Slashdot Mirror
Is Overclocking Over?
MrSeb writes "Earlier this week, an ExtremeTech writer received a press release from a Romanian overclocking team that smashed a few overclocking records, including pushing Kingston's HyperX DDR3 memory to an incredible 3600MHz (at CL10). The Lab501 team did this, and their other record breakers, with the aid of liquid nitrogen which cooled the RAM down to a frosty -196C. That certainly qualifies as extreme, but is it news? Ten years ago, overclocking memory involved a certain amount of investigation, research, and risk, but in these days of super-fast RAM and manufacturer's warranties it seems a less intoxicating prospect. As it becomes increasingly difficult to justify what a person should overclock for, has the enthusiast passion for overclocking cooled off?"
For me, It's fun and I could care less what some dude did with liquid nitrogen.
First computer, I just used Asus Overclock and felt I got more for my money.
Second computer, I started fiddling with manual settings.
Third computer I pushed it until I couldn't get rid of the heat with air cooling.
Fourth and current computer, water cooled and running awesome (6 cores at 4.3 GHz).
Each time I felt the progress, it's like leveling your character, but the character is you, and the game is real life!
Look, digital electronics are still subject to analog limitations. When you overclock, you squeeze the hysterisis curve, increasing the probability that your chip incorrectly interprets its the state of a particular bit as the opposite value. i.e. you get random data corruption. This is why you eventually start crashing randomly the more you overclock.
While overclocking a chip that has been conservatively binned simply to reduce manufacturing costs but is actually stable at higher clock rates is reasonable, trying to overclock past the design limits is pretty insane if you care at all about the data integrity. Also, you tend to burn out the electronics earlier than their expected life due to increased heat stress.
I never overclock.
That's like saying competitive soccer going broke would impact on EVERYONE EVER from playing soccer with their friends.
Not everyone overclocks to beat a record.
Hell, "overclock" a toaster if you have to. 2 second cold toast anyone? (the best toast)
But really, there are still plenty of things you can overclock to beat records, such as what iB1 mentioned up there, overclock a smartphone or tablet.
Overclock a Beagleboard, or a Raspberry Pi when it comes out, Arduinos. All these compact computers are pretty much sitting around waiting to be hit by the overclocking spirit.
Your logic is wrong.
Every time a FET switches, it requires a certain number of electrons to move to or from the gate to create an electric field in the substrate to open or close a conducting pathway. This is a current flowing through a reistance and it dissipates power as heat. Assuming that the leakage current on the gate is very small compared to the switching current, the energy required to switch the FET (call it Es) is constant regardless of the clock speed. So the power dissipated by each FET (call it Pf) is:
Pf = Es x fc
where fc is the clock frequency in Hertz.
Why do you suppose that frequency scaling is an effective way of saving power?
Slashdot - News for Nerds, Stuff that Matters, in ISO-8859-1 Has just realised that beta makes this signature redundant
I'm sorry that you have barely passed EE 101 a week ago. The information you have is very inaccurate and quite outdated.
In modern CMOS geometries, a large amount of power is wasted on leakage. That means that while the dynamic power scales linearly with frequency (at a constant voltage), the static power (leakage) does not.
However, if you *can* overclock significantly at a constant voltage, there probably is power headroom the manufacturer did not use properly, or expected the devices to be unreliable with reduced voltage at the original frequency. Dynamic voltage scaling is not new.
Actually, voltage matters substantially.
The gate of a FET is effectively a capacitor. Even with the FET in the on state, if you keep increasing the gate voltage it'll still keep taking electrons. And like a capacitor, energy stored in a FET gate = 1/2*C*V^2. You also have source/drain and gate/drain (miller) capacitance - source/drain has to be discharged (another 1/2CV^2 loss) and the miller capacitance has to be discharged and then charged at the opposite polarity (a CV^2 loss).
Overall, neglecting leakage current, power loss is proportional to frequency, but it's also proportional to voltage squared.
Power loss is also proportional to transistor count, which is why ARM is such a low power processor.