If you've read my posts and articles you've heard me use the phrase "ripple suppression" in reference to a PSU's performance. But do you know what this really means?
Ripple is a phenomenon that occurs with all SMPSUs (switch-mode power supply units). It is the tiny fluctuations in a power supply's output voltage, on the order of tens of milivolts, millions of times a second. It is an artifact of the process of converting a sine wave AC input to a constant DC output. Why it occurs is beyond the scope of this article; take an EE course if you are really, truly curious.
Also closely related, and often grouped together with ripple, is the phenomenon of electrical noise, which is fluctuations in voltage caused by exterior electromagnetic fields. From the user's point of view it's the same thing, and when I say "ripple" I will be referring to both ripple in the EE sense, and to electrical noise.
Ripple looks something like this:

As you can see it causes what would ideally be a smooth line to be a jerky, zig-zaggy line. In this instance you can also see a sine wave element occurring. Things like that are an artifact of the particular design and topology used; other designs may produce sporadic jumps and spikes, pronounced "dual wave" patterns, or even stranger patterns. Again, a full explanation would be beyond the scope of this article.
Ripple is measured as a peak-to-peak value in milivolts, and is measured using an oscilloscope (software can not detect ripple, nor can a multimeter or voltmeter; you need an oscilloscope with a scanning frequency of at least 30MHz). The ATX specification sets these limits on computer PSU ripple levels:
+12V - 120mV
+5V - 50mV
+3.3V - 50mV
+5VSB - 120mV
-12 - 120mV
My personal preference is to see numbers in or under this range:
+12V - 80mV
+5V - 30mV
+3.3V - 30mV
+5VSB - 50mV
-12V - 80mV
Any power supply which has ripple outside of those proscribed by the ATX specifications can and will cause short term hardware damage (over a period of days to months, depending). Anything outside of my preferences may cause long-term reliability degradation and other issues, though it won't damage your hardware in the short term.
Why is having good ripple suppression important? The first reason is hardware lifespan. Ripple wears on several types of electrical component, the most relevant being electrolytic capacitors. The closer ripple levels are to a capacitor's rated ripple, the shorter its lifetime will be, and ripple exceeding its rated level can cause it to blow or perform poorly. Since electrolytic capacitors are found in your power supply, motherboard, graphics card, and many other components, having a power supply with good ripple suppression is a must for long-term reliability.
Another problem poor ripple levels can cause is one that will surely strike fear into the heart of every overclocking enthusiast. Higher ripple levels means lower overclock.
Hub-a-duh-WAH?
Your motherboard's VRMs (voltage regulation modules) convert the mid-voltage DC from your power supply into very low-voltage DC for itself and your components (+12V powers CPU and main chipsets, +3.3V powers RAM and GPU control circuits, +5VSB controls BIOS and CMOS and basic mobo functionality, etc.). Though your VRMs are very good at converting slightly out-of-spec voltages to the correct voltage, they aren't always good at removing ripple. Thus if your PSU has poor ripple suppression on the +12V rail...
Let's jump to the inside of the CPU. Here it is, ticking away the clock cycles at its stock clockspeed. The transistors inside the CPU need a certain voltage to pass through them for them to do their job (red line). The voltage your motherboard attempts to supply to the CPU is the Vcore (green line).

As you can see the Vcore takes a bit (a fraction of a nanosecond) to reach its intended voltage.

When you overclock you increase the clock speed, thus reduce the time the voltage has to reach the target Vcore. It may never even reach that voltage, but will still exceed the transistors' required voltage. When the clock speed gets high enough, the voltage may not even have time to clear the required voltage, forcing us to increase our Vcore.
However, if your PSU is giving out high ripple levels, you may run into a problem...

Rather than an ideal logistic rise in voltage, instead you have a jagged rise. And in a certain percentage of clock cycles... BAM. A dip caused by ripple coincides with the peak of a clock cycle, and the voltage doesn't reach the required level, causing an error, causing a crash, meaning an unstable overclock. You have to increase Vcore to compensate for the dips, which causes increased heat. And if ripple is bad enough you may never find a stable clock.
The same concepts apply to your motherboard's integrated circuitry (like the northbridge) and your GPU and its VRMs.
So what can you do to stop bad ripple from harming your components or even (gasp!) reducing your potential overclock? Well, higher-end motherboards and graphics cards have better VRMs, with more phase units, which use capacitors to filter ripple. Thus you could theoretically reach a higher overclock with a high-end motherboard and a low-end PSU, than with a mid-range motherboard and a mid-range PSU. However, this is a bad idea because, as mentioned before, ripple degrades capacitors and can cause your expensive parts to fail.
The simplest solution is to get a quality power supply. Read reviews (jonnyguru.com, hardwaresecrets.com, hardocp.com, etc.) that use professional equipment to test their power supplies, including quality oscilloscopes. That will tell you what ripple levels a PSU is likely to get at varying loads. Another important thing to note is the quality of the capacitors used. Low-end Chinese capacitors will degrade quickly, or even blow, and as a capacitor degrades its ability to filter ripple diminishes, causing your PSU's ripple levels to increase. Japanese-brand capacitors from companies like Nippon Chemi-Con, Panasonic, and Rubycon are much more reliable and longer lived. Also, the higher the temperature the capacitor is rated for the longer it will last and still filter effectively.
So the bottom line? If you want to overclock to the maximum and keep your components running for as long as possible, get a quality power supply with good ripple suppression. You won't regret it.
Further Reading
Edited by Phaedrus2129 - 12/3/10 at 10:04am
Ripple is a phenomenon that occurs with all SMPSUs (switch-mode power supply units). It is the tiny fluctuations in a power supply's output voltage, on the order of tens of milivolts, millions of times a second. It is an artifact of the process of converting a sine wave AC input to a constant DC output. Why it occurs is beyond the scope of this article; take an EE course if you are really, truly curious.
Also closely related, and often grouped together with ripple, is the phenomenon of electrical noise, which is fluctuations in voltage caused by exterior electromagnetic fields. From the user's point of view it's the same thing, and when I say "ripple" I will be referring to both ripple in the EE sense, and to electrical noise.
Ripple looks something like this:

