TR Forums

sschaem wrote:

So tell me again how I can reach 4.4ghz at 1.2v with a third party cooler ?

You are missing that point. You seem to have gotten an uncooperative CPU that can't do 4.4GHz at 1.2V, we all know that. Going for a better cooler allow you to offset the heat generated by increased voltage in order to hit your target. There are only so many knobs to turn and if one of them won't, you have to dial the others to compensate.

nanoflower wrote:

He's not the only one to try and OC a G3258 with the stock heat sink. The OP in this thread http://www.overclock.net/t/1500524/intel-pentium-g3258-performance-and-owners-thread was doing much of his testing with the stock cooler but I think has since upgraded. That's probably the best thread on the G3258 and OCing with numerous people reporting their results using different coolers/MBs/RAM.

As for the temps I did hit the low 90s when I first started testing with my Pentium. I think it was a combination of the TIM paste being fresh and perhaps a bit too much being applied to the heat sink. Now it's staying under the low 80s during the same runs of Intel Burn Test. That's with the voltage set at 1.3 and the multiplier set at 44x for a 4.4GHz run on a MSI PC Mate MB. Though I will admit I'm leaving the cache alone as setting it to the same 44x mutiplier causes an instant crash even with a boosted voltage and I'm not willing to spend a lot of time playing around to figure out what OC level it can support since I'm not sure how much of a boost in performance it would provide.

I will have to read this one, because I have been running both part of the chip at 4.2ghz (cores & Cache/Uncore) and it seem people are running the Cache/Uncore at much lower speed then the cores...

Actually it seem 4.2ghz for the Cache/Uncore is decent

Dont have the system handy. But I will have to try 4.4ghz core & 4.0ghz uncore at both at 1.2v

Thanks for the hint!

You are right, but you are also wrong. While using a better cooler doesn't always make lower voltages or higher clocks possible, it usually does, and it almost certainly would in the situation described here.

It's not a simple boolean of "more cold = lower voltages plus higher clocks", at least not at the thermal difference two air coolers generally provide, but there *is* a clear relationship. Ask anyone who has done any LN2 or dry ice overclocking. A CPU that manages 4.4GHz and 1.2V on a good air cooler can do more clocks with less volts under extreme cooling. Of course, then you can add the volts back and go even further, but that's a different point. Normally the difference between two air coolers isn't enough to show this difference, but the Intel cooler is pretty lousy, and the Hyper 212 is really good, despite being very cheap.

sschaem wrote:

nanoflower wrote:

He's not the only one to try and OC a G3258 with the stock heat sink. The OP in this thread http://www.overclock.net/t/1500524/intel-pentium-g3258-performance-and-owners-thread was doing much of his testing with the stock cooler but I think has since upgraded. That's probably the best thread on the G3258 and OCing with numerous people reporting their results using different coolers/MBs/RAM.

As for the temps I did hit the low 90s when I first started testing with my Pentium. I think it was a combination of the TIM paste being fresh and perhaps a bit too much being applied to the heat sink. Now it's staying under the low 80s during the same runs of Intel Burn Test. That's with the voltage set at 1.3 and the multiplier set at 44x for a 4.4GHz run on a MSI PC Mate MB. Though I will admit I'm leaving the cache alone as setting it to the same 44x mutiplier causes an instant crash even with a boosted voltage and I'm not willing to spend a lot of time playing around to figure out what OC level it can support since I'm not sure how much of a boost in performance it would provide.

I will have to read this one, because I have been running both part of the chip at 4.2ghz (cores & Cache/Uncore) and it seem people are running the Cache/Uncore at much lower speed then the cores...

Actually it seem 4.2ghz for the Cache/Uncore is decent

Dont have the system handy. But I will have to try 4.4ghz core & 4.0ghz uncore at both at 1.2v

Thanks for the hint!

Yeah, FWIW, I left my uncore alone at 3.2GHz. The auto-clocker app put them there, and I remembered from LGA1366 overclocking that the old i7s liked the uncore to be at 2X ram speed, more being better, but more than 2X ram speed showing greatly diminishing returns.

On an old Core 2 laptop, I undervolted it using RMClock.

Then I replaced the thermal paste and cleaned out the heatsink.

