New Setup, Need council on OCing. Am I maxed out or can I try diff settings?

[Johan45]

High aweaver, I think it would be good if you were to make a sig like you see in Trents post above your last. I've been out a couple days and had to jump back and forth to your first post so I knew what I was dealing with. I checked your QVL list for the mem/board and it doesn't list any 32 Gb kits at 2133. If it's not too late, I'd exchange them for a 2x4 GB kit of G.Skill in the 1866 range bought as a dual channel pair.

 

[aweaver255]

Was that a yes? You did buy each stick one by one or........?

I bought 2 8GB kits

 

[aweaver255]

High aweaver, I think it would be good if you were to make a sig like you see in Trents post above your last. I've been out a couple days and had to jump back and forth to your first post so I knew what I was dealing with. I checked your QVL list for the mem/board and it doesn't list any 32 Gb kits at 2133. If it's not too late, I'd exchange them for a 2x4 GB kit of G.Skill in the 1866 range bought as a dual channel pair.

Done. Ok so about the RAM. I purchased two of their 2x8GB kits. Crucial sells their 8GB RAM sticks advertised as 2133MHz. The MSi mobo supports up to 2133MHz(OC) The RAM I bought is capable of speeds up to 2133MHz(OC). That doesn't mean I have to run it at that speed, which is why I've tested it running at 1600MHz, Windows and CPUz verifies that speed as well. So far I'm not having compatibility issues with them and I wouldn't expect it have any issues since everything is compatible.

With all of this being said, the first response I received from AMD Support about the FX 8350 was that the RAM Capacity is dependent on the Motherboard. I've written them back asking for a more specific response, but this gives me reason to believe, along with my tests of removing 2 sticks and re-testing stability, that while overclocking in my case it has proved to be an insignificant variable. Once I hear back I'll surely post that info here for everyone's future reference.

 

[EarthDog]

Yes, the board has a say in how much capacity you can use... Your board can allocate up to 32GB. That was never in question.

What is actually in question is if the IMC on your CPU has the ability to run the sticks at such fast speeds (2133) with that much ram (and potentially being mismatched). Typically it cannot which is what we are saying here. Does that make more sense?

Are all the S/N on the ram the same (all 4 sticks)?

 

[Lochekey]

If you see using 2 different kits of ram that can cause issues that are very hard to track down. I know you said you went to 2 stocks but are you sure those sticks were from the same matched kit originally.

Yes if you have not run it already run memtest to verify the ram but even if you pass you could still have compatability issues. If you really want to remove the ram as a variable take it down to one stick and test that way. Just make sure the ram is in the correct slot on the motherboard for single channel use.

 

[aweaver255]

Yes, the board has a say in how much capacity you can use... Your board can allocate up to 32GB. That was never in question.

What is actually in question is if the IMC on your CPU has the ability to run the sticks at such fast speeds (2133) with that much ram (and potentially being mismatched). Typically it cannot which is what we are saying here. Does that make more sense?

Are all the S/N on the ram the same (all 4 sticks)?

Ok let's figure that one out. Does anyone have the spec sheet for AMD's FX Series CPU's? Has anyone been able to find any concrete definition for what the IMC's capabilities are in these FX CPU's? We need to find an answer to that. I'm not even sure it's an issue but hell let's figure it out already. All I can find is:

Memory controller
The number of controllers: 1
Memory channels: 2
Supported memory: DDR3-1866
Maximum memory bandwidth (GB/s): 29.9

Nothing with capacity as of yet...... still researching......

Yes all 4 are the same

 

[EarthDog]

The 'quality' of the IMC will vary. Meaning, if you have the same board/bios/ram, but swap out CPUs, not all will reach the same memory speeds. Again, the more sticks/capacity you have, the more stress that is placed on the memory controller. It is what it is. I don't think I have ever seen a whitepaper that says this CPU will do XXXX speed with XX capacity. But, ALL IMC's WILL VARY in what it can do.

 

[aweaver255]

The 'quality' of the IMC will vary. Meaning, if you have the same board/bios/ram, but swap out CPUs, not all will reach the same memory speeds. Again, the more sticks/capacity you have, the more stress that is placed on the memory controller. It is what it is. I don't think I have ever seen a whitepaper that says this CPU will do XXXX speed with XX capacity. But, ALL IMC's WILL VARY in what it can do.

