Archon

[Grinder]

Wow. I must be getting senile or something.

I pulled the cpu and was examining the block to see how flat it appeared to be, when I noticed that it still had the plastic on it. I thought I removed it, but I must have been thinking of the two GPU blocks. It seems plastic doesn't make a fabulous thermal interface...

So, before I delid, I'll check temps WITHOUT the plastic!

[Bartacus]

LOL! Grinder, you're not the first man to do that, and you won't be the last. Kudos to you for having the stones to admit it.

[Grinder]

Thanks Bartacus, I'm so proud... But at least we get to have a good laugh. And to think I've been watercooling for 10 years! Just goes to show, you shouldn't mix assembly with old age and

[10e]

Good job!

Problem solved!

[Grinder]

Thanks, it does seem to have made a minor improvement.

Load temps in prime after 1.5 hours are about 25 degrees less...

Now to try that OC again!

[Grinder]

It's not a stellar chip, but much better results than before (duh). It was at 4.4 with 1.25ish load, and I was seeing 96 degrees on the hottest cores after 40 min of prime.

Now I have it more or less stable (ran prime95 for almost 1.5 hours) at 4.7, although voltage is higher than I'd like to see at 1.37ish. With the rad fans on max I'm seeing max core temps of 86. Mind you the wood stove is on in the room, so ambient is 26C. 4.6 requires 1.312 and gives a max of 78 degrees during a similar prime run and ambient.

[10e]

That's about normal/average for these chips. I'd probably try and stick with lower than 1.35 volts vCore just to eliminate any possibility of degradation.

The best 3770K I had was one that did 4.9 at 1.33v vCore but that was a pretty special chip, only hitting 60 celsius after a de-lid.

[Grinder]

I'm still toying with the idea of delidding it. If it lowered temps enough that I could hit the same frequency (4.7) at reduced voltage, or 4.8 below 1.39 or so, I'd probably go for it. Although being 1.5 percent over the generally accepted max volts is probably acceptable for me, especially since I'll set it up in offset once I finish tweaking, so it won't be at max volts all the time (not folding or anything).

At first I had turned down PLL to 1.5 to keep temps lower, however I did find that bumping it back up to 1.8 increased stability a tiny bit.

How did delidding affect the voltage you needed at a given frequency, or your ability to clock higher?

[10e]

It only seemingly affected temperatures. I went down from nearly 80 celsius to low 60s on the hottest core. Like I said, I had a very good chip, though. Unfortunately either dropping it (the only time I've ever dropped a chip) or putting on the naked ivy kit with the wrong block killed this chip. I would mourn it, but I have a new one that seems pretty good, doing 4.3Ghz undervolted.

In some cases I have seen lower temperatures allow less voltage to be sent to a chip, so this might have been a side benefit, but I never went above 4.6Ghz prior to de-lidding it, and then I made it fairly easily to 4.9Ghz after the de-lid.

5.0Ghz was a no-go needing more than 1.39 volts. As long as you aren't constantly stress testing the chip or running it at 100% then you should be able to max out a bit higher, so if it's worth it, you might want to go for it.

The question begged here is whether a few extra mhz are worth it in the long run.

[Grinder]

Thanks for the info. I'm still mulling it over.

Meanwhile, I finally ordered the 5mm LEDs for the fan mod, plus some bits and pieces to get ready to tackle my Blood and Guts build again (finally).

I was thinking of ordering an aquaero, but money is just too tight. I had problems with my t-balancer miniNG, it has the power but I couldn't adjust the fan curve where I wanted it with the rad fans and water temp sensor. After spending an entire day wrestling with the big brother, the t-balancer bigNG, I finally got the software working again in Blood and Guts. So I attached the miniNG to the bigNG, and through that I should be able to change the curve and then transplant it back to Archon, since it is supposed to store the curve values in flash memory. Will post back after I've tuned the curve and tried it standalone.

[Grinder]

I was successful in connecting the miniNG on Archon to the bigNG on Blood and Guts. Then I tuned the curves for the miniNG, channel A for water temp/fan rads and channel B for GPU VRM temp/fan mod. The resulting curves were stored in flash memory. They look so cute connected together. Hopefully they will reproduce!



You can really see how much bigger the XL R2 is than the Midi R2.



I've since separated them again, and the new miniNG curves were preserved. The issue with the default curve was that it was off below 35C, and above that it ramped from 50 percent at 35C to 100 percent at 60C. At idle and normal room temp, the water temp was only 24C. Even at load in 3DMark using the cpu at 4.7 and both 290s at 1150/1350 the temp only peaked at 34C with the ambient increased to 24C. I adjusted the water curve to kick the fans on 25% at 24C, and ramp to max at 36C.

The VRM curve wasn't as far out of whack, but still I changed it to turn on at 25%/39C and ramp to full speed at 60C. Here's an example of the software showing the curve, midway through the tuning process.



Saved me from having to buy a new fan controller capable of custom curve definitions!

[xmanrigger]

It only seemingly affected temperatures. I went down from nearly 80 celsius to low 60s on the hottest core. Like I said, I had a very good chip, though. Unfortunately either dropping it (the only time I've ever dropped a chip) or putting on the naked ivy kit with the wrong block killed this chip. I would mourn it, but I have a new one that seems pretty good, doing 4.3Ghz undervolted.

In some cases I have seen lower temperatures allow less voltage to be sent to a chip, so this might have been a side benefit, but I never went above 4.6Ghz prior to de-lidding it, and then I made it fairly easily to 4.9Ghz after the de-lid.

5.0Ghz was a no-go needing more than 1.39 volts. As long as you aren't constantly stress testing the chip or running it at 100% then you should be able to max out a bit higher, so if it's worth it, you might want to go for it.

The question begged here is whether a few extra mhz are worth it in the long run.

You have me confused. Last I heard, temperature was a function of voltage, not the opposite as you put it. Please explain.

[Grinder]

If a component is cooler, it takes less voltage to achieve the same frequency.

[xmanrigger]

If a component is cooler, it takes less voltage to achieve the same frequency.

First I have ever heard of that. The reason it is running cooler is because clocks are lower, thus, lower voltage. Raise the clock speed and/or voltage, and temperature will rise accordingly. Even a stock un-overclocked GPU or CPU will ramp voltage for higher/maximum clock speed, generating more heat, even if that rise is negligible.

Not trying to be Jonny Noitall, but that just dont make sense to me. I have been benching and overclocking for years and first I have ever heard that statement or 10e's.

[Grinder]

That's been my experience over the years, and I've heard it many times as well. Not that I have any hard data to back it up Like you, I've been building/overclocking/watercooling for many years. Building since the 80's, OCing since the 90's, water since Jan 2005.

We aren't talking just a couple of degrees here making a difference. But I operated a radiator in my (unheated) attached garage, for the computer in my office. After dialing in an OC in the winter, I found I had to raise the voltage come summertime. However I was able to lower it again when it got cold. This was with an AMD 1090T. Unfortunately, I don't have my notes so can't tell you exactly how much of a voltage difference there was operating the chip at 4Ghz winter vs summer.

My hope was that delidding my 3770k and lowering peak operating temps by 15-20C would be enough to lower the voltage. Still haven't decided whether to go for it.