Do temps under 0*C really affect the CPU?

[MaJik]

once your under 0*C does having it -20* or even like -40* or -60*C really allow any further overclock? how cold can it run before it blows up ?
-100? -200?

 

[jinsonxu]

It'll help in overclocking since CMOS runs faster at low temperatures. Should see pm's opinion first though.

 

[Archon777]

Well, CMOS doesn't run faster, just more efficiently. You need that extra efficiency for big overclocks (such as 200Mhz over the default speed.) Depending on the percentage of your overclock, you can determine the percentage increase in efficience you need. For an 800E, say, you need about 15%-20% more efficiency to overclock, and that means you need to reach about -20C. Oh, and the cpu won't "blow up" when it gets extremely cold. At -120, current CMOS's double in efficiency, allowing two-fold and greater increases in speed. At -200, you could probably triple your cpu speed. (Old Cray supercomputers were just big procesors cooled to more than -250 degrees C).

 

[pm]

JinsonXu, I only got your PM now... I bought something from an Anandtech member and was checking for a reply late last night, but I just popped on and off.


Yes, transistor performance continues to improve with colder temperatures down to an absolute limit (which I can't remember, but it's really low). Whether you want to call the improvement in transistor current drive as being more efficient or simply faster (it's both), the end result is a device in which signals reach their destinations sooner at colder temperature, and thus calculations happen faster, so the device can be clocked at a higher frequency without introducing max-time errors.

There's a big concern regarding rapid contraction of the materials and the stress this places on the BGA/FC bumps that connect the processor to the package. Due to variances in thermal properties, the coefficients of expansion and contraction can vary between the materials, so you end up with one material rapidly contracting in the presence of a sharp drop in temperature, while the material on the other side of the bump grid contracts more slowly. This can result in shearing off of the bump pads in the worst case (dead chip), and increased resistance due to metal fatigue in the best case. This can be avoided by not introducing sharp temperature gradients.

Second, the limit on temperature tends to be a design limit rather than a physical limit. At some point, the transistors will switch faster than the clock can (since the clock is limited by a huge metal network and this will switch more slowly at low temps compared to the transistors) and you end up with signals literally flying through the circuitry and arriving before the latches close because the signas have "raced" the clock and won (which is bad). This is called a "hold time failure" or, more commonly in the industry, a race condition. Designers are pretty good about fixing all races down to a certain limit but beyond that you are increasing design time (because races can be a big pain to fix) to achieve race-free operation at conditions that most people don't care about. If you start using cryogenic cooling then you will probably run into this at some point.


Standard disclaimer: I am not encouraging you to run your CPU's at cryogenic temperatures - I'm just answering the question. You can do what you want with your CPU that you bought, but (a.) be careful, and (b.) don't say that I didn't warn you if your CPU stops working.