fx-8320 fine tuning my OC

[Trollyy]

Hey everyone. I posted a couple months back regaring overclocking the fx-8320 on a asrock 970 extreme4. I know that its not "wise" considering its a 4+1 power phase on this motherboard. Keeping that in mind I carefully planned out my case airflow and motherboard airflow. I moved my GPU to the secondary pci-e x16 slot as the top pci-e x16 was greatly restricting airflow to all of the problematic hot-spot areas on my motherboard (vrm, nb, socket). This lowered my temps by ~15c. I added a top-front intake mounted in the 5.25 drivebays blowing cool air straight onto the CPU/HSF/Socket and Northbridge. I also mounted a spot fan blowing cool air directly onto the VRM. Directly above the VRM/socket I mounted another 140mm exhaust fan. This has managed to lower all my temps considerably (~20c atleast!) and I think they are below my motherboards thermal limits now. Ambient system temps are 28-30c here in early spring with the window open. Socket temp never goes over 55c and on a nice and cool spring day core temps dont go past 35c. Normally they are about 37-41c full load after running p95 for 1hr ish.

Current OC settings:

in bios ive disabled all throttling settings (CnQ, c1e, c6, apm, turbo)

Multiplier - Manually set to 21.0
vCore bios - manually set to 1.300
NB frequency - manually set to 2200 (default)
CPU/NB voltage - 1.2v
HT Link - Manually set to 2400 (default and max on my motherboard)
LLC setting - manually set to 25% (asrock LLC settings seem to be screwy. This was the only stable one that didnt cause massive vdroop or overvolting)

I also manually set DRAM timings/voltage/frequency.
DRAM voltage - 1.5v
DRAM frequency - 1600mhz

All other voltages are manually set to the default values.

Its stable at 4200mhz after running p95 for 2 hours and I'm now in the process of fine-tuning everything for best performance/temps. Any tips or adjustment suggestions would be greatly appreciated . Also what should my NB frequency set to? default for me is 2200 but this is lower than my HT link. Would upping my NB to 2400mhz give any real-world performance increase? I have also attached cpu-z and hwmonitor screenshot after p95 for 2 hours. Ignore the temp sensor auxillary considering it just goes inbetween 4c and 115c randomly all the time. also vin1 vin4 vin5 vin6 arent attached to anything, just some random readings.

11.68000

 

[RGone]

Trollyy, your temps and voltages all seem well enough if the HWMonitor image you attached was gotten by having HWMonitor open and monitoring during the time that P95 Blend mode was being run.

Cannot see what ram might need adjusting since the CPUID Memory tab is not shown. Likely though for most average DDR3-1600 ram the settings are okay enough.

Most AM3+ motherboards originally had bioses that had the HT Link speed set to 2200Mhz when the AMD spec is actually 2600Mhz. But there is no way the average person would notice any difference whether set to 2000Mhz or 2800Mhz. Benches are the only thing that "might" show any difference in varying HT Link speed and then 'only' with very elevated cpu speed, ram speed and the use of more than one video card.

The same actually holds true in general for CPU_NB speed. Out of 'pure old school' thinking I generally run CPU_NB and HT link speed in lockstep as in they run the same speed. Again faster CPU_NB is not going to really have anything like a visible difference to you except in benches and then usually above DDR3-1866 ram speed and honestly again with more than one video card.

The deal is you can adjust the various bus speeds as you wish and maybe use the CPU Performance test in Cinebench 11.5 or 15.0 and see what varying HT Link and CPU_NB does for your test results. Elevated CPU_NB nearly always requires more CPU_NB voltage than the normal stock voltage of 1.185 or therebouts depending on board maker and bios engineers ideas.

Just honestly, the tweaking deal when running at cpu speeds less than 4.7Ghz and with DDR3 ram running less than DDR3-2000 or so is just not going to suddenly turn my piston powered aircraft into a jet in any particular way. Adjusting is part of the fun but should always be followed by the P95 Blend mode thing for a couple of hours to see if all is still as stable as expected.

So in closing, I would just say to adjust and test to see if there is anything extra you get from adjusting. Test for stability for certain and in so doing the trial and error thing, you actually learn quite a bit about your system.

Good luck and have some fun. Oh by the way, more cpu speed is far more power producing to do work than the items you asked about tweaking. In general the only real adjustment maybe vistible in day to day use or in gaming will be to run the cpu itself faster. Not sure that is overally advisable with 4 + 1 VRMs but there is where you will see some extra ooomph. I tend to notice a difference in how apps complete when I move from 4.3Ghz to say 4.5Ghz and really not again though until I hit 4.8Ghz, so cpu speed to performance increase does not seem to give me the human experience of 'faster' in anything like a linear fashion. YMMV.

RGone...

 

[Trollyy]

Thanks rGone. hwmonitor screenshot was open while p95 blend ran for 2 hours. Even tho its stable it I just wanted to double check that I hadn't overlooked any settings that were terribly important or didnt set something right. never hurts to be a little paranoid right? My day-to-day use is video encoding and 3d rendering and somtimes a bit of streaming games to twitch.tv at 1080p/60fps.

I'll check out cinebench and test out various ht link and cpu/nb settings. the auto settings on my motherboard for HT link defaulted to 2400. when i increase the CPU_NB to 2400 how much of a voltage bump should I give the cpu/nb? also I have read that you should give DRAM voltage a small bump just to make sure its not the stability killer?

Thanks again!

oh and here is that cpu-z memory screenshot..

 

[Silver_Pharaoh]

Can you get the Memory tab too not just the "SPD" tab?
Otherwise I like it. looks pretty nice for a 4+1 phase board.

 

[Mandrake4565]

Can you get the Memory tab too not just the "SPD" tab?
Otherwise I like it. looks pretty nice for a 4+1 phase board.

See post #1 SP

 

[Silver_Pharaoh]

Big 'ol failure on my part.

Nevermind, carry on

 

[RGone]

Tidy up last asked questions...

...From post #1 > CPU/NB voltage - 1.2v

From post #3 > increase the CPU_NB to 2400 how much of a voltage bump should I give the cpu/nb? That will certainly be trial and error since it will be cpu and board dependent and we have not seen more than about 2 of that model of mobo in about 3 years. It may not need any voltage rise or try 1.25V CPU_NB and see.

From post #3 > should give DRAM voltage a small bump just to make sure its not the stability killer? I always run my 1.5V ram at 1.55V or that little boost you ask about.

I am just going to say this from experience. I find that 4.0Ghz is one speed plateau. 4.3Ghz is the "next" plateau. Then 4.5Ghz and finally generally 4.8Ghz cpu speed. By that I mean I can sort of recognize the changes in those speeds and move to the next. Benches sort of give a little extra or make that rise in score. This on Vishera Core FX processors. So if you can get 21.5 multiplier and not have to add much Vcore, which means not much more heat which means...well you might be able to run the, for me, slightly more powerful 4.3Ghz and be there without much more fuss and bother and without too much more strain on that 4 + 1 VRM circuit.

But hey that is just me and having watched this stuff 'and' done it for nearly 3 years.

RGone...

 

[Trollyy]

so i upped the multi to 21.5 and i kept crashing on prime95 blend at 1.300v and 1.315v. I upped the voltage to 1.335 (stock) and im 15 minutes into the p95 blend test. Ill post a screenshot incase I lock up.

 

[RGone]

Certianly nothing wrong with those temps at the time you took screenshots for posting.

RGone...

 

[Trollyy]

I had to revert to my 4.2ghz OC settings. I didnt lock up on voltage 1.325 but I actually ran into this error.

FATAL ERROR: Rounding was 0.5, expected less than 0.4

upped my vcore to 1.335v and p95 was running fine for 30 minutes. I stopped the test manually and had no errors. After restarting my computer and trying to run the test again I ran into the fatal rounding error again within a few minutes.

I tryed a different ram speed of 1333mhz just to make sure that wasnt the problem... then i got a bunch of illegal sumout errors. Any ideas? Should I run a memtest? I noticed in cpu-z my "JEDEC Timing tables" are completely wrong (i think) because I think im missing the jedec timing tables 1-4 for speeds lower than 1600mhz..... Or am i completely mistaken

 

[Mandrake4565]

Trollyy, in your first post you stated that you were 2 hours prime stable at 4.2, if you tested using prime blend then I doubt you're receiving errors due to your memory. Additionally, since you're only running at DDR3 1600 you're not putting a lot of stress on the IMC, you can try and raise the Cpu Nb voltage and see if it helps but I do not feel your issue is the memory. You are most likely receiving errors due to lack of voltage to the cpu, now here lies the issue. Your board having a 4+1 Power Phase Vrm may not be up to the task of providing enough "clean" voltage to the Chip. Therefore you may raise the voltage and it's still unstable due to the fact that it's not clean. Unfortunately, these Fx 8xxx chips require stout motherboards when overclocked which I learned first hand when I purchased my Fx 8350 when they first came out. I was trying to run it on an Asus M5A99X Evo board which has a 6+2 Vrm section and my chip stopped dead in it's tracks at 4.5 ghz. Now it is possible that by giving the Cpu more volts you may find stability at 4.3 but at some point you are most likely going to run out of power with the motherboard and may have already.

RGone spent a lot of time putting together an Info and tip thread here that you may want to read through. It's basically what the group of us that run these Fx 6/8 series processors, have learned since they came out. A lot of us that have them, have pushed them pretty hard for benchmarking, daily usage and real work such as video rendering. We have learned a lot about running them overclocked and two things that really stick out to me are: they require, as I said previously, a stout motherboard when pushed and they need high end cooling.

 

[RGone]

Okay some of the method to my madness...

...perhaps a little madness in wanting to see if 4.3 Ghz was going to be stable and how much more Vcore it might take to get stable along with 'would' it get stable relatively easy. See back in the early days of that series of mobo, the users had a terrible time getting stable the furhter toward 4.5Ghz they went. Around 4.0Ghz was not too much of a problem, but the faster the cpu was pushed the harder gettting stable became.

So now my memory about the EXT 4s is sort of REfreshed. Been so long since we actually had EXT 4 in the forum and did some serious work with and with a user that could do a good bit of the work on his own. Had one here in Trollyy. He went thru ram speed change and all to see if that were a stability issue and I agree with "manny" that I doubt a ram problem. Likely just not enough cpu voltage or it not staying steady and clean to the cpu.

I am fairly certain now that most of those boards did quit around 4.1 to 4.3Ghz and Trollyy you are right dead in the middle there. You have done an excellent job in getting 4.2Ghz stable and I now believe that is where you should likely stay. You "could" continue to up the Vcore to the voltage range that does allow 4.3Ghz to get stable, but personally that may well put too too much load on the VRMs on the board and then after some use...well maybe a mosfet goes boom and takes out the cpu or ram, or video card and it is not likely worth that sort of risk for just 100Mhz more since the voltage seems to take that big jump for 4.3Ghz stable.

Trollyy, I appreciate your skill in testing 4.3Ghz and for bringing my 'remembering' back into more clear focus. 4.2Ghz seems very stable and cool and uses little cpu voltage. For 100Mhz more, it does not seem worth it to push the 4 + 1 VRMs any harder. You did well in testing. Thanks man.

RGone...