I've searched and couldn't find a thread for 3950x owners to post JUST 3950x results. So, here I start this thread, and will post my OC's in a few. Please, all users of 3950x's post up your screenies of OC's!
OFFICIAL 3950x Overclocking thread....
newls1
281 - 300 of 984 Posts
Joined
·
47 Posts
I might be quite lucky then
Every step of the way I've tested with Cb20 and Y-cruncher. The 1.181 load voltage is cb20 load. Y-cruncher pushes voltage down to 1.175. Once I found my limit I also ran the entire Blender Open Data Benchmark suite without issue.
Perhaps my FIT measurement was scewed because the chip was overwhelming my cooler? -its only a 240mm AIO in push/pull configuration. With my current configuration the chip is idling around 33c and in normal high intensity workloads (like video rendering) it will hit 68c-ish temperatures. Blender Open Data benchmark pushes temps to 79c max, and Y cruncher to 85c.
I've run this configuration for days without any problems. And I've been doing a lot mixed work, from browsing, gaming and quite a lot of video rendering.
CB isn't much of a stress tester, but Y-Cruncher should do the job, so yeh I'd say you had one of the best 3950x chips ever made. Well above the top specs for the golden chips SL sells. You could probably flip it for 4 or 5 times what you bought it for, should that interest you (I'd probably just be lazy and keep it, but still).
What cooler are you using?
I suggest doing Realbench for 5 hours, and then Prime95 Small-FFTs with all AVX turned on for another 5. If it passes both then I doubt anything can make it crash. Congratulations!
Joined
·
95 Posts
So I have a question for you all.
I currently am in possession of 2 3950x and I'm trying to decide which to keep. Is there a good and easy way to test which is better?
I compared the core boosts to eachother but the differences are super marginal, it's mostly a matter of one having the first two as the fastest while the other was more like core 4.
I seem to be thermally limited very strangly in that prime95 small fft, doesn't hit my cpu very hard at all, but Aida and Occt do, which is the opposite of how it's always been for me. It's also hard to tell what voltage in running at, as even with LLC 5 I'm getting notable low voltage compared to what I set.
Currently I set it to 1.25v with a 4.3ghz OC all ccx and Aida is at 1.237v according to hwinfo. Prime95 can get to 1.244, but also is a joke in that it doesn't get temps above 70C, while Aida hit 84c. Double checked, prime is the newest. Cinebench seems to drop the most, which likely explains why it crashes the most.
I'm having trouble getting a 4.5ghz ccx with anything that is decent temp wise. So is there a quick and dirty way to
Figure out how good your chips are?
Neither of mine can hit 4.7 in stock config, with PBO on or off, the special power profile or any combination I can think of. The best I've seen is 4.66ghz. so I'm curious if my mobo is like just not good. Asus x570 TUF .
If a Overclock crashes Cinebench is it no good? I've had some fantastic OCs that seemed to run fine outside of CB. Is it fine to use Performance Bias to try and get it to work?
I currently am in possession of 2 3950x and I'm trying to decide which to keep. Is there a good and easy way to test which is better?
I compared the core boosts to eachother but the differences are super marginal, it's mostly a matter of one having the first two as the fastest while the other was more like core 4.
I seem to be thermally limited very strangly in that prime95 small fft, doesn't hit my cpu very hard at all, but Aida and Occt do, which is the opposite of how it's always been for me. It's also hard to tell what voltage in running at, as even with LLC 5 I'm getting notable low voltage compared to what I set.
Currently I set it to 1.25v with a 4.3ghz OC all ccx and Aida is at 1.237v according to hwinfo. Prime95 can get to 1.244, but also is a joke in that it doesn't get temps above 70C, while Aida hit 84c. Double checked, prime is the newest. Cinebench seems to drop the most, which likely explains why it crashes the most.
I'm having trouble getting a 4.5ghz ccx with anything that is decent temp wise. So is there a quick and dirty way to
Figure out how good your chips are?
Neither of mine can hit 4.7 in stock config, with PBO on or off, the special power profile or any combination I can think of. The best I've seen is 4.66ghz. so I'm curious if my mobo is like just not good. Asus x570 TUF .
If a Overclock crashes Cinebench is it no good? I've had some fantastic OCs that seemed to run fine outside of CB. Is it fine to use Performance Bias to try and get it to work?
Joined
·
47 Posts
If you have voltage set to 1.25v, then it should be dropping way lower than 1.237 under Aida, unless you have your LLC set to something very high. At which point the instability could be caused by all the overshooting the chip is doing. But then your temps seem weird too. Prime95 will only act that way if you have all the AVX settings turned off. Thought it's hard to compare when I don't know which AIDA settings you have turned on in the stress test.