As you can see it causes what would ideally be a smooth line to be a jerky, zig-zaggy line. In this instance you can also see a sine wave element occurring. Things like that are an artifact of the particular design and topology used; other designs may produce sporadic jumps and spikes, pronounced "dual wave" patterns, or even stranger patterns. Again, a full explanation would be beyond the scope of this article.
Ripple is measured as a peak-to-peak value in milivolts, and is measured using an oscilloscope (software can not detect ripple, nor can a multimeter or voltmeter; you need an oscilloscope with a scanning frequency of at least 30MHz). The ATX specification sets these limits on computer PSU ripple levels:
+12V - 120mV
+5V - 50mV
+3.3V - 50mV
+5VSB - 120mV
-12 - 120mV
My personal preference is to see numbers in or under this range:
+12V - 80mV
+5V - 30mV
+3.3V - 30mV
+5VSB - 50mV
-12V - 80mV
Any power supply which has ripple outside of those proscribed by the ATX specifications can and will cause short term hardware damage (over a period of days to months, depending). Anything outside of my preferences may cause long-term reliability degradation and other issues, though it won't damage your hardware in the short term.
Why is having good ripple suppression important? The first reason is hardware lifespan. Ripple wears on several types of electrical component, the most relevant being electrolytic capacitors. The closer ripple levels are to a capacitor's rated ripple, the shorter its lifetime will be, and ripple exceeding its rated level can cause it to blow or perform poorly. Since electrolytic capacitors are found in your power supply, motherboard, graphics card, and many other components, having a power supply with good ripple suppression is a must for long-term reliability.
Another problem poor ripple levels can cause is one that will surely strike fear into the heart of every overclocking enthusiast. Higher ripple levels means lower overclock.
Hub-a-duh-WAH?
Your motherboard's VRMs (voltage regulation modules) convert the mid-voltage DC from your power supply into very low-voltage DC for itself and your components (+12V powers CPU and main chipsets, +3.3V powers RAM and GPU control circuits, +5VSB controls BIOS and CMOS and basic mobo functionality, etc.). Though your VRMs are very good at converting slightly out-of-spec voltages to the correct voltage, they aren't always good at removing ripple. Thus if your PSU has poor ripple suppression on the +12V rail...
Let's jump to the inside of the CPU. Here it is, ticking away the clock cycles at its stock clockspeed. The transistors inside the CPU need a certain voltage to pass through them for them to do their job (red line). The voltage your motherboard attempts to supply to the CPU is the Vcore (green line).

As you can see the Vcore takes a bit (a fraction of a nanosecond) to reach its intended voltage.

When you overclock you increase the clock speed, thus reduce the time the voltage has to reach the target Vcore. It may never even reach that voltage, but will still exceed the transistors' required voltage. When the clock speed gets high enough, the voltage may not even have time to clear the required voltage, forcing us to increase our Vcore.
However, if your PSU is giving out high ripple levels, you may run into a problem...

Rather than an ideal logistic rise in voltage, instead you have a jagged rise. And in a certain percentage of clock cycles... BAM. A dip caused by ripple coincides with the peak of a clock cycle, and the voltage doesn't reach the required level, causing an error, causing a crash, meaning an unstable overclock. You have to increase Vcore to compensate for the dips, which causes increased heat. And if ripple is bad enough you may never find a stable clock.
The same concepts apply to your motherboard's integrated circuitry (like the northbridge) and your GPU and its VRMs.
So what can you do to stop bad ripple from harming your components or even (gasp!) reducing your potential overclock? Well, higher-end motherboards and graphics cards have better VRMs, with more phase units, which use capacitors to filter ripple. Thus you could theoretically reach a higher overclock with a high-end motherboard and a low-end PSU, than with a mid-range motherboard and a mid-range PSU. However, this is a bad idea because, as mentioned before, ripple degrades capacitors and can cause your expensive parts to fail.
The simplest solution is to get a quality power supply. Read reviews (jonnyguru.com, hardwaresecrets.com, hardocp.com, etc.) that use professional equipment to test their power supplies, including quality oscilloscopes. That will tell you what ripple levels a PSU is likely to get at varying loads. Another important thing to note is the quality of the capacitors used. Low-end Chinese capacitors will degrade quickly, or even blow, and as a capacitor degrades its ability to filter ripple diminishes, causing your PSU's ripple levels to increase. Japanese-brand capacitors from companies like Nippon Chemi-Con, Panasonic, and Rubycon are much more reliable and longer lived. Also, the higher the temperature the capacitor is rated for the longer it will last and still filter effectively.
So the bottom line? If you want to overclock to the maximum and keep your components running for as long as possible, get a quality power supply with good ripple suppression. You won't regret it.
Further Reading
Edited by Phaedrus2129 - 12/3/10 at 10:04am






Now i know more +rep



the OP is why we cap GPU's to run stupid clock speeds too...input and output phases... and do other things like add inductors.. all in the name of canceling or minimizing ripple in the circuit.
