FX8350 overclocking performance improving with time?

Hey guys,

I apologize to begin with for the long post but I think to explain the situation fully it’s required.

I have recently noticed something really strange going on over the last two weeks while I've been trying to overclock my 8350. It almost seems like every time I get the spare time to work with the overclock the CPU acts like it is more tolerant to higher clock speeds at the same voltage.

For example, I started out with just pushing the multiplier higher to see how far it would go with the default voltage of 1.32v and initially it failed a 10 minute P95 small FFT test at 4.428 GHz at a multiplier of 21.5. I did try to play with the core voltage to try and make it pass P95 and ran out of time that night to find a voltage to make that clock speed stable. I started again 3 days later and for grins and comparability sake I reran P95 at stock 3.99, 4.212 and 4.320 with the same stock voltage and my temps were down a degree or two but I attributed that to slightly cooler ambient. But when I started to try to find a stable voltage for the 4.428 clock I started at 1.36 or so and continued to back it down and wound up back at 1.32 and a completed run in P95. Out of surprise I ran it three more times without errors and just chalked it up to luck (or lack of the few nights before).

So now, after a few more days I came back to working with the overclock and had a similar situation happen where the initial voltage I found days before for a stable clock speed could again be reduced a little and still pass P95.

Is it really possible that my chip is just getting more tolerant to overclocking? I've had it for a few months and just recently put on a Noctua NH-D14 cooler on it to start working on overclocking but I never expected it to become a better overclocking chip as days passed, I always thought that if it needed 1.35 volts to run at 4.428 GHz one day it would still need that same voltage a week later….

Thanks in advance to any help you can provide to clear up my confusion here.

Honestly, a lot of people have conflicting views. Some people say that break in periods are necessary, others think its a bunch of bull, and then there are others who believe your PC will never be as good as when you get it at first. I'm rocking an 8350 on my build, its a monster of a processor, the highest I managed to get it up to was 4.7GHz. I ramped my up over time, and I actually hit a few unstable frequencies, 4.6 was a no with the settings i used, but 4.7 worked fine. It's all about the quality you received your processor in and the capabilities of your motherboard.

From what I understand you moving to a better cooling solution is most likely why you can do the same with less

Better cooling solution allows for lower volts as there is far less heat / resistance

these days I dont bother pushing past 4.5ghz as I like the thermals / noise / voltages (holy trinity of overclocking :D) as they are.

Skurtov I had read a few places that there may be that break in period that you just mentioned but couldn't seem to find anymore than two or three posts only mentioning it so I wasn't sure if that was going to be the case or not. Thank you for at least confirming that it might be the cause lol. That seems really strange too that you are stable at 4.7 but not at 4.6, Ill have to keep that in mind when I hit a wall.

Flazza I guess I wasnt real clear on adding the NH-D14, I only started to overclock after I had it installed. Before I had actually done a slight under volt on my 8350

There isn't such a thing as a break in period when it comes to a microprocessor, but it can change characteristics somewhat with age, both for better or for worse.

Good example of that was with my old 8800 GTX, when I got it in 2008 it did about 620mhz on the core from 580mhz, you can't change the voltage via software/firmware but after a couple of years of usage it suddenly was fine with a 650mhz overclock.
It could just as easily have gone the other way around as well.

Jayztwocents talked about this during one of his broadcasts; when he hung out with JJ from ASUS, during the the whole X99 launch, that, similar to CPUs degrading from an overclock, they can also "burn in" to certain frequency range.

Doesn't mean it will always be better, however. Just by using a CPU, you're slowly, ever so slowly, destroying it.

The concept of slowly destroying the cpu makes sense to me, especially when overclocking it. I was just really confused by the improved stability. And i guess i havent done full 4+ hour P95 tests so it might maybe have been a fluke but it seemed like if a core was going to fail the test at all in the first 30 mins it often did within the first 3 or 5 minutes after starting

It also seems like my socket temp is my limiting factor. I seem to push past the 70° mark before the core gets to 54°. Any thoughs on keeping the socket cooler? I do have a noctua nff12 blowing behing the mobo already

Stability tests are great arcane mysteries. I remember when I used to be really hardcore about them, running multiple 8-hour test runs to ensure stability. I remember on something like, my fifth run I had a single error. Oh! The horror! Of course, by that time, I just bumped up the voltage one notch and called it good.