4.5ghz on a CCX is extremely rare for 2950x, especially at low voltages. You'd need to be running nearer to 1.3v (or higher) under full load to be likely to manage a stable clock at that speed. But I wouldn't recommend it, personally.
Quick test to see which chip is best... none that I'm aware of. You'd need to run 8hours+ per stress test for each setting option on both chips before you can figure out the maximums they can tolerate. I'd have my chip for almost a month and I still am not finished fine tuning it. If you can set up both chips in two identical rigs so you can run tests simultaneously, that will cut the time in half.
Joined
·
95 Posts
Perhaps I've been wrong on LLC, I figured rising lv5- the highest on my board, would result in getting the voltage I set it to for my benchmarks, which would improve the likelihood of it being stable.
I have all the cpu settings on for Aida, disabled hard drive and you, but kept memory on.
I have a setup of 4.4 ccx one/4.25 the other three, that I've been using for awhile but I wanted to optimize it. It runs at 1.28 bios closer to 1.24 in aida, but it seems like 4.4 requires a lot of voltage as I was able to get a a copy djustiz settings of 4.3, 4.3, 4.2, 4.2@1.2v bios with at least preliminary stability using llc5. I'm sure it'll be unstable, but it clearly shows me that 4.3 needs much less then 4.4 to be stable.
Would be really nice to be able to trust my prime95 runs as useful but when its not burning my cpu, it feels like it's probably not a valid test. It's a good like 10C+ below Aida running the small fft. Looks to be using 130w vs 170w based on hwinfo.
Attached is my prime95 running, with temps. I was getting 75C with Aida right before this. So confused...
Attachments
For me on stock PBO settings, small-FFTs on Prime95 runs at 3.8GHz with voltages hovering around 1.13V. My ram is fully overclocked to 3800MHz CL14-15-13-26. Temps are in the 50C range.
Even a 4GHz CPU overclock crashes for me on Prime95 Small-FFTs, regardless of the voltage or LLC I set in the bios, with temperatures in the 60C range.
It would be very useful if the folks posting 4.2/4.3/4.4GHz CPU overclocks could run Small FFTs and report.
Joined
·
95 Posts
You're using the latest version of prime? 29.8 build 6? It seems we both are having strange behavior with Prime, for me, my temps are way too low, but it does run and doesn't crash. As seen above in the picture I included, there is me using 4.3/4.2 in small FFT. I haven't had a crash in prime like ever, which I think is due to mine just not working properly.
Joined
·
47 Posts
Hang on... you said Prime was running at 1.244? but on that HWINFO you have prime running at 1.167.
Joined
·
47 Posts
When you say 'stock PBO settings', are you talking about PBO being on 'auto' or on 'enabled'?
Cos Auto means off. Enabled turns it on, and increases the limits.
3.8ghz is the stock chip behaviour, turning on PBO (it's off by default) should increase it to around 4.1ghz all-core.
What cooler are you using? If your cooler is being overwhelmed the chip will run slower.
Joined
·
95 Posts
The screenshot was taken while doing the 4.3/4.3/4.2/4.2 setup as I've mostly abandoned trying to get high clocks. The 1.167 was due to lowering LLC to 2, which caused insta crash in prime95.(it also seemed to be too low voltage to get full performance as my CB score was lower with it) I'm back to LLC5 and it's not crashing at the 1.2V bios setting. (gonna try lowering to LLC 4 in a bit.) Curious what voltage is real though, cpuz shows a higher voltage then hwinfo. 1.216 vs 1.194.
Thanks for the reply.
PBO is enabled with the 200MHz bump and limits removed. My cooler is Corsair H150i 3600mm AIO. I don't think this is a cooler issue, but I'll dig around some more.
Ack, thanks.
I find it really strange all this positing about LLC level. Look I was there and always set my LLC to the highest level as I checked with a multimeter and found the least vdroop. But taken the advice I got I decided to do the same test with LLC set top Auto and I found that I was wrong. Auto was the most stable with the least vdroop. With my current overclock of 4.2ghz using 1.1875vcore I get zero vdroop to the processor with LLC on Auto. The moment I touch LLC my system would crash and I cannot even check vdroop. Yes checking with Hwinfo64 there is a vdroop of between 0.0125 and 0.0250 from what gathered.