I was then able to lower the voltage by an extra .025V on all of the turbo levels except for the max one. A shame I couldn't OC the CPU though. Not that I would want to play games on it because it was using the pre-SandyBridge Intel GMA IGP.

Flying Fox wrote:

sschaem wrote:

So tell me again how I can reach 4.4ghz at 1.2v with a third party cooler ?

You are missing that point. You seem to have gotten an uncooperative CPU that can't do 4.4GHz at 1.2V, we all know that. Going for a better cooler allow you to offset the heat generated by increased voltage in order to hit your target. There are only so many knobs to turn and if one of them won't, you have to dial the others to compensate.

The thread is about me reporting the clock speed attainable at 1.2v with a G3258.
Max overclock was never the goal. And heat never was an issue in any of my testing with the stock cooler.

So getting a third party cooler will do nothing to help in this situation. Again, this never was about max overclock at any voltage. But what can the chip do a 1.2v
Anand part can reach 4.2ghz at 1.15v , my part requires 1.27v

I can only guess that the gap is bigger at the extreme. I wouldn;t be surprised if some people find out that they need 1.3v to run stable at 4.2ghz

Note: if I did have a third party cooler it might have been more interesting if I had tested how far the chip can go at insane voltages. Most chip seem to plateau at 4.7 / 4.8ghz
But anand could not go past 4.7ghz even with 1.5+v , even so he seem to have a better sample... So I dont have much hope for the one I got.
In the end it possible that I would have maxed out at ~4.5ghz...

Forge wrote:

You are right, but you are also wrong. While using a better cooler doesn't always make lower voltages or higher clocks possible, it usually does, and it almost certainly would in the situation described here.

It's not a simple boolean of "more cold = lower voltages plus higher clocks", at least not at the thermal difference two air coolers generally provide, but there *is* a clear relationship. Ask anyone who has done any LN2 or dry ice overclocking. A CPU that manages 4.4GHz and 1.2V on a good air cooler can do more clocks with less volts under extreme cooling. Of course, then you can add the volts back and go even further, but that's a different point. Normally the difference between two air coolers isn't enough to show this difference, but the Intel cooler is pretty lousy, and the Hyper 212 is really good, despite being very cheap.

We are talking about a much smaller difference between Intel stock cooler and even the Hyper 212.
LN2 cooling VS stock is an extreme case, so I can see having a measurable effect on voltage requirements.
(But never saw/read report anything about it)
I wonder also if a third party cooler would help in the same manner...

I also dont recall the BSOD being heat related.
Prime95 test 1 and 2 would pass (well 1 hour each during this testing phase) and those gave me the highest temp reading.
But it was the third test, the one that involve main memory, that would randomly fail hard.

sschaem wrote:

We are talking about a much smaller difference between Intel stock cooler and even the Hyper 212.
LN2 cooling VS stock is an extreme case, so I can see having a measurable effect on voltage requirements.

Big differences, big results. Smaller differences, smaller results.... BUT THE PATTERN HOLDS.

sschaem wrote:

I also dont recall the BSOD being heat related.
Prime95 test 1 and 2 would pass (well 1 hour each during this testing phase) and those gave me the highest temp reading.
But it was the third test, the one that involve main memory, that would randomly fail hard.

You *wouldn't* know if heat was the cause, because heat is NOT alone! This is the point I keep making! If the heat is higher, the chip is less stable, even at the same clocks and voltages. If you then INCREASE voltages, you have COMPOUNDED the heat increase!

This is usually where most overclocking projects end, people say "It just wouldn't get stable, no matter what voltage I gave it". More voltage does two things. It can make the signalling inside the chip "louder" and thus get a borderline signal through, or it increases heat. When OCing you want the first effect, but want to avoid the second. That's why you shouldn't start by putting your CPU at the max voltages you'll allow and backing down. If you want to OC correctly, you start slowly increasing the clocks until things become borderline unstable, and then you try increasing the voltage. If that doesn't work, sometimes you should actually DECREASE the voltage. In particular, "hot bins" like the 4790K I'm running now, they often ship with voltages higher than they really need, though this is much less common than it used to be. By using a better cooler, or more controlled environmental temperatures, you can decrease voltages below stock and still run stable.