Understood. Which again is why I set the speeds at 1600MHz and NOT 2133MHz. Agreed more stress may be applied to the IMC with higher capacity but in my opinion if that was an issue, AMD would surely note that, especially in boards that support over 32GB, hence the reason why we're unable to find that sort of info as it seems to be an insignificant variable.

So say I ran a single stick at 1600MHz, I have a Samsung DDR3 compatible, and tried Ocing, according to everyone's suggestions here, I will be able to get a better clock at a lower voltage?

 

[DNRDustin]

In theory, it would at least remove the likelihood of it being a variable, unless something else is going on. Like you have a below average chip, which it seems you do if you need [email protected]

 

[EarthDog]

but in my opinion if that was an issue, AMD would surely note that, especially in boards that support over 32GB, hence the reason why we're unable to find that sort of info as it seems to be an insignificant variable.

That is your opinion...I dont remotely agree with it though. You keep calling it(things) insignificant variables, yet people here, with experience, are telling you otherwise. Now, I don't believe everything I read on the net either, but, come on man, to me it feels like you are challenging everything we say...perhaps I am being too sensitive though or losing site of whatever you are trying to accomplish here.

So say I ran a single stick at 1600MHz, I have a Samsung DDR3 compatible, and tried Ocing, according to everyone's suggestions here, I will be able to get a better clock at a lower voltage?

Memory overclocking? Yes. Not sure what you are asking there though really. But yeah, the less sticks/ranks typically the higher it can go versus more capacity/sticks. It does affect the CPU on the high end too because of all the other added voltages (among other things).

It is possible you just have a dud overclocker. But there is a process to figure that out... which is what these people are trying to help you do.

 

[aweaver255]

I'm not trying to challenge anything honestly. But the same goes here, I can't believe everything I read on the net, especially without supporting evidence in the form of some kind of documentation. I'm basing things solely on what I'm testing with my system. It is possible that I have a dud chip and didn't get as lucky as others for its OCing abilities. But the facts are I've reached a stable 4.5GHz at a voltage within limits and temps that are well below threshold.

If I am able to reach this with all 4 sticks installed I'm happy. I've tested it with 2 sticks and there weren't any differences in voltage levels to make it stable and that alone is enough for a trial and error test to assume it hasn't anything to do with the RAM. I'm half inclined to waste another few hours of my time to prove this by installing the Samsung DDR3 1600MHz and testing, I'd be willing to bet the voltage levels will still need to be up to 1.5v for me to be stable at 4.5GHz. Seems that's just what my particular chip wants.

Regardless, I have the answers to my questions thanks to everyone! I appreciate the help and I am glad to have been able to reach my goal. Good luck in your endeavors. If I get some extra time to fudge around I'll throw that other single RAM stick in and run some tests and post the results. I am waiting for AMD's final response as I did call them as well. They're going to tell me the info for the FX 8350 in regards to it's IMC to see what ratings it has for capacity for single and dual channel setups and what it can handle. I'll def be giving you all the update on that as well for your future posts and knowledge base.

 

[EarthDog]

If I am able to reach this with all 4 sticks installed I'm happy. I've tested it with 2 sticks and there weren't any differences in voltage levels to make it stable and that alone is enough for a trial and error test to assume it hasn't anything to do with the RAM.

If the chip is a dud, there wouldn't really be a way to figure it out with your current setup...

I don't know AMD well enough, but I can tell you from the Intel side of things, its sure as heck wont drop .05V+ or more on the Vcore. Perhaps on whatever controls the IMC on AMD chips though... again, this is really swimming out of my lane, the AMD side of things.

 

[DNRDustin]

Speaking of supporting evidence, we still haven't seen anymore screenshots of settings or stability tests.

Edit. never mind, I'm sure you're good.

 

[trents]

Understood. Which again is why I set the speeds at 1600MHz and NOT 2133MHz. Agreed more stress may be applied to the IMC with higher capacity but in my opinion if that was an issue, AMD would surely note that, especially in boards that support over 32GB, hence the reason why we're unable to find that sort of info as it seems to be an insignificant variable.

So say I ran a single stick at 1600MHz, I have a Samsung DDR3 compatible, and tried Ocing, according to everyone's suggestions here, I will be able to get a better clock at a lower voltage?

This is not likely to be a problem at stock speeds and voltages but when overclocking. I'm sure the FX CPUs will handle 32gb when the CPU is at stock and the RAM is being run within the evnvelope of what the IMC is rated for, which is 1600 mhz. But there can be a performance penalty in running high speed RAM at lower than the XMP rating and that is at lower frequencies the timings may actually be more lax.

 

[aweaver255]

Here's a thought then..... start quote:
1. > When a speed seems too easy...it is probably telliing you something about the future. Hehehe.

2. > With super cooling I have found that at the seeming steps of an FX cpu as you move up in speed, there are general amounts of Vcore that are needed for the next 200Mhz or so. On good water that Vcore add was usually about 0.02V. When I put my bench CHV and my original FX-8350 on BiG Air, It took right at 0.04 for each of those Vcore jumps with 200Mhz or so jump in cpu speed. This is until I got a different CHV and the latest bios available for it.

3. > I knew that there was a problem with this cpu when I had to begin increasing the Vcore in an effort to get 4.6Ghz stable and when I reached 1.49Vcore which is a LOT for an FX 8350 at only 4.6Ghz...well I knew I was in trouble. That Vcore that was nearly 0.1 over the Vcore for 4.3Ghz was a big arse clue that something else besides just ramming voltage at the cpu was necessary.

4. > From above and paying attention to others setups and my extensive log of cpus I have tested, it came to me that it was time to drop the multiplier and bring my speed up by upping the FSB/HT REF FREQ. This sort of tactic has been said to maybe get a little more cpu speed without upping the Vcore. Since i was fairly sure this Air Cooling experiment was going to dead-end at about 4.6Ghz; it was surely time to employ the drop multiplier and up FSB tactic. So I did.

5. > I dropped the multiplier by one full number and added enough FSB to bring me back to 4.6Ghz. I had to monitor ram speed and CPU_NB, and HT Link speed and reduce them as necessary to keep them within supposed safe speeds. Around 2600Mhz on CPU_NB and HT Link Speed and of course ram close to its' rating. Well just shett if it did not fail faster in OCCT.

6, > Fail faster? What changed. Okay I wound up with a slightly faster CPU_NB speed. HT Link Speed can make a differnce but not as much as dorked CPU_NB and Overclocked ram can make trouble. However the full multipliers for Ram, CPU_NB and HT Link Speed would not allow me to set those three busses at the speed I desired so I lowered the multiplier some more and banged up the FSB some more. Well this allowed me to have a lower CPU_NB and underclocked ram and OCCT seemed to run a little longer before failing. WTFO?

7. Okay now stop and think Think on what has been seen with some FX cpus. Then it was a bolt of lightning or similar. Some of these FX processors just don't seem to be able to run high CPU_NB speed without a LOT of CPU_NB voltage and I am trying to keep the temps down on this Air Cooled exercise. So I am loathe to try more than about 1.27V on the CPU_NB.

8. > So what is the next logical step? REduce the CPU_NB speed. Drop it one multiplier below the speed of the HT Link Speed. I don't want more heat and I do want CPU Speed. When all else fails...get cpu speed. Hehehe.

9. > So I dropped the CPU_NB by one multiplier making it less in speed than the HT Link Speed. Shett shett. Staying with my moves and keeping them in logical order seems to have won the day. I had to stop and think about what can happen with a particular FX if you don't get one that is super or not even as good as a previous one that was used. You have to stop and realize that more voltage is not helping and then think. That is why I keep logs of the most important things I have come across. I did not have to make a post asking for help. Thinking and moving steadily in a direction until I see it is not going to succeed. I do nOt go to thrashing around changing this or that or the other so that I am lost. That is why I made a baseline test to begin with. I had an idea how the cpu was actiing from that baseline test.

10. > So now with the CPU_NB lagging the HT Link Speed and enough FSB to bring the ram back up to its' speed rating and without having to add anymore CPU_NB Voltage and with lowered Vcore...I am now running OCCT at 4.558Ghz and for 1hr 20mins without failure. If I had listened to everything that says CPU_NB and HT Link Speed must be the same, I would still be failing. This cpu seems not to have such a good IMC. I had to pursue the raised Vcore until I just said, crap I am not adding anymore just to get hotter and still fail OCCT testing.

11. > You must make decisions that are for the best as CPU Speed is King. Surely I would love more CPU_NB as it helps performance but I am not willing or able with current cooling to just keep upping the CPU_NB voltage to try and raise the CPU_NB speed. Now that I have diagnosed this particular CPU as not having as good an IMC as my other one AND being on a different CHV, I am guessing that I might n fact reach my 4.6Ghz target speed. But it is just that; a target speed. 40 more Mhz is not going to do all that much and I am not going to let 40Mhz put a bad taste in my mouth. I am not depending on what somebody else has done but on what living and learning with my current setup has shown to be realistic and also real.

12. > I had temps in reason and had reached the point where Vcore was not bringing an improvement so it was time to check all the other bus speeds and adjust according to how some few of the lesser FX processors might and can act. I feel am learning more as I have to try weak air and a cpu that is not but about 70% as good as the one I have used up to now. I resisted the urge to go with some water and it has showed me once again that the whole seup thiing is a trade-off of speeds and voltages. Briing them all within the range of a particular cpu and things get more fun and less time consuming. The jump from 4.3Ghz to nearly 4.6Ghz has taken at least 1.5hours of methodical testing, but the results and feel of the results are a good thing.

Luck to all that get down to tweaking their own rigs. Nothing like that feeling of accomplishment when the computer world gets righted on its' axis again.
RGone...ster.

End quote.

Thoughts?

 

[caddi daddi]

pretty simple, no two cpu's act exactly alike, you have to explore the limits of each one.
that cpu and motherboard combo are now a 5ghz, 8 core file server and even under big water the cpu/nb take a major voltage step (leap) to break 2600 and the fsb has yet to break 308 and boot.

I need to correct that, the cpu is in the file server but in a crosshair5-f-z board, the ch5-f is pushing a thuban right now.

 

[caddi daddi]

what operating system are you using?
I just had a real wrestling match with a processor that I have used many times and we know each other very well.
the issue was a higher vcore for a known clock and not being able to use the rig on a kvm, would not pick up the keyboard and mouse anytime the cpu/nb was set to over 2400.
the cure was to give win ten the boot (pun intended) and reinstall win 7 and the rig is back to its old self again.

 

[aweaver255]

what operating system are you using?
I just had a real wrestling match with a processor that I have used many times and we know each other very well.
the issue was a higher vcore for a known clock and not being able to use the rig on a kvm, would not pick up the keyboard and mouse anytime the cpu/nb was set to over 2400.
the cure was to give win ten the boot (pun intended) and reinstall win 7 and the rig is back to its old self again.

I am on win 10. In my earlier tests I tried for 4.4GHz with just the multiplier. Found I could reach that at 1.52v. So I tried the same. I tried dropping the multiplier back down to x20, setting the CPU FSB to 220, my NB freq at 11 which made it 2420MHz, set the DRAM at 1866MHz which adjusted the DRAM freq to 2052MHz, HT link speed set at 12 which adjusted it to 2640MHz. CPU voltage kept at 1.52v with the CPU-NB voltage at 1.3v and the DRAM voltage set at 1.64v. NB voltage I bumped to 1.2v from 1.1v. Default CPU-NB speed is 2200MHz, default HT link speed is 2600MHz. That was as close as I could get before I came here. Ran it prime 95 for a while and said shoot, there's gotta be something going on, I shouldn't hafta go that high on the CPU voltages.

 

[aweaver255]

Just a thought.... Am I crazy to think that the IMC can handle as much voltage as the Cores just because it's all the same silicon? I'm still within the safe zones all around and am peaked at 4.5GHz. Unfortunately though, in my opinion, I feel like I should keep it as is and within this window as I feel there really isn't much else I can do lol. I suppose I had a higher expectation that's all haha

 

[Silver_Pharaoh]

Even I wouldn't push 1.6v into the CPU/NB (IMC) Thats way to high. I never push more than 1.45v. Even then thats a lot of volts and not really necessary.