Delidded and lapped G5328

This is my second G5328. Not the grand prize silicon lottery winner by any means but a significantly better overclocker than my first one. Got it new off ebay for about $65.

It overclocked easily to 4.5 ghz on vcore of 1.90 1.29, Ring Ratio of 39x with Ring voltage at 1.178. With a tiny bump in System Agent Voltage to help the IMC a little I got it Prime95 stable with the RAM running at 1600. I'm pretty sure I could get the chip to 4.6 ghz stable with vcore around 1.325 but I'm not going to run it 24/7 with that much vcore anyway so I really haven't tested that assumption yet.

All that to give you some context for the thread title.

Under good all in one water (Swiftech H320) I was getting Prime95 max core temps of 72c with a spread of 12c between the two cores. Certainly a safe temp but the tinker bug hit me as it frequently does. I just can't leave well enough alone these days.

So, I delidded it and put Liquid Ultra under the hood. I also took the lid from the 4790k I fried last week in a botched attempt to go naked, lapped it and slapped it on the new delidded G3528. On the top side I just used AC MX2.

The result: Lowered max Prime95 temps a full 20c and brought the separation of the two cores within 1c of each other. I'm very pleased. The combination of delidding with Liquid Ultra is something I have done several times now and it has turned out to be a big time temp lowering strategy that is just more than anecdotal hype. Temp improvement will vary from one trial to the next but for me it has always been a significant improvement.

I feel like the speed/voltage given is incorrect. 1.9vCore is LN2 territory

Thanks for catching that typo. I'll change that vcore. I meant to type "1.29".

*edit*
err nvm, for some reason i thought i found this thread in the FS section.

any way! whats the batch info on it?

1.29V for 4.5 on a G3258 is solid, IIRC.
Mine takes 1.65V for 4.8GHz in Cinebench R15.

Evilsizer said:

*edit*
err nvm, for some reason i thought i found this thread in the FS section.

any way! whats the batch info on it?

Click to expand...

3418C454

Looks like first half of 2014 manufacture date. People say the G3258s produced early after intro generally overclocked better than later ones.

Man Mr T you are going to make me pull the lid off my 4770k, I'd love to see a 20c drop in temps. Mine is an absolute beast when it comes to heat, it will do 4.4 @ 1.23 V with temps in the high 80's on a Megahalems but I cannot get it any higher even on a custom loop because of the heat.

Just do it, Manny! I believe in ya' kid!

trents said:

3418C454

Looks like first half of 2014 manufacture date. People say the G3258s produced early after intro generally overclocked better than later ones.

Click to expand...

how you figure that out? wonder about mine, 3418B993, still running along at [email protected], tried lower but didnt like it. i might have other settings to mess with, after some reading seems my settings are a bit off. im using either 1.8v IN or 2.00v IN. forget about the uncore but i swore it was 1:1 not 200-400mhz lower.

Evilsizer said:

how you figure that out? wonder about mine, 3418B993, still running along at [email protected], tried lower but didnt like it. i might have other settings to mess with, after some reading seems my settings are a bit off. im using either 1.8v IN or 2.00v IN. forget about the uncore but i swore it was 1:1 not 200-400mhz lower.

Click to expand...

https://rog.asus.com/forum/showthread.php?278-HOWTO-Read-an-Intel-CPU-FPO-Batch-Code

With the two G3258 CPUs I have messed with it was not true that all I had to do was raise the CPU ratio and the vcore in order to get a good clock as many have said.

I had to also increase the Ring (aka, uncore) ratio and the Ring voltage, and bump the System Agent voltage up a smidgen. I also had to increase the I/O dgital and the I/0 analog voltage. This article was helpful to me: http://www.overclockersclub.com/guides/intel_pentium_g3258_oc_guide/2.htm in understanding what settings needed to be altered to get the most out of my chip. Though, I will have to say his numerical values for those settings had to be significantly modified for my chip but the concepts were the same. With my CPU I needed to use considerably more vcore but considerably less Ring Voltage and considerably less I/0 digital and analog voltages than the guy who wrote that article. That didn't surprise me because everybody knows these Haswells are all over the map and each one needs individual attention.

One thing I have found that makes Liquid Ultra application considerably easier: take a pair of scissors and trim the applicator brush bristles back about half way. Not only will you use less LU but you will have much more control over the spreading process.

Another tip is to cover the resistors next to the die on the PCB with nail polish top coat. The biggest risk in using LU is that you will splash some over onto the circuitry elements that reside on the PCB.

Lapping the lid like I did is not essential but in doing so I found the top to be considerably more concave in shape than what met the eye. Worth a couple degrees perhaps.

trents said:

Just do it, Manny! I believe in ya' kid!

Click to expand...

LOL

Hmmm...wondering if delidding would make a noticeable difference at around -100C. Have had my $65 g3258 over 6GHz...would love to see it give more (or die trying).

Since my cooling's busted am using the time to clean up, organize, get a handful of boards out for rma, hardwire my 780ti matrix cards, etc, etc...maybe I'll ask Shrimpy to do a delid for me.

Think it'd make a difference when running cold?

I should think it would make a temp difference with any kind of cooling. Delidding simply removes the factory obstacle to efficient heat transfer between the die and the lid which is the thick layer of gunky TIM paste. As the thermal differential between the die and the cooler interface or the coolant increases (in the case of "cold") the more effective delidding (and replacing the TIM with Luquid Ultra) proves to be. At idle the improvement is small. Under load the improvement becomes obviously significant. At least that's how I see it with the cold question. To put it another way, regardless of the kind of cooling you are doing, delidding/LU simply allows you to remove heat from the die faster.

trents said:

3418C454

Looks like first half of 2014 manufacture date. People say the G3258s produced early after intro generally overclocked better than later ones.

Click to expand...

My experience tells otherwise. All early G3258 which I had were really bad like 4.5GHz 1.4V. All new ( end of last year/early 2015 ) could make 4.5GHz 1.28-1.32V. My last one can make 4.7GHz 1.3V and it's from about march this year.

Delidding doesn't help much on LN2. Can try anyways. The biggest problem is to find a chip which is scalling good with voltage above 5GHz. Most haswells are bad regardless if it's pentium or i7. In most cases they're scalling good up to ~4.7-4.8GHz and above that there are bigger voltage steps till you see something like 0.1V per 100MHz.

Thanks for the input, Woo. But actually, I was not addressing the issue of whether or not it would make a difference in the overclock when going cold but I was just addressing what I thought was a question about temp improvement. Yes, the two don't necessarily equate and often do not.

I just added that LN2 comment to what funsoul said. I should have quoted him in my last post to be clear.