Question about Uncore when overclocking the G3258

So I'm in the process of overclocking this poor excuse for a piece of silicon G5328 on a Gigabyte Z97 chip set motherboard. Running a beta bios which I installed in order to run Devils Canyon CPUs, though I realize the G3258 is not a DC CPU. The 4790k I just borked was, however.

Here's the issue. When I raise the core multiplier as far as 38x the uncore ratio stays in lockstep with it. As soon as I bump the cores to 39x the uncore ratio reverts to 32x. Is this normal? Until I noticed that was happening I was having all kinds of headaches overclocking this thing. So at a core multiplier of 39x and over I manually am setting the uncore ratio to 38x and am able to progress once again in the overclock. This reminds me of overclocking the AM3 Phenom Thuban core CPUs which did not like it when the CPU/NB got too far out of sync with the frequency of the cores.

Try to keep uncore about 300-400MHz below core clock. It won't affect performance much but when it's too close to core clock then you can notice stability issues. The same rules as on all 1150 haswells.

Woomack said:

Try to keep uncore about 300-400MHz below core clock. It won't affect performance much but when it's too close to core clock then you can notice stability issues. The same rules as on all 1150 haswells.

Click to expand...

Maybe I wasn't clear in my original post but my experience is the opposite on this G3258. That is when the separation gets too large between core and uncore frequencies, that's seems to be when I get the instability. But there must be a limit on how fast the uncore can run without creating instability as I don't think it can keep up with the core frequency once you get to a certain overclock level. And as far as rules for overclocking Haswells I've read the stickys and the threads of many forums and there doesn't seem to be a lot of agreement. Each chip behaves so differently.

I notice on my 4790K I can run the multi @ 47, but uncore multi @ 45. Any attempt to get the uncore to match core speed will result in instability for me. Unsure if this will happen to your G3258.

Every board is setting different ratio after OC so when you are overclocking then you keep uncore low ( ~32-39 ) or manually at about 200-400MHz lower clock than the cores. Usually with higher core clock difference between cores and uncore has to be higher. Like at 4GHz you can keep uncore ~3.8GHz but at 5.5GHz sometimes it won't run stable above 5GHz uncore.
If CPU is unstable when you set uncore at ~32-39 or auto and cores 4.5GHz+ then something isn't right with BIOS. Uncore at auto is usually within this limit ( 32-39 ) even if you set 5GHz cpu clock.
Pentium G3258 is exactly the same as any other haswell. There are better or worse chips but OC rules are not changing.

My 3258 set the uncore to 5Ghz when I was around 5.8Ghz+ with it (on LN2). I was unable to raise my memory past 1333 without lowering the cache to 4.5GHz which worked perfectly fine (with a 1.3Ghz difference). I really think the cache has to be XXX Mhz apart is hooey. Through 2 4790K's, a 5820K, and now this 3258, I have never kept it close.

Why do I experience the opposite of what all these guides tell you?

ED, I tend to agree with you. I have found the guides for overclocking Haswells to be of limited help. There are so many variables: Different boards, different bioses, differences in silicone. Things were more consistent in the AMD world it seems to me.

All guides will have difference nomenclature in them because its based off of the board they had at the time (or some list different names for things even when its generic). Different BIOS' don't make a huge difference how to overclock, and the silicon is always a variable.

To me, its not so much that (limited help) as to find some of the stuff in them is not true. Particularly the cache being within XXX MHz. Not once was that a requirement for stability for my purposes. Be it ambient overclocking on air/water, or even LN2.

Here is 800Mhz - http://hwbot.org/submission/2862007_
900 MHz - http://hwbot.org/submission/2862007_
1.8GHz - http://hwbot.org/submission/2862014_

So far I tested maybe 30+ haswells and I never had any problems with uncore up to at least 5.1GHz but I'm setting everything manually. Auto is sometimes weird. Still in all cases cpu clock has to be higher by about 200-400MHz than uncore and I think it's in most OC guides regarding haswell. Uncore can be much lower than cpu clock ( like ED linked 1.8GHz or even more ).

Most my chips couldn't run at more than 5.1-5.2GHz uncore with cpu clock up to 5.9GHz. My current G3258 is working up to 5.1GHz uncore when cpu clock is 5.5-5.6GHz. 5.2GHz uncore wasn't really stable and I don't want to push it at too high voltages as it's not that bad comparing to my previous chips and with my luck it can die or degrade.

here is my last 4790K [email protected] , cache @5.3GHz


Btw haswell-e is something totally different, you can run uncore much higher than cpu clock and it can be still stable.

here is one result from XTU competition CPU @4GHZ , cache @4.75GHz on water as I remember

Great info, but not what I am saying (in fact I am saying the opposite)...

I don't have to be anywhere close to the clock speed with my cache to be stable, like nearly every guide seems to mention. See hwbot links...

I have been, as a hwbot link shows, 1.8GHz away from my CPU speed and was stable enough to complete benchmarks.

Tbh I fully read only one guide which I found good since the beginning. Most other were really bad or only average. Later I was testing everything on my own.