Joined
·
95 Posts
Do you find hwinfo to have the most accurate voltage numbers? Hwinfo shows a tiny bit of droop for me at max LLC, cpuz actually shows a tiny bit of overvoltage compared to what I set. I think at least in terms of its value, its motherboard dependant on how effective it is. The highest setting for me either is overkill, or does exactly what you're getting with auto, depending on which monitor software is correct. Comparatively the lowest setting greatly lowers my voltage under load and results in instability. Lv 1 gave me a 0.031 droop or so compared to the 0.019 of lv 5- the max.
I generally prefer auto, with the C7H I used before and now the Msi Meg X570 Ace I found that leaving LLC on auto gives the least vdroop. As for the difference between cpuz and Hwinfo64 I'm not really sure, but from what I heard, the voltage shown in Hwinfo is what is fed to each core where as the voltage shown by CPUz is what is fed to the cpu itself, if that makes sense. If I so much as touch LLC my stability goes out the window no matter whether I set it high very high or low.
Joined
·
22 Posts
I'm running a NZXT Kraken x52 in push/pull configuration. Pump at a fixed 97% which yields the best results according to my own tests. (see attached thumbnails)
I'll give it a whirl if it doesn't overwhelm my cooling solution. Wouldn't want to hover at uncomfortable temperatures for 10 hours straight ;-)
Attachments
Joined
·
95 Posts
I personally use mining software Xmrig as a pseudo stability test. The fact is it raises temps higher then prime95 does for me, and is about on par with Aida64. An added benefit is I get to make some money on the side. I'm sure it's not as intense as benchmarks, but it's still pretty decent. It draws about the same amount of watts as other benchmark tools, although my config has 4 unused cores so it's not perfect. But the cpu is like unusable at all cores utilized, so that kind of tells you how intense it is. It's proven my last like 3 days worth of overclocking tests to be unstable by crashing overnight. Strangely it would crash within an hour of me turning my display off, yet I could have it running for hours while I used the computer without a crash. Might have to look into some kind of idle voltage drop issue or something. I'd love to figure out my prime95 problem and be able to use that to be sure, but even a redownload didn't help.
Is realbench a good metric for stability? Haven't heard much talk about it.
Is realbench a good metric for stability? Haven't heard much talk about it.
Joined
·
1,799 Posts
My Per CCX OC. I haven't updated my Sig, but using Watercooling, 1 360 Rad, 2 280 Rads, 1 Loop, GPU and CPU in Loop.
HWBot result: https://hwbot.org/submission/4327847_
Screenshot for OCN:
![Image]()
HWBot result: https://hwbot.org/submission/4327847_
Screenshot for OCN:
Joined
·
95 Posts
My Per CCX OC. I haven't updated my Sig, but using Watercooling, 1 360 Rad, 2 280 Rads, 1 Loop, GPU and CPU in Loop.
HWBot result: https://hwbot.org/submission/4327847_
Screenshot for OCN:
![Image]()
Yeesh, can it even run a sustained test or is it just going to throttle? If its hitting 93C from just the CB duration, that's pretty darn hot. Pretty good for a - get the highest score you can metric, but seems pretty useless for any actual tasks. Unless that maximum is a freak outlier and it was much lower after hitting it.
Joined
·
95 Posts
So I tried Auto, and it results in my voltage dropping from 1.2V bios set to 1.125 when running cinebench... Which is the largest drop I've seen so far, and also caused CB to crash. so yeah, i stand by my belief that auto only works on some motherboards. yeah, it does run at 1.2 when not at load, which is great! But it also drops more then anything for me.
From what I've seen, when my vcore is set at 1.1875 that is what HWinfo64 will show with LLC on AUTO. CPUz will show 1.2v, but the moment load is applied the vcore will show in cpuz between 1.187 and 1.192 but in Hwinfo64 under load it will drop to 1.162vcore.
Joined
·
22 Posts
Ok - so I made an observation. My 43/43/42/42 @1.175V (full load voltage) is only stable with VSOC at 0.950V (which is the voltage I used when I dialed it in) ...
I upped my VSOC to prepare for some memory OC and made the following observations:
VSOC: 1.004V - Y-cruncher would insta-crash my pc
VSOC: 0.975V - Y-chruncher would crash my pc at the end of the 5,000.000.000 decimal digits cycle (I have 32Gb of ram, so thats the option I use)
VSOC: 0.950V - Y-cruncher runs without issues
Can anyone explain this behavior?
I upped my VSOC to prepare for some memory OC and made the following observations:
VSOC: 1.004V - Y-cruncher would insta-crash my pc
VSOC: 0.975V - Y-chruncher would crash my pc at the end of the 5,000.000.000 decimal digits cycle (I have 32Gb of ram, so thats the option I use)
VSOC: 0.950V - Y-cruncher runs without issues
Can anyone explain this behavior?
281 - 300 of 984 Posts
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
