Grouding your heatsink = Free MHz

[pHaestus]

Am I crazy, or did I read something about special EMI shielding being needed for Hammer CPUs?

[bigben2k]

hum... 7 full waves within one unit, where one unit is 50 ns, that would be a 140 GHz signal. My math must be off.

Maybe you could try the same with the PSU's +3.3 line, to see if there's any residual 60Hz signal in it. If you can, you could also check if the mobo's regulator circuit does filter out that signal.

[MeltMan]

Roger, but it will have to wait till late tonight. I have a ChemLab exam at 7:15, and work and class until then. The o-scope is a 1Mhz unit, so at its lowest setting, those should be 1mhz pulses? 50ns MTB is the lowest period it will read out.

[KnightElite]

Just out of curiosity, is your heatsink annodized? Annodizing isn't conductive, so you may have to scratch some of it off in order to actually ground it. You can test if it's grounded with a multi-meter. I hadn't really thought of that, until I was trying to ground my swiftech water block. Fortunately, I was still able to, since where "Swiftech" is engraved into the block, the annodizing is scraped off.

Excellent work though.

[Yo-DUH_87]

KnightElite, did you get any OC increase by grounding your CPU die (just out of curiosity)?

[KnightElite]

Not sure with the water block I have now, but with my CAK-38, I definitely did. Prior to installing the ground wire, it would BSOD after a few hours of uptime, after installing it I ran if for weeks at a time without a reboot. I even bumped it up 10MHz, and it was still stable.

[KnightElite]

Originally posted by bigben2k
hum... 7 full waves within one unit, where one unit is 50 ns, that would be a 140 GHz signal. My math must be off.

Maybe you could try the same with the PSU's +3.3 line, to see if there's any residual 60Hz signal in it. If you can, you could also check if the mobo's regulator circuit does filter out that signal.

I just checked your math, it should be 140MHz. I wonder...
Meltman, what was your FSB speed at the time that you did this? Possible correlation?

[Squeaky1]

It definitely helps. To what exact extent I am not sure. Grounding a shim with or without the hsf shielded, it produced the same results.

I can get the fsb to 200MHz with grounding, and only about 180 without, however the final CPU speed obtained is only 50MHz or so higher. Unclipping the grounding wire will result in a system crash quickly.

The thread was going pretty well, it's just a shame a few individuals were on a mission to prove it didn’t work, with nothing but text-book theory, and no practical evidence.

[bigben2k]

Originally posted by KnightElite
I just checked your math, it should be 140MHz. I wonder...
Meltman, what was your FSB speed at the time that you did this? Possible correlation?

Thanks! Where did I go wrong?

1 millisecond = 0.001 second
1 micro second = 0.000'001 second
1 nano second = 0.000'000'001 second, right>

so for 50 ns, we have 0.000000050 second. Frequency = 1/wavelength, so...

Hey it works!

But let's not be fooled, I'm not convinced that we're on track here. (Sorry Squeaky, we're going to beat on this one a little bit more).

So there's an FSB signal that can be detected, from the core to the ground. That's not surprising, in fact, it's pretty normal. However, if grounding the core will help filter it out, then all the best: it might make other components on the mobo act better.

I still think that the interference that this "grounding" is preventing, is in the 60 Hz range. That's the "mains", as our fellow OC'ers down under call it. I'd like to see if there's a measurable 60 Hz signal, at the core, the 5V line, the 3.3 line, and the CPU regulated output. Ah heck, let's throw in the 12V line as well, since it seems to feed some mobos.

Meltman, you've been found out: your FSB is 140, right?

[KnightElite]

MeltMan... we need the reply-age! I just had another thought... what if the smaller pulses are the DDR signal, with the large ones being the actual clock?

[Vector86]

i noticed that on my roomates compcrap er... i mean compaq the heat sink is grounded, also in a monitor i took a part once, the sinks in it were grounded as well.

another thing about that compaq, it died last week of unknown causes

[Can O' Beans]

well.. it WAS a Compaq..


and nice to see people diggin this back up, I wanted to see more evidence pointing to it actually working.

[GigaFrog]

What did you say was the max freq of the oscilloscope? 140MHz is already high for this kind of analog scope, you might be seing the tip of the iceberg (higher frequency pulses being missed by the scope). I wish I could borrow one of the 400 MHz digital scopes we have at work :-(
Nice job on the pics.

EMI filtering is not a precise technique, and there is often multiple parameters that may make a solution work great on some systems, and not at all on others, which makes some people say that a valid solution is BS.

I don't think interferences are coming from outside (unless you have a microwave oven in the area, maybe), so the closed case and side windows is IMO irrelevant.

I don't think either that the most sensitive signals are the FSB, DDR or PCI bus, but the internal busses of the CPU core that run at core clock. I am not saying these buses are not sensitive to interferences, just they are less than the CPU busses.