Here is one of the good guides:
http://www.overclock.net/t/1401976/the-gigabyte-z87-haswell-overclocking-oc-guide

Most guides were created by people who base on theory rather than tests/personal experience. Later when you give advice on some other forums then they call you troll etc as their local gurus say something else

Some guides were created by ASUS or based off of information from ASUS and say the same thing...

All I know is nearly every guide says it needs to be within XXX MHz or its not stable. I have plenty of submissions to the contrary.

EDIT: That is the first guide I read that doesn't specifically state XXX MHz or not stable. But hell, if I am 1.8GHz off and stable (enough to run a very stressful benchmark on all cores) is it really worth a mention?

You want the Uncore frequency to be anywhere from 300-500MHz below the CPU Frequency to un-bottleneck the CPU,

Click to expand...

That was a gem there...

In ASUS real overclockers are not writing guides. Only Shamino made a guide for some boards but not general guide about everything. Others at ASUS make guides based on info from Intel and some basic overclocking tests ... in real Intel has only theoretical data and is not helping in overclocking in any way.
There are too many bad guides in the web and later on the forums you see all these guys who think that memory voltage above 1.5V can kill memory controller and memory and are afraid to touch BIOS etc. ( it's only one of the examples ).

The only real rules are:
1. keep uncore at ~300MHz below CPU clock or lower - too close to CPU = stability issues, too far = may affect performance but it's not a rule
2. don't set too high VRIN but it's recommended to keep it 0.4V above vcore, 2V is still fine on water, 2.4V+ can kill cpu or cause much faster degradation ... for anything up to 5.5GHz you shouldn't need more than 2.0V.

Everything else depends from cpu or motherboard. Some boards have problems with BIOS, even these designed for overclocking like GB Z87/97X SOC. If they can't make perfect BIOS for their top series then I don't think that everything is working fine in lower series.

On my GA Z87X-UD3H I have to use a beta bios to get "better" overclocking on Devils Canyon CPUs and they seem not interested in developing it any further. That bios been out for a while and still no new bios. There are indications that I was getting the same overclock on my current G3258 with an under $50 3+1 powerphase mATX MSI H81 chip set board and on the same or a little less voltage. And the Z87 Gigabyte has 8+2 powerphase. Maybe I should spring for a decent Z97 based board.

Even 4 phases are enough but without good BIOS you won't make much. I had Z87X OC and later Z97X SOC. Both boards were working exactly the same with haswells even though Z87 wasn't designed for DC. However OC series have usually much longer support and available beta BIOS versions. In regular boards, manufacturers don't care much about improving overclocking etc. You can see that especially on AMD where except 1-2 top models, all other motherboards get 1-2 BIOS releases and manufacturers are forgetting about them. Even ASUS ROG series for AMD got couple of updates in first months after premiere and that's all.

On the other hand UD3 is not low series so it's weird that there are no updates. Recently I got ASRock Z97 OCF and it's a great board. I had GB OC/SOC but after couple of RMA I had enough of them. Yesterday I got info from store that I get money back for my last Z97X-SOC ( 2 months in RMA ).
Sometimes it's good to pay premium just to be sure that you get additional support for longer and that manufacturer actually cares to improve its product.

+1... things are a lot different with Intel, particularly with a dual core CPU.

Well, 4.3 ghz turned out not to be truly stable. So I cut it back to 4.2 ghz but I also discovered that the CPU would run the 1600 mhz RAM at the full 1600 mhz stable. So the CPU is not a total disappointment. Nonetheless, I got another one on the way. Hopefully, it will be a better clocker.

You know, I find it strange that Intel does not make an i3 k series chip. I'm certain if they did they would be in big demand.

Well, I did finally get it stable at 4.3 ghz on 1.34 vcore but I'm not going to run it 24/7 on that much voltage. Obviously, a better piece of silicon would not need that much vcore at 4.3 ghz. I read all these reviews of this chip being overclocked to 4.7-4.8 on pretty low voltage and wonder if those were cherries provided to professional reviewers by Intel.

Most G3258 can make ~4.5GHz 1.35-1.40V. It's the same as with early 4670/4770K. New batches are better but still most are not overclocking good. There was a G3258 thread after these CPUs were released and most forum members were posting not so good results comparing to i5/i7.
As I said some time ago, I had a chance to test about 15 G3258 couple of months ago and two of them could make 4.5GHz 1.30-1.32V. All other required 1.38-1.42V for 4.5GHz. Also one of these two "good" chips had much higher voltage steps above 4.7GHz and on cold couldn't pass 5.2GHz while better one reached 5.8GHz on dice cooling.
G3258 which I have now is from the new batch ( got it maybe 2 months ago ) and is able to pass most tests @4.7GHz 1.3-1.32V. I didn't test longer stability.

Thanks for the input, Woo. New chip arrived yesterday but I've also got a new motherboard on the way. ASRock Z97 Fatalaty Kiiller. Egg had it on sale for under $100 after the rebate. The GA Z87X UD3H I've been using had significant bios issues. And judging from the date of the last bios release, a beta at that, they seem to have moved on. Too bad, it was a really stoutly built board. But the last non-beta board prevented one from increasing the CPU ratio above stock and the beta bios would no allow you to save more than one oveclocking profile. To make any changes stick you had to reset the cmos every time.