At any point, you can say "I don't want to listen to you anymore" and I'll go away, but telling me I'm wrong won't do that. I've done a lot of tweaking and clocking and tinkering and I'm telling you straight; Using an underperforming cooler while overclocking WILL have a serious impact on the clocks you can achieve and the voltages you can use. The stock Intel coolers are the very definition of "barely adequate", and running at any clocks or volts higher than stock while using the Intel cooler is counterproductive.

Forge wrote:

sschaem wrote:

We are talking about a much smaller difference between Intel stock cooler and even the Hyper 212.
LN2 cooling VS stock is an extreme case, so I can see having a measurable effect on voltage requirements.

Big differences, big results. Smaller differences, smaller results.... BUT THE PATTERN HOLDS.

sschaem wrote:

I also dont recall the BSOD being heat related.
Prime95 test 1 and 2 would pass (well 1 hour each during this testing phase) and those gave me the highest temp reading.
But it was the third test, the one that involve main memory, that would randomly fail hard.

You *wouldn't* know if heat was the cause, because heat is NOT alone! This is the point I keep making! If the heat is higher, the chip is less stable, even at the same clocks and voltages. If you then INCREASE voltages, you have COMPOUNDED the heat increase!

This is usually where most overclocking projects end, people say "It just wouldn't get stable, no matter what voltage I gave it". More voltage does two things. It can make the signalling inside the chip "louder" and thus get a borderline signal through, or it increases heat. When OCing you want the first effect, but want to avoid the second. That's why you shouldn't start by putting your CPU at the max voltages you'll allow and backing down. If you want to OC correctly, you start slowly increasing the clocks until things become borderline unstable, and then you try increasing the voltage. If that doesn't work, sometimes you should actually DECREASE the voltage. In particular, "hot bins" like the 4790K I'm running now, they often ship with voltages higher than they really need, though this is much less common than it used to be. By using a better cooler, or more controlled environmental temperatures, you can decrease voltages below stock and still run stable.

At any point, you can say "I don't want to listen to you anymore" and I'll go away, but telling me I'm wrong won't do that. I've done a lot of tweaking and clocking and tinkering and I'm telling you straight; Using an underperforming cooler while overclocking WILL have a serious impact on the clocks you can achieve and the voltages you can use. The stock Intel coolers are the very definition of "barely adequate", and running at any clocks or volts higher than stock while using the Intel cooler is counterproductive.

Show me proof that an aftermarket cooler would allow me to significantly lower my voltage a 4.2ghz... If you dont have any data, how can I take anything you say about this seriously.

For the rest, you are just arguing with yourself. Because I never outline overclocking techniques or else.

I think its misleading to tell people "spend money on a cooler, it will lower your voltage requirement"
"Oh, your G258 need 1.4v on the stock cooler? get a Hyper 212 and you can lower it to 1.2" . this is nonsense. no matter what story you tell about your 4790k.

On that note, the intel stock cooler got my 2.4ghz Q6600 at 3.2ghz running flawlessly for the past 7+ years... (and its also undervolted) So no, stock cooler are NOT always "barely adequate".

All I'm telling you is that the stock cooler is more then adequate for this chip at 4.2ghz.. Its a freaking 22nm dual core CPU with AVX disabled, its a tiny CPU . (BTW, this CPU is paired with a 7970, the GPU side is 'disabled')
So spending more money on a third party cooler will not reduce this chip voltage requirement and is just a waste of money.

If you have proof to the contrary, link please. (not a link to some LN2 experimental cooling on some I7 CPU...)

LOL. You should look through that thread. There was one post from someone that did use liquid nitrogen on his G3258. I think his final results were a 6GHz overclock but he had to push the voltage up to 1.5 (or maybe higher.) I'm not positive of the results but they were insane. Of course he isn't running at those speeds normally but it just goes to show what the combination of better cooling (insane in this case) and higher voltages (also what I would call insane values in this case) can accomplish.

For my own systems I like to see a good overclock value but also to keep the idle temps low. Pumping up the voltage gets in the way of lower idle temps, especially if you lock the multiplier.

sschaem wrote:

So tell me again how I can reach 4.4ghz at 1.2v with a third party cooler ?