Grounding is a major issue when you design PCBs or ICs at frequencies above 1GHz. Not only for noise, but also for the correct control of impedances.

I did not do the grounding trick, because I am water cooled and I thought water would already ground the water block. Although there is an important parameter that could make this grounding method (by the water) not efficient: the grounding loop between the water block or Heatsink and the ground connections of the CPU must be as short as possible, otherwise the grounding loop might pick up more noise, plus its impedance will defeat its purpose (making the HSF a good ground).

For this reason, I am not sure using a switch would be a relevant test: because of longer connections, plus the switch itself, the ground loop may be unefficient. As a result, you would think that because you see no difference with the switch ON or OFF, grounding the HSF is BS, while I would think the test is un-conclusive, because of possible aleration of the ground connection by the switch and long wires.
Maybe it will make a difference with the switch ON and OFF, and then you will proof it works.
It is always interesting to make this test, but be carefull of conclusion you will take from the result.

Maybe I should try on my next system.

I hope this was not too boring.

[KnightElite]

I'm grounding my waterblock, and I know that STEvil for one got large gains from grounding his.

Water will not be a very good ground, since it's a rather poor conductor. It might be worth it to try it out with this technique. I'm almost tempted to take my computer to the lab at the university, since we have some really nice scopes there, but I don't really want to lug this monster that far.... .

[Since87]

Interesting thread. (I only read about 5% of the hard thread.)

There are so many different things that can come into play here.

The signals displayed on a 1 MHz scope are almost totally irrelevant. Most of that signal is probably ringing in the scope leads in response to much faster events that the scope is unable to measure accurately. The tip of the iceberg as GigaFrog said.

I have a 1GHz scope at work and 1.5 Ghz probes to go with it. It would give a much more accurate picture, but I don't think that seeing a more accurate picture of the signal on the heatsink would indicate much about what was going on.

One thing that would be very interesting to try, is to connect the heatsink to the CPU's ground with a MUCH shorter connection. Find the ground end of one of the capacitors on the CPU itself. Then solder a spring of some sort to that ground. (A thin piece of brass shim stock would be ideal.) Bend the spring so that when the heatsink is put on, it will make contact with the spring. This will provide a much more effective ground path for the high frequency signals radiated by the CPU. Whether this would improve things, make things worse, or make no difference, I wouldn't even venture to guess.

Because all wires have inductance; at the high frequency that CPU's run at these days, the length of signal paths is an extremely critical issue. It's reasonable to believe; that if a long ground lead from the heatsink to the chassis has an effect, then a short lead from the heatsink to as near the CPU as possible will have an even greater effect.

[KnightElite]

Here's the XtremeSystems clone of this thread. If you check post #34, STEvil did some experimenting and seems to have gotten a 40MHz gain on a 1600+ and 60MHz gain on an 1800+.

[KnightElite]

Originally posted by KnightElite
I may have an explanation as to why this works. I mentionned it to a 4th year EE student today, that I was able to clock higher with a grounded heatsink, and he was not at all surprised. I didn't entirely understand everything he told me, but he said it has to do with the fact that at the clock speeds that modern computers run at, all the traces have to act as waveguides, so one electrical signal creates a companion signal in the other traces.... (I probably got that completely wrong ). But in any case, that design feature causes an electric field in the heatsink, and that grounding it gives a better waveguide. He says that in theory, connecting the heatsink to the positive voltage line should work just as well as a ground, but it's not quite as safe .

I just posted this over at XS. Anyone know anything about waveguides?

[mfpmax]

Hey, are the AMD holes on Socket A boards tied to the ground layer on the motherboards?

[bigben2k]

Originally posted by KnightElite
I just posted this over at XS. Anyone know anything about waveguides?

What he was referring to is interference from one line to another. I don't see how that relates to grounding a heatsink though.

[cybrsamurai]

That settles it I'm going to wrap my head in copper foil and then run a wire to the nearest outlet ground! I should kick some ass on my finals with that kind of advantage!

[Yo-DUH_87]

I think that your mere organic processor would not suffer from any preformance increace from that meathod.

Although, you might get a slight shock, and a cool hairdo

[MeltMan]

Originally posted by KnightElite
MeltMan... we need the reply-age! I just had another thought... what if the smaller pulses are the DDR signal, with the large ones being the actual clock?

Jeez! Thought this thread was dead. FSB speed at 100. No DDR. Asus A7V motherboard, Duron 1000 cpu. Stock frequencies and voltages. Dont know how the math ends up with 140mhz, because the O-scope says 100mhz on the front of it?

[Rattlehead]

I hope this helps some....

mfpmax: I'm not sure about other motherboards and the four holes being grounded but I'm using an ASUS A7V8X with a heatsink that mounts using those four holes and when I connect the ground wire to my heatsink, it won't even boot.

I finally got out my meter and tested one of the ground wires where the power supply connects to the motherboard and to my heatsink and I show a connection.

Later all.