Maybe. I was using FFmpeg 4.0.2 and I think the new microcode plus Linux optimizations required more Vcore.
I think you misread. I was trying to point out that while you may feel this overclock is necessary and should work, at the end of the day, if it’s either persist and not have a usable machine, or downclock and do, you should probably downclock. You are free to keep trying, but personally I will take less performance over more hassle any day of the week.
My temps reach nowhere near TJmax. Especially when doing FFmpeg with a max of only 65C on the hottest core. Even bumping up Vcore didn’t change max temps.
Current draw is also the same between the 2 voltages. I checked with HWinfo. 1.35 was at 0% Vdroop curve and 1.4 was at 30% Vdroop curve making the load voltage 1.39V.
I know what I’m doing. I also bought a CPU still covered under warranty until 2020. Amperage kills, but voltage if kept cool doesn’t as badly do the same effect. That’s why it’s 1.4V with Vdroop, cause without Vdroop would mean death.
Edit: Nope. Just got lucky on medium. 1.4 isn’t stable on slow neither. Gonna change BCLK back to 100 and see if it’s literally something funky with BCLK tweaks. If BCLK isn’t it. I’m giving up.
SUCCESS.
Funky BCLKs don’t play nice with FFmpeg x264 AVX on Linux. Initial instability was still Vcore, but I kept the BCLK on the wrong value for stability, cause it only plays nice at 100mhz BCLK on Ivy Bridge-E.
In fact, looking on overclock [dot] net, BCLK overclocking is a definite no no for 24/7 OCs. Looks like this is the first time and last time I’ll be doing that.
2 Likes
NOPE.
Tried encoding again today and it started crashing again.
Just gonna encode with a AVX -1 offset I guess. I give up on getting 4.4 stable with FFmpeg.
It’s weird because I encoded my ViewSonic video just fine but it waited until now to re-introduce instability.
Hey. I have or should I say had OC problems too. My CPU is an old x3440 maybe the newer gen is different but here is what I found.
I didn’t catch if you are running prime95 on windows but if you do. Windows is more tolerant to overclocking more “stable” if I can say so. Sure it will do some silent error but in my almost 5 years of 4.2Ghz overclock I needed to reinstall only once never did I loose data. Crashing occurred also only once upon a blue moon. Workloads ranged from gaming up to blender renders, substance painter and unreal shader compiling.
As soon as I switched to linux Fedora I couldn’t even get into the loading circle and the kernel would stop. I thought that’s it my PC is too old for linux, until I defaulted the BIOS and it started working. Now I am running 3.6Ghz and 1200Mhz memory and it’s finally stable.
Just few days ago I did some ffmpeg with 4K youtube-dl downloads and it’s stable.
I know it’s not an answer but my suggestion is to scale back OC for linux and make two profiles Windows “stable” and Linux stable.
Hope it helps.
youtube-dl downloads don’t encode, but they just stream from the network. It’s not a CPU intensive thing.
I’m almost certain the Intel microcode updates increased CPU stress. 4.3Ghz has no issues encoding, but 4.4Ghz has non-stop issues encoding, going into the dangerous voltage areas if you want to stabilize that.
1 Like
Sorry for not elaborating more as I wrote a lot already. I used the merge audio video parameters in youtube-dl. This uses ffmpeg to combine the audio and webm video stream into one video format manly mkv. That’s where it encodes.
I see… Well it’s interesting nobody is mentioning anything outside of some overclocking forums. I mean Wendell is doing overclocking and using all the OSes under the sun. Isn’t he encountering this difference…?
Older hardware at higher clocks might be more sensitive to microcode changes for Spectre and Meltdown and L1TF.
Merging webm/mp4 is a I/O intensive task, but not a CPU processing intensive task.
1 Like
Downgraded my microcode, didn’t work.
Disabled all Spectre mitigations… didn’t work.
IDK if the version of FFmpeg would have anything to do with it… I’m on 4.0.2.
Well unless you are an OC expert and you exacly know,
what to expect the value you change on the BCLK.
Otherwise i would highly advice to leave BCLK at its stock 100.
Because BCLK overclocking affects allot of other things to.
43 x100= 4300mhz is a pretty decent number for a 4960X.
I run 44 x 100 just fine EXCEPT for FFmpeg. That program is it. I have to AVX -1 manually to get stable.
Did you already try to play arround with some LLC?
Yes, within reason. Didn’t work. What worked before was extra Vcore but now it would require dangerous amounts to keep stable, so I’m opting for a manual AVX -1 offset.
Yeah or honnestly like i mentioned if 43 x 100 is perfectly stable,
then i would leave it at that personally really.
I mean that 100mhz difference in real world performance gain,
will be really minimal.
Single core performance is what I care about, so for games, 44x is fine.
43x is for FFmpeg…
Yeah i get that, but even for single core performance,
it wont really make a huge difference.
I mean 100mhz isnt going to make your games perform amazingly better.
But of course if you dont mind switching between OC profiles all the time,
then of course you could totally do that.
Of course it is also a matter of silicon luck really.
I mean you can have two exact the same systems cpu + mobo combo´s,
one can be a great overclocker, the other one could be an absolute bust.
Yeah, I guess I got a tiny bit unlucky if this is happening.
I think the profile switching is a better way forward, but I prefer automatic AVX offsets to do the work and that only got introduced in X99. X79 didn’t have AVX offsets.
Yeah well if you know what you are doing,
You could try to back off your cpu multiplier to like 42,
and then bump the bclk to like 105.
But the probablly going to need some fine tuning with LLC.
I have sensitive PCIE devices and this is a 24/7 stable situation so not really gonna mess with BCLK.
1 Like