Question How Far to Push Overclocking for Professional Film Editing -- Advice Needed

Sep 23, 2019
3
0
10
0
Hi all,
I just completed my first build for a new editing rig because it seemed far more cost effective than purchasing one. I do professional film editing, and buying a custom rig seemed wasteful in comparison to building my own. First off, thank you to EVERYONE on this forum -- there is so much great information, and I wouldn't have been able to do it without the community. Seriously -- huge props.

I'll include my specs below, but my question is this: I've overclocked my CPU to a stable state using the standards/tests outlined in the forums and guides, BUT my daily workload (even 4K) doesn't come close to the load and stresses of the tests. In light of this, should/could I push the overclock further, or is it best to play it conservative for long-term CPU health? I do not game at all, so I mostly want to just get the most out of my money, while also making sure I don't do something stupid that shortens the lifespan of the CPU (thus making me not "get the most out of my money" by cutting the CPU's life down).

HARDWARE:
CPU: i9-9900K
Motherboard: Gigabyte z390 Designare
RAM: Crucial Ballistix Sport LT 62 GB (4 x 16 GB) DDR4-3200 Memory
Cooler: Noctua NH-D15
OS: Windows 10
Case: Carbide 275R

Current Overclock Settings:
4.9 GHz @ 1.30v (Medium LLC) AVX offset -1

Prime95 26.6 Small FFTs 15 Minute Torture Test:
-Passed with no restart issues.
-Max temp (range across all cores): 75-85c
-Avg temp (range across all cores): 74-81c
-MaxV: 1.315

RealBench 1 Hour Stress Test:
-Passed with no restart/issues
-Max temp (range across all cores): 79-84c
-Avg temp (range across all cores): 74-80c
-MaxV (range across all cores): 1.306-1.318

For data collection, I used HWinfo and CoreTemp.

Other info:
-2 cores consistently run a few degrees hotter than the others
-I have three intake fans at the front of the case and I am considering moving the bottom fan (which essentially blows directly on the HD trays and PSU basement wall) and relocating to the top of the case so that is above and slightly behind the Noctua cooler as a second exhaust fan (currently there is one exhaust fan located at the back of the case, in-line with the Noctua fans). I'd imagine this would give a few degrees more cooling IRL tests, or is the priority to have cool, incoming air?
-While rendering and exporting 2K video today, the highest temps across all cores were 60-65c (fans didn't even kick into high). Interestingly, though, the maxV was higher than during any of the stress tests at a range of 1.329 - 1.364. Is this an indicator that I need to make an adjustment to CPU voltage?
-I realize that the Noctua air cooler will ultimately limit my cooling abilities, but if my workloads (and thus, heat) are less than the stress tests, then can I increase OC? The cautious side of me says stick to the settings that could pass the tests with good thermals, but the part of me that is leaving a lot of headroom on the cutting room floor, so to speak, would like to fully take advantage of the hardware.

As I am the quintessential n00b, any detailed explanations and/or links so I may continue learning/educating myself are greatly appreciated. Thanks to all in advance!
 
Sep 23, 2019
3
0
10
0
Haha, I think all filmmakers have had that experience to one degree or another. Most pros are going to have redundancies and backups (at least I do and everyone I've worked with), but this raises a good point that I don't know the answer to: to what degree does OCing introduce instability? I know the standard caveats always apply and that stock is guaranteed to be the most stable, but it also seems that stock purposefully limits the CPU's potential to achieve this. If you do the proper testing, are we talking about increasing instability by 10%? 90%?

I fully realize a measured answer will be "it depends," but using the assumption that you really hit every stress test long and hard, what is that supposed to indicate for stability?
 

USAFRet

Titan
Moderator
Mar 16, 2013
113,328
1,882
155,190
18,564
The other 1/2 of the equation is how much actual performance gain might you get, in your particular tools and workflow.

For a particular OC level, might you see a 5%, 20%, 50% reduction in rendering times.

It is probably much closer to the lower end of that range.
And at what cost. The more you push the OC,the greater possibility of a crash.
 
Reactions: BrooksPuppings
Sep 23, 2019
3
0
10
0
Ahh. That's quite helpful. At those levels of performance gain I'd be inclined to leave it a lower speed than push higher.

In terms expected returns is the RealBench and CineBench scores good predictors? I.e. incremental gains in bench marks = projected incremental gains in real world editing, rendering, exporting?
 
...
I fully realize a measured answer will be "it depends," but using the assumption that you really hit every stress test long and hard, what is that supposed to indicate for stability?
As much as I hate Prime95 small FFT, the 'power virus', I do feel it's the test you should put a 'production' system through before turning your back and letting them start to make money with it. If it can't pass 6-8 hours of it, de-rate the system until it will.

Depending on where/how it's installed, especially if I'm not the primary user, I'd suggest even SABATOGE the system by putting a blanket over the case and run the test. You have to know someone's going to use it for their coat rack.
 
Reactions: BrooksPuppings

USAFRet

Titan
Moderator
Mar 16, 2013
113,328
1,882
155,190
18,564
Ahh. That's quite helpful. At those levels of performance gain I'd be inclined to leave it a lower speed than push higher.

In terms expected returns is the RealBench and CineBench scores good predictors? I.e. incremental gains in bench marks = projected incremental gains in real world editing, rendering, exporting?
Not really.
It also depends on your particular software.

Benchmarks are sort of good to compare 2 different systems, or 2 different settings.
Not good for "10% in Cinebench = 10% faster render".
 
Reactions: BrooksPuppings
....
Not good for "10% in Cinebench = 10% faster render".
Except, if I understand the arguments correctly: Cinebench may indeed be somewhat synthetic, probably because how it controls the interface to make a reliable benchmark utility out of it and the score being something you can't intuitively relate to. But the rendering engine (Cinema 4D) is very much the same engine used in Maxon's product so benchmark results do co-relate to actual rendering performance in their application at least.

In other words: while they're not numerically proportional to render times, higher scores do reliably predict faster renders.
 
Last edited:

ASK THE COMMUNITY

TRENDING THREADS