Are there sweet spots for overclocking 5960x?

doublehamm wrote on 2/8/2016, 11:55 AM

Hello, I just did a complete overhaul on my PC this weekend. I have tried doing some minor overclocking to get the most out of my system. My mind is boggled with my results, as it seems the faster my system is, the slower my renders.

I have ASUS Rampage V Extreme and i7 5960x. I am in no way a great or even fair overclocker. First I tried the "EZ Tuning Wizard" just to get somewhere while I tried to educate myself on better overclocking. It simply asked what kind of system it was intended for - So I picked Gaming/Media Creation. Then it asked what kind of cooling - So i picked water (H100i). It gave me some quick results stating CPU performance would increase 32%, and DRAM performance up 14%. Alrighty, all looked good, I restarted my PC, and Windows loaded faster than I could complain.

HWiNFO64 showed I had CPU clock at 3.978Ghz, and RAM running around 2400. ASUS AI suite showed nearly identical results.

With this setup, I did a test render of a project in Vegas Pro 12. The clip was 15 seconds long with heavy noise reduction and other effects on it. Total render time was 03:41. Max Temp was 60C

Not bad for a quick start I thought, but I wanted to get my RAM up to the 3000 it is rated at and increase the CPU a bit. Here is a SS collage of what I felt was important to share from these settings:

I Semi followed the guide here: http://rog.asus.com/365052014/overclocking/rog-overclocking-guide-core-for-5960x-5930k-5820k/ to try and see where I could get. After some trial and error I was able to get my CPU to 4.375Ghz, and DRAM to 3000.

I was ecstatic when I finally saw windows load, and happy when I opened HWiNFO64, and AI Suite to check and make sure what I set was what was what was going on. All looked good I thought so I opened up the same project from Vegas and did a test render of the same 15 second clip. It looked quite a bit slower than what it was doing the previous time but I was hoping it was just my mind playing tricks. Nope!

When all was said and done, the render was 30 seconds longer than when my PC was showing slower settings (4:11 vs 3:41). Max temp was also 71C, 11C higher than the previous settings.

Here is a SS collage of the higher OC settings:

Does anyone know WHY I am getting worse results with higher OC, not to mention burning more watts and adding more heat?

I took some crappy cell phone pictures of what was going on during MY tweaking in the bios.

Thanks in advance for any help!

I am currently back in the "dummy OC mode" and things are as fast as they were the first time I used these settings.

Back to post

Comments

musicvid10 wrote on 2/10/2016, 10:22 AM

Have you tested the exact setup with Premiere, Media Composer, or Edius?
Handbrake x264 encodes tend to run a little hotter CPU than Vegas.
The same may also true of ffmpeg.
You can download trial versions of any Sony upgrade or product to try for 30 days.
I suspect you will eventually draw some conclusions about video encoding in general and overclocking, but I can't predict your outcomes on a level playing field.

I am quite sure Sony has no policy on overclocking encoding machines, other than they may run hotter and less stable.
Do let us know how it all turns out, then try it on a 4-day encode.

BruceUSA wrote on 2/10/2016, 11:30 AM

Musicvid10,

Since you are such a anti overclocking and have often claimed overclocking is bad for your computer, bad for editing/encoding etc. I have been working on a project every month and I have never encountered any rendering crash. None.

All I want to say is this. Are you willing to put your money where your mouth is? Do you have any project that required a 4 days rendering? If you do. Please send it to me and I am willing to accept your challenge and put my system 1 in for a 4 days rendering test for you. What do you say eh?

OldSmoke wrote on 2/10/2016, 11:42 AM

doublehamm

How is your 64GB ram configured? It should be 8x8GB. Did you buy quad channel sets and made sure that you don't mix them up during installation and also put it into the right slot? Try and take out half and OC again.

doublehamm wrote on 2/10/2016, 12:06 PM

Corsair Vengeance LPX. CMK16GX4M2B3000C15 64GB total (4 kits) 8x8

musicvid10 wrote on 2/10/2016, 12:10 PM

Bruce,

Eh, I explained that my direct experience is dated, using fewer cores than you undoubtedly are using, and yet silicon has changed very little over the last twenty years.

The analysis is simple: Hobbyists love speed and benchmarking; production folks like reliability. I don't pick the poster children for either daycamp; however, I prefer the reliability angle for multi-day encodes, because doing my work on a deadline more or less dictates not having to do it again.

Here's a test for you, however:
a. Set your BIOS and everything else back to defaults. Run a 24+ hrs encode.
b. Now overclock your system to what seems to run stable, and run exactly the same encode.

Now, if the second encode succeeded, subtract time b. from time a. to get your speed increase
If the second encode didn't succeed, back off your voltages, and encode again. Now double the new b. time and subtract your actual a. time to see your speed loss.

If you don't instantly see the factors an indie producer faces when time and money is at stake, it may be a good idea to forego that path and continue to enlighten us with your bleeding-edge testing.

Best to my overclocking friends.

doublehamm wrote on 2/10/2016, 12:16 PM

Thank you for your concern. My main focus is Wedding Videography, so this is a slower time of the year. If this was mid-summer I would not keep testing, and I would just go get the job done.

My goal is to get it nice and solid for this summer. My 3930k I have run 24 hours renders at 4.4 and 4.5GHz with no issue. I have it at 4.4 for now just because I have it in a smaller case than before and with a smaller cooler, but seems to be doing fine. I may up that in a few days if I see the temps behaving. That ended up being a solid system, and i will still use that for rendering while I edit on this machine.

doublehamm wrote on 2/10/2016, 12:20 PM

OldSmoke,

Also I know you stated 2133MHz is best, and that is what it was at when I was doing the 4.1-4.5 tests with Vegas crashing. The current setup is running at 2400MHz and no issues at all with the memory or CPU at 3.978GHz

musicvid10 wrote on 2/10/2016, 12:25 PM

Yes, and where ambient room temperatures reach 85+ this summer, I suspect you'll lower it again.
The rough translation of 0.1 GhZ or even 0.5 GhZ to encoding speed gains are not only less than you think, they pale in comparison to doing any render twice.

The speed gains from backing off high profile settings in an otherwise good h264 encoder will exceed those from overclocking, with only small size collateral and no quality loss, and I'll go head to head with anyone at that sandlot.

doublehamm wrote on 2/10/2016, 12:31 PM

My location is wonky. The vents are whacked. My office is in the basement. In the winter, it is MUCH warmer down here because all of the vents have no control to open or close them and they are huge. I swear 75% of the airflow comes to the basement. In the summer, I freeze down here and have been known to run a space heater due to overactive AC trying to cool the upper level.

OldSmoke wrote on 2/10/2016, 2:25 PM

doublehamm

The reason asked is that many motherboards tend to behave better with less memory slots occupied.

Edit:
How strong is your power supply and do you have an UPS too? A strong and stable voltage supply is a must for any system but even more so for one that is overclocked.

doublehamm wrote on 2/10/2016, 3:34 PM

OldSmoke,

1200W PSU.

I do not believe my RAM was quad channel, plus they were 4 kits, which is not always a great idea.

I just ordered a 64GB kit that I believe is listed in the QVL for the board. The Corsair I have is not, and I do not believe they are quad channel.

This is what I just ordered: http://www.newegg.com/Product/Product.aspx?Item=N82E16820231832

Maybe I am just throwing more darts at the board, but I have my fingers crossed.

What do you mean about not mixing them up?

I will try removing half later to see.

OldSmoke wrote on 2/10/2016, 3:57 PM

When you buy a kit of 2 sticks for dual channel or 4 sticks for quad channel and more then one kit, you must NOT mix the sticks between the kits. These are matched sticks per kit. If you want to use quad channels without problems, you buy either 2x dual channel or better one quad channel set.
If I understand it correct, you initially bought 4 kits of 2x 8GB each, hopefully dual channel kits. Still ok, as long as you put the sticks from each kit in the respective dual channel slots on the board, A1+D1, B1+C1, A2+D2, B2+C2 according to your boards manual.
The new GSkill you ordered should be fine as it is a kit of 8 matched sticks and it doesn't matter how you put them in.
1200W PSU is great, just make sure you got stable and "clean" mains supply too; I always run a UPS on all my systems.

doublehamm wrote on 2/10/2016, 4:19 PM

HA! Weill I just put in the RAM slot by slot next to each other. I guess I have never purchased multiple kits before, usually all in one but that is all Microcenter had. I will try this. I Just need to remember how I put them in...

In any case, the 64GB kit I ordered will save me $100 over the other kits, but I will see what I can do before those arrive.

OldSmoke wrote on 2/10/2016, 4:37 PM

Taking out half or even reduce to 16GB could improve system stability.

astar wrote on 2/10/2016, 7:31 PM

When you installed your memory, did you run a full pass on Memtest86+? This will determine if you memory settings are stable and error free. This test could take an hour.

I am not a huge fan of overclocking past the base features on the board. If the intel processor supports Turbo boost great run it. If the memory has an SPD that supports a given freq that matches the memory controller? Great that works. If you are custom overclocking your motherboard beyond this, you are spending your life re-testing things Intel and the board manufacturer has spent thousands of man hours arriving at a stable solution. This is not smart to me.

doublehamm wrote on 2/10/2016, 7:51 PM

OldSmoke

THANK YOU! I re-arranged my sticks per kit like you said, and now it seems to be overclocking like a dream!

Granