• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

My 1st serious overclock... with an arguably questionable stability (newest Prime95 displays issues)

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

MiracleMatter

Registered
Joined
Aug 23, 2022
Hi, as my first post ever, I'm gonna try to remain concise because a lot happened. So... to cut through the chase I'm now at 4.4 GHz (CPU ratio only) on my FX-6300 with just the boost voltage and some good LLC (Load-Line Calibration) tuning but as my first committed overclock implies, I'm really not sure how to guarantee stability. Previous mistuning of the LLC to "high" instead of "ultra-high" failed after a long while (I don't recall exactly, 1+ hour ?) the stress test from the latest Prime95 version (30.8).

As I'm trying to figure out if my system is truly stable, I searched a lot for advices on which tests to run and basically came roughly to the same conclusion as this guy but with somewhat different tools (and far less confidence): https://www.overclock.net/threads/stability-testing-in-2019-the-right-way.1722272/
I now have installed OCCT, RealBench, AIDA64 Extreme, CPU-Z, Linpack Xtreme and of course Prime95 ! I genuinely don't recall exactly how much I ran each software on my current config so this is approximate: I did very little of AIDA64 and just minutes of CPU-Z as I don't know how they are supposed to report hardware errors and have been told somewhere they aren't good for the job. I cumulatively ran more than 2 hours of Linpack Xtreme, maybe 2.5 or 3 hours of OCCT in three separate occasions (I think) and maybe a total of 45 minutes of RealBench. OCCT was the one that caught instability at lower voltages iirc (I used it first in most situations). My current 4.4 GHz config passed ALL these stability tests...

... except for ONE ! The latest version of Prime95 crashed with an error reported but only under very dubious and questionable circumstances !

So the first thing I need to explain is before that happened, I noticed something pretty crazy with this version of Prime95 (that turns out to not be the case with the only other version I tried (v28.7) btw): If I select a "custom" stress test and pick a value between 24576 and 24581 MB with a large enough blend of FFT sizes (max can be just below 480K for convenience), any FFT size at or below 160k or so fails immediately ! I didn't check the exact limit though, 256k didn't have this problem iirc. And the best part is that happened to me regardless of the clock settings, it even happened at stock settings with a massive 25 % underclock of the base clock (150 MHz). I don't know what to make of this. Is it my rig that is flawed in an irreparable way or the latest version of Prime95 software that is with this oddly specific amount of RAM. I know that probably sounds like blasphemy to dare question the almighty Prime95 but... Can someone reproduce this problem in his own rig please ?

The second thing that makes me question whether this version of Prime95 found instability in my rig is that the first time I launched it for an overnight stress test. I found out the next morning that the Prime95 software - not the entire OS - crashed but on the washed out Prime95 window, I could see that a single error was reported last. Did it actually found instability or something went completely haywire with the software... again ? So I chose yesterday to search for and download older versions of Prime95 I could find on the web and I ran so many hours of the version 28.7 (from 2015 - 3 years after the release of my CPU in case that's relevant). I can confirm none of the issues I experienced with the newer version happened. I ran version 28.7 for several hours during the day and 9 more hours during the night, no problem, no error reported !

So the million dollars question now is "Is my rig stable enough !?" I'm running it right now with the O.C. in question since the weekend without issues during work days (and play nights in ChilloutVR).
I have some more questions but my post is already enormous as is so let's just focus on the most important for now.

Rig:
AMD FX-6300 @ 4.4+ GHz with no more than 1.4 volt (199.4~202.0 MHz base clock)
Cooled by an OCZ Vendetta 2 (and a gigantic A/C on wheels in the room :cool:)
M5A99FX PRO R2.0
4 x 8 GB (Patriot PSD316G1600KH) - And yes, this new RAM completed well over 4 full passes of Memtest86+ flawlessly
2 x AMD Radeon R9 280 (no longer used in crossfire, my current games don't support it anymore :()
A Corsair TX750W PSU

P.S. If I should do some O.C. "back off" for stability, what would you experts recommend, -1 % of the base clock or +20 mV on core voltage ? Or both ? Which is better ? I really don't want to age up my CPU too quickly. Also, which one property between voltage and temperature ages the CPU more quickly ?
 
Last edited:

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
"Is my rig stable enough !?"
Well, that's an interpretation for each user, IMO. I just run AIDA64 stress test (CPU/FPU/Cache/Memory) for a couple of hours and if it passes, it's good for how I use my PC. I don't custom this, custom that, DL 800 programs.. nothing like that. For me and my uses (gaming, some photo editing)... admittedly light, it's fine.
 

Blaylock

"That Backfired" Senior Member
Joined
Jun 5, 2013
Location
Go Blue!
My biggest question for you is what are you using this rig for? If it's for casual gaming of 8 hours a day and it's stable for 9-10 hours then I'd say you are all set.

As you have noticed, your system could be perfectly stable under 3-4 different tests but one specific one likes to crash it. This is pretty typical actually. Is your rig stable enough? That depends on your usage and if it's stable when using those applications. Generally speaking, if your system can survive 8 hours of testing, it is stable.

Side note. You write about your voltages but never mention your temperatures. FX CPUs are hot running chips. The Cooler you have is quite small for that CPU. I understand you have an A/C unit in the room, but if the cooler is unable to transfer the heat out of the CPU the A/C unit will not be able to help. Also, the FX CPUs are notorious for running very high socket temps. placing a fan directly over your VRM heatsink and one at the back of your motherboard at the socket can help tons. IIRC the maximum CPU temp should be 55C 61°C with the socket kept under 61C 71°C. Anything above this can and will cause instability.
 
Last edited:
OP
MiracleMatter

MiracleMatter

Registered
Joined
Aug 23, 2022
As you have noticed, your system could be perfectly stable under 3-4 different tests but one specific one likes to crash it. This is pretty typical actually. Is your rig stable enough? That depends on your usage and if it's stable when using those applications. Generally speaking, if your system can survive 8 hours of testing, it is stable.
I ran another overnight test with the older version of Prime95. A good range of "in-place" small FFTs this time and my CPU sustained it without flinching for over 7 hours. I think I'm declaring my CPU victorious at this point based on this and the replies I got around.

Side note. You write about your voltages but never mention your temperatures. FX CPUs are hot running chips. The Cooler you have is quite small for that CPU. I understand you have an A/C unit in the room, but if the cooler is unable to transfer the heat out of the CPU the A/C unit will not be able to help. Also, the FX CPUs are notorious for running very high socket temps. placing a fan directly over your VRM heatsink and one at the back of your motherboard at the socket can help tons. IIRC the maximum CPU temp should be 55C with the socket kept under 61C. Anything above this can and will cause instability.

Sure ! I never mentioned temperatures because I knew they never were a problem. However, that being said, you're the first one I see saying the temperature limit should be this low. The peak temp recorded during all these days my CPU was overclocked and stress tested is 62.1 degrees CPU and 60 degrees socket according to HWiNFO64. It's that low mainly because that powerful A/C is aimed straight into the computer. It really is not good for a human being to be subjected to this cold wind directly for long periods of time anyway so why not aim it at something that definitely benefits from it :giggle:. It still cools down the room efficiently, it just cools my PC first and foremost and thus the lowest temperature recorded is 22 degrees for the socket lol.

(Yeah, my CPU won but it got a big helping hand from my A/C ngl)

That being said, I have another question that directly relates to temperature: the temperature recorded directly in my CPU appears to be unreliable because it changes erratically especially at low temperature. As a consequence, I've been relying on socket temperature so if I want to extrapolate the temperature of my CPU from my socket temp, how many degrees should I add (or subtract) to the socket temp approximately ?
 

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
Temperature sensors on CPUs are most accurate the closer they are to the maximum. At idle, they can be off. It's also quite normal for significant temperature fluctuations at idle. Load temperatures are the primary concern anyway. ;)
 

Blaylock

"That Backfired" Senior Member
Joined
Jun 5, 2013
Location
Go Blue!
The max temperatures I posted were off memory. It seems my memory isn't spot-on (who knew). After a quick Google, the Max temps are actually 61°C for the CPU and 71°C for the socket. It seems I was a bit off.

Either way, I'm glad you got it handled.
 
OP
MiracleMatter

MiracleMatter

Registered
Joined
Aug 23, 2022
Either way, I'm glad you got it handled.
Yeah, it's looking great now (so far)... keeping my fingers crossed. It seems to me like the M5A99FX PRO is a good mobo for that.

However, last gaming session the other day, I noticed that - again - the FPS were low without using close to 100 % any of the CPU or GPU activity or the RAM. I think it might be due to the puny 3 GB of VRAM that is overflowing into the regular RAM. If that is the case, can that be helped by overclocking the north bridge ? Or the DDR3 RAM ? Both or neither ?
 

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
he FPS were low without using close to 100 % any of the CPU or GPU activity or the RAM.
That old CPU should probably be pegged at 100% in any modern 1080p game. As far as system RAM, that's filled on an as-needed basis so if you're using 100% of that, you need more.
 
OP
MiracleMatter

MiracleMatter

Registered
Joined
Aug 23, 2022
That old CPU should probably be pegged at 100% in any modern 1080p game. As far as system RAM, that's filled on an as-needed basis so if you're using 100% of that, you need more.
What I meant is none of these components appear to be the bottleneck. I am not playing intensive AAA games but the one I'm invested in right now runs short in VRAM (system RAM is plentiful, check again, I have 32 GB). So in light of that, what should I do to ameliorate the performances assuming the GPU is using the system RAM too and assuming that is where the bottleneck is ? Is it a reasonable assumption ?
 

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
(system RAM is plentiful, check again, I have 32 GB).
I just mentioned that if it's near capacity, you need more. ;)

Ameliorate........ I learned something today... :attn:

Anyway, I don't know offhand if increasing memory bandwidth/NB would improve gaming, sorry. Makes sense it should, but by how much and if it's worth the effort of tweaking, no idea. It's time to save for new hardware that I do know. :)

Your GPU should run at 99%, if it isn't and you're gaming at 1080p (or less), it's CPU bound. You can get around that by turning up the graphics, but you have a limited amount of vRAM.

What game are you playing that's using more than 3GB at 1080p? They certainly exist, but, many don't and you aren't playing AAA titles so.... there's that to confirm if you haven't already. Are there games where you aren't banging off the vRAM limit? What is your GPU use there?
 
OP
MiracleMatter

MiracleMatter

Registered
Joined
Aug 23, 2022
What game are you playing that's using more than 3GB at 1080p? They certainly exist, but, many don't and you aren't playing AAA titles so.... there's that to confirm if you haven't already. Are there games where you aren't banging off the vRAM limit? What is your GPU use there?
I don't know about that. I'm an old gamer with old games that's out of the loop and haven't played for a long while because of my job, thesis and everything else... except for this rather "indie" game called ChilloutVR and its community who I'm spending more and more time with. I don't have VR yet and it's not happening this year either with the current price of silicon !! But that's beside the point, CVR is a well optimized game performance wise but it is rather (V)RAM hungry. The more populated the lobby, the hungrier it gets but the most the system RAM got filled up is to 22-23 GB with approx. 50 people. VRAM thought is lacking.
 

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
I don't know about that.
About what? I made a couple of points up there... even in what you quoted. :)

Are there any games you play that don't bang off the vRAM limit? What are the usage % then?

A game that regularly uses 20+ GB of RAM today is a bit insane, especially considering most users are around 8-16GB total. I wouldn't call that part well optimized at all, lol. I DL'd the game (~980MB) to see what it's all about. I don't have VR and joined some world with 10 people. RAM use went from 11GB to 12GB after starting the game and ending it. The CPU used a few cores and threads and the GPU used 25% (this is at 2560x1440). :)
 
OP
MiracleMatter

MiracleMatter

Registered
Joined
Aug 23, 2022
About what? I made a couple of points up there... even in what you quoted. :)
I don't know if the other games I played in the past (and that I'm still interested in) have hit the VRAM limit but probably not since I never really noticed or bothered to look. I'm only looking to optimize for CVR as it is right now.

A game that regularly uses 20+ GB of RAM today is a bit insane, especially considering most users are around 8-16GB total. I wouldn't call that part well optimized at all, lol.
If you knew what CVR is, you'd understand. There's a lot of player created content and it only gets that high in enormous lobbies. CVR itself is way better performing than it's main competitor VRchat but that doesn't mean the Unity avatars players upload are optimized too or made to be lightweight in memory.

Anyway, we can't argue against the facts. There are common situations where neither the CPU, GPU or RAM are hitting their limits (and it's a well multi-threaded software BTW) but the VRAM definitely does so I suspect the bottleneck could be data flowing between the system RAM and the GPU as it is.
 

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
As I said, I just played it... I guess it has to be in specific situations you see such significant RAM and vRAM use. I'm curious, so I'll try it during prime time and see if I can get into some more populated rooms/lobbies and see what my system does.

Anyway, we can't argue against the facts. There are common situations where neither the CPU, GPU or RAM are hitting their limits (and it's a well multi-threaded software BTW)
Be careful... in most situations, the GPU is supposed to be pegged at 99%, but more often than not, a CPU isn't. It's going to be situation and game-dependent, but generally, GPU gets pegged and CPU does not.

Maybe more cores light up with more people, but this seemed to only tickle a couple of threads on my system...That said, your CPU is quite old and doesn't have a of cores and threads, so I can see why you feel it's heavily multi-threaded.

EDIT: Oh yes, situation-dependent. It's lighting up a few cores/threads... Da..!!

vRAM in this area was 2.5-3GB. CPU use is 30% with 12900K. :)

EDIT: Have you checked out this guide of ours yet?
 
Last edited:
OP
MiracleMatter

MiracleMatter

Registered
Joined
Aug 23, 2022
Hi, let's say I've achieved an overclock I want to settle with. How would I go about re-enabling the voltage-lowering function there is by default in my motherboard when the CPU becomes idle or under a very light load ? I'd like to expand the lifespan of my CPU as much as possible.

As it is, I achieved a stable overclock of 4.4 GHz at 1.4 V with cool'n'quiet enabled the whole time. I like that this feature keeps the frequency low under light or no load but it doesn't do the same for the voltage which always stays millivolts away from 1.4 V. I tried enabling every other power-saving features except the C6 thingy but my voltage didn't budge.

How would I go about doing this... and should I even do this or will it cause instability that'll be hard to detect ?

Rig:
AMD FX-6300 @ 4.4+ GHz with no more than 1.4 volt (199.4~202.0 MHz base clock)
Cooled by an OCZ Vendetta 2 (and a gigantic A/C on wheels in the room)
M5A99FX PRO R2.0
4 x 8 GB (Patriot PSD316G1600KH)
2 x AMD Radeon R9 280
A Corsair TX750W PSU
 

Nebulous

Señor Senior, Senior
Joined
Oct 11, 2002
Location
The Empire State
From what I remember, the motherboard should have have the C-states enabled which are the power saving features. So when the cpu is not running full bore overclocked, it should downclock the cpu which results in lower power consumption.

Also that cpu is quite old so you can pick up replacements really cheap. https://www.ebay.com/sch/i.html?_from=R40&_nkw=amd+fx+6300&_sacat=0&_sop=15
 
OP
MiracleMatter

MiracleMatter

Registered
Joined
Aug 23, 2022
Fair enough, but I still would like to avoid the potentially critical downtime and trouble of replacing the part if things go wrong. I already have the underclock when saving power. What I am asking for is the corresponding undervolting.
 

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
Depends on how you overclocked. If you used a fixed voltage, then it won't drop. If you used offset or PBO, make sure windows is set to balanced and all your power saving is enabled.

Also, undervolting and letting the chip's voltage drop on idle are two different things. Yes, the voltage drops, but undervolting most would think is manually adjusting the voltage lower than stock.

PS - I merged threads since this is still about your overclock and the information in it could be helpful to answer the question (like how you overclocked, exactly). ) :)

PPS - I wouldn't worry too much about undervolting it... just let the power savings do its thing. Also, have a spare CPU handy if you're really concerned about minimizing downtime. They are incredibly cheap as they are quite old and, not terribly performant (even when it was launched 10+ years ago) so there are tons available dirt cheap. Spend $5 getting a backup not hours undervolting in this case.
 
Last edited:
OP
MiracleMatter

MiracleMatter

Registered
Joined
Aug 23, 2022
Ok, sorry. I thought separate questions on separate threads were more appropriate since a future search with the search engine would make them easier to find.

I did set my voltage to "manual" instead of "offset" on my mobo. If that's is why the voltage always stays the same despite power saving features, that's a little unfortunate. I was told "Manual" is much preferable for overclock stability and such. What would be the downsides of using the "offset" voltage type based on your expertise ?