Meanwhile, my friend is still sporting his I5-2500k. I can't remember his overclocked settings, but he ran a single test of Prime95 for fifteen minutes and called it good. Hasn't had a BSOD or any crashing, and it's been over a year. So, yeah, I changed my modus operandi since then.

Anyway, as for your question about keeping your socket temperatures down, I can only suggest try messing with your airflow. What case do you have, with what fans, and where? Also, how close is the Noctua NF-F12 to your CPU socket, and is at full RPM?

Wow with just one 15 min test. The more OC guides and experiences I read the more it really seems to come down to chance almost on how well a stabilty test actually judges the long term stability of the machine.

I have an Antec 1100 v2 with 2 stock 120mm intake fans at the front, one stock 120mm exhaust just behind the nh-d14 and then one NF-F12 behind the motherboard as a side intake. I do have room to put two 120s at the top but i wasnt sure if theyd just mess with the airflow over the d14 or actually synergize with it. And the 140mm center fan on my d14 is as low as it will go. Mobo is the asus m5a99fx pro r2.0. Thanks for all the help and advice guys!

On the subject of stability - yep. Great arcane mysteries. Now, obviously, my friend "got lucky" with his test. I don't recommend doing that for anyone, but I have any issues (yet) with doing one-hour tests. shrugs

That should be a pretty decent airflow set-up. Adding more fans probably won't help. Have you tried setting the center 140mm heatsink fan to maximum speeds, along with the behind-mobo-NF-F12?

Well not exactly I have a FX8320 @ 4.5Ghz with a H80i running a 990FX chipset, with 1.5v lol. It could prob be lowered to 1.475v but I noticed that if you don't have a steady beefy stream of power to the vishera core cpu's then they get alittle loopy.

Im pretty sure that I do have the 140 and 120 fans for the nh-d14 running at max all the time since they arent PWM and my motherboard doesnt seem to want to control them via voltage but the nff12 on the back will max at 2000 rpm when the socket hits 66° according to my custom fan curve and pwm's fine off of the cpu opt header.

And i think i recreated the anomaly that I started this thread over. Last night I was running at 4.4GHz and 1.34v going for a 1hr test and bumped it up to 4.5 afterwards and then tried for 4.6 but it wasnt stable so i tried 1.343 and 1.348 or so unsuccessfully. I then jumped it up to 1.365 or so and it was stable for 30 mins and then backed it off with ~20 min test and found it all the way back at 1.343 and it ran clean for 40 mins where i called it quits since it was 3 am lol.

So it almost seems like its more stable if I start high and back off the voltage a bit which would explain the situation the first time too. Is it possible this is a false positive since it wouldnt run stable at that voltage initially but then it will after bringing the voltage down? I guess ill test some more tonight to see if i can run it at the 4.6 at 1.343 from the start or if I have to bring it down to that again for stability.

Not what to say about the phenomenon; might be a fluke, might be a temperature related thing.

As for fans and airflow, if you can somehow replace that rear exhaust with a fan that has better airflow, that might help a bit. Maybe. The idea that I am trying to go for is to remove heat from around the area of your CPU, and if the fans around there are already working their best, perhaps there needs to be a way help remove the hot air from that area.

Ok sounds like a good idea, I'll try to borrow a fan from a friend or pick one up to see if that makes a difference in the socket temps.

And I guess it could be a temperature thing since it always seem to be after I've tested for a while that I can get this to show up.... hmmm more deliberate testing tonight i guess

I have a 9370 locked at 4.5ghz on a h80i... its currently at 1.38 something v and completely stable

i think i could push a fair bit higher while staying under 1.5v but I like the noise and thermals as they currently are.

2 Likes

Yeah thermals aren't really as bad as they sound with 1.5v, noise level is really about about the same though for me any way. Also, your voltage is around stock so that's a decent voltage.

How is the performance of the FX-9370 is it worth the upgrade??

Had the same thing happen to me. Managed 4.7ghz with a 5820k at 1.3v a while back (now it's 1.1v and 4.2 although it could handle 4.3) and right before that I struggled to get anything above 4.4. It happens.

I wouldn't recommend the upgrade unless you don't plan on upgrading when the Zen architecture comes out. I splurged last tax season and got a FX-9590 and a R9-290x from the FX-8350 and HD7870 I already had. Yet, I will be upgrading at least CPU/Motherboard and RAM again when Zen launches, even if it's an Enterprise part with 16 cores 16 SMT threads.