Leakage (or electrical resistivity) increases with temperature, as OP in this thread shows with his experiment.
2600k running at a constant 2GHz and 1.29V increases power consumption by 23W just by raising the temperature from 47C to 96C.

So you could actually gain stability at the same voltage just by reducing temperature. It's not nonsense, but physics.

Oh, $220 board, $2 retail-box cooler?
I'm pretty sure the Pentium and Celeron stock coolers are even cheaper than the i3 ones, in that they don't even have a copper core - so $2 is probably being generous.....

/facepalm
/thread

jihadjoe wrote:

sschaem wrote:

So tell me again how I can reach 4.4ghz at 1.2v with a third party cooler ?

Leakage (or electrical resistivity) increases with temperature, as OP in this thread shows with his experiment.
2600k running at a constant 2GHz and 1.29V increases power consumption by 23W just by raising the temperature from 47C to 96C.

So you could actually gain stability at the same voltage just by reducing temperature. It's not nonsense, but physics.

No, clearly sschaem is right and I'm an idiot. I'll stop polluting his thread now. Plonk.

Forge wrote:

UnfriendlyFire wrote:

If there were the resources, someone could design a CPU that has only the very basic functions and let the mobo have the north/southbridge, memory controller, L2 cache, and etc.

Watch that motherboard's cost skyrocket.

For a long time, all motherboards were that way. Welcome back to Socket 7!

Yeah, and I remember (like you, probably) how moving the memory controller and cache to the CPU improved both performance and reliability.

Ahh, the 256k (upgradeable to 512k) and then 512k 486 mainboards across eight chips. And then the Pentium boards with 256k or 512k of pipeline burst-cache on a fingered module. I remember having the occasional cache chip or module go bad and result in some wild, weird stuff.

To the OP: An inexpensive aftermarket cooler like an Arctic Freezer i11 or a CoolerMaster Hyper 212 Plus is really worth the modest prices they cost. Heck, I bought a low-end Arctic Freezer 7 Pro for the SO's machine running a stock Core i5-2500 non-K. It was one of the least expensive tower heatsinks at the time (and on sale). The difference between stock and the F7P is noticeable. The i11 is $23 plus a buck shipping at the `Egg, and should be a better cooler for less than the F7P, and will allow the chip to run significantly cooler, and you get MX-4 thermal compound with it, also better. It's a bargain.

http://www.techpowerup.com/196918/arcti ... olers.html

LoneWolf15 wrote:

Forge wrote:

UnfriendlyFire wrote:

If there were the resources, someone could design a CPU that has only the very basic functions and let the mobo have the north/southbridge, memory controller, L2 cache, and etc.

Watch that motherboard's cost skyrocket.

For a long time, all motherboards were that way. Welcome back to Socket 7!

Yeah, and I remember (like you, probably) how moving the memory controller and cache to the CPU improved both performance and reliability.

Ahh, the 256k (upgradeable to 512k) and then 512k 486 mainboards across eight chips. And then the Pentium boards with 256k or 512k of pipeline burst-cache on a fingered module. I remember having the occasional cache chip or module go bad and result in some wild, weird stuff.

This is a worthwhile read about some fraud committed regarding motherboards with fake cache, generic chipsets with fake labels, and retailers not paying sales taxes: http://redhill.net.au/b/b-bad.html

Quotes from the webpage:

"Want to buy some 486 boards?"

"What price?"

"If you take 100 pieces, $76."

"OK. What chipset are they?"

"What chipset would you like?"


"That made it one of the cheapest boards on the market, but it still should have been costing them somewhere around AU$120 a piece to land boards in Australia. They were already the poorest quality motherboard you could buy, but how did they manage to make a profit selling them for AU$80?

The answer is very, very simple: fraud."


EDIT: On a side note, I should mention again on an old Core 2 laptop, I undervolted the CPU, then later replaced the thermal paste and cleaned the heatsink.

With the lowered operating temperature, I was able to lower the voltage on some of the turbo levels by .0125V even further.

Forge, you got me to do more playing around with the system and now I've got my system down to 1.225V core at 4.3GHz and it's stable for over 4 hours of Prime95. I can probably take it down further if I keep going (I was at 1.25 Volts from my setting and 1.3volts from the default OCGenie overclock.)


That keeps the temps down at a high of 81C during the Prime95 run and sitting around 37-38C when idle. One other little thing is I turned on Speedstep which MSI's OCGenie turns off by default. I see no reason to run at 4.3GHz all the time since most of the time the system is close to idle.

UnfriendlyFire wrote:

LoneWolf15 wrote:

Yeah, and I remember (like you, probably) how moving the memory controller and cache to the CPU improved both performance and reliability.

Ahh, the 256k (upgradeable to 512k) and then 512k 486 mainboards across eight chips. And then the Pentium boards with 256k or 512k of pipeline burst-cache on a fingered module. I remember having the occasional cache chip or module go bad and result in some wild, weird stuff.

This is a worthwhile read about some fraud committed regarding motherboards with fake cache, generic chipsets with fake labels, and retailers not paying sales taxes: http://redhill.net.au/b/b-bad.html

Quotes from the webpage:

"Want to buy some 486 boards?"

"What price?"

"If you take 100 pieces, $76."

"OK. What chipset are they?"

"What chipset would you like?"


"That made it one of the cheapest boards on the market, but it still should have been costing them somewhere around AU$120 a piece to land boards in Australia. They were already the poorest quality motherboard you could buy, but how did they manage to make a profit selling them for AU$80?

The answer is very, very simple: fraud."


EDIT: On a side note, I should mention again on an old Core 2 laptop, I undervolted the CPU, then later replaced the thermal paste and cleaned the heatsink.

With the lowered operating temperature, I was able to lower the voltage on some of the turbo levels by .0125V even further.

Ahh, the memories...my boss (cheapskate) bought boards like that at our PC shop to build bargain systems. After an astronomical defect/instability rate, we beat him up on it and only sold AOpen and ASUS boards after that. A bargain isn't a bargain if you end up getting them all back and have to replace them with better boards. Ours all said UMC.

EDIT: P.S. Always remember where PC Chips went...they became part of ECS/Elitegroup. And then think about whether you want to buy an ECS board. We now return you to your regularly scheduled topic on the Pentium G3258 Anniversary edition.

LoneWolf15 wrote:

EDIT: P.S. Always remember where PC Chips went...they became part of ECS/Elitegroup. And then think about whether you want to buy an ECS board.

The ECS K7S5A was legen... wait for it.... dary.

Flying Fox wrote:

LoneWolf15 wrote:

EDIT: P.S. Always remember where PC Chips went...they became part of ECS/Elitegroup. And then think about whether you want to buy an ECS board.

The ECS K7S5A was legen... wait for it.... dary.

Had a few of them. So was the RMA percentage. Good design, bad implementation; for anybody that owned one good one, that wasn't noticeable, but for someone who saw a bunch of them fail, it showed.

LoneWolf15 wrote:

Flying Fox wrote:

LoneWolf15 wrote:

EDIT: P.S. Always remember where PC Chips went...they became part of ECS/Elitegroup. And then think about whether you want to buy an ECS board.

The ECS K7S5A was legen... wait for it.... dary.

Had a few of them. So was the RMA percentage. Good design, bad implementation; for anybody that owned one good one, that wasn't noticeable, but for someone who saw a bunch of them fail, it showed.

i had 2 at one time (built 2 pcs one for me and one for my brother) that was DOA right out of the box, the replacements was also DOA - the computer shop was so suss on us they checked the replacements replacement before we got them and although they wasnt DOA they wasnt working correctly (only recognized memory in one slot of one mobo and the usb ports in another wasnt working)

the replacement replacement replacement was working fine but i diddnt trust it so i sold it and bought a different board instead

we probably just had the worst luck ever but since that time i couldnt trust ECS

It's a common story.
PC Chips (purchased by ECS) had RMA rates of over 50%
My local shop stocked them because they were so cheap they were popular, but they said they'd get a pallette of boards and test them all on the spot, sending back dozens before they even made the store shelf.
I made the mistake of buying one and it failed in a week, so I bought a Chaintech board that lasted about 8 years before being retired gracefully.

Since getting mine and using OC Genie, it's ran trouble-free at 4.3GHz. I've played around with the Command Center App to overclock on the fly, which is kinda neat.