cooling per watt with a vapour-change compressor system.
The geek will run out of oxygen in his basement.
Interconnect resistance increases by over 20% when you go from 25C to 100C.
I've heard that if a chip gets cooled too much... it's starts superconducting... is this true?
||| I still can't believe Parkay's not butter.
Go back to GBS
Ill make an off-the-cuff assertion that modern hardware is already exceptionally well engineered, and that modern software is bloated crap.
For most tasks, I reckon that if the effort was applied to profile, hand optimise and re-factor the underlying software, much greater gains can be acheived than by ever dicking with the hardware.
Its like if you have a car which is not going so well, because the handbrake is stuck on, the oil is crudded up into lumps in the engine, and there is 100kg of dirt and mire under the floorpan to carry around - solution - whack a supercharger on the engine and make it faster.
Better idea: get one card to generate image for the left eye, the other one for the right eye, and feed the head motion sensors right into the cards so the position update is couple milliseconds faster. We're finally getting the machines good enough for VR while still being affordable. Just hope the displays will get cheaper soon.
If I remember correctly, it's rather that the electrons form Cooper pairs and they almost stop interacting with the crystal lattice of the material. There are materials - notably the YBaCuO ceramics - that get superconductive at the liquid nitrogen temperature, which is quite hot - enough to keep the lattice vibrating like mad. I forgot a lot about it, though...
You can only pump the heat around. It usually ends up in the heatsinks at the back of the cooling machines, from where it dissipates into the lab, rising its temperature slightly. Like a common kitchen refrigerator.
Or you could somehow use the on-chip diode or the temperature sensing diode mounted on the motherboard under the CPU.
Never tried it, it's just a theory, but it should work. :)
You know, overclocking two cpus in an SMP situation might be neat.
I don't know a whole lot about SMP architectural hardware requirements, but it'd be great to read about it.
tasks(723) drafts(105) languages(484) examples(29106)
Well, use something like HyperTransport over a short distance for a couple(or more) PCI cards.
What...is PCI a dirty word now?
Configure as many cards as you like to monitor the graphics commands on the PCI bus. Then have the cards dump rendered video back over the HT link to the master card. The master card would be decided automatically, since it would be the only card with a display plugged in.
Also, instead of mirroring data, you could have the cards pool their memory. That would be especially advantageous if you wanted to link four or five cards together, but wanted to run two displays, since each display could, with optimised drivers, primarily work with the data closest to it, yet be able to take memory from other cards as needed, you wind up with a synergistic (ack!) relationship.
While you technically won't multiply your performance by the number of cards you have, you'll still do better than the average card.
Finally, once PCI Express becomes standard (in, oh, I'd guess three years), AGP won't have an advantage over PCI, anyway.
tasks(723) drafts(105) languages(484) examples(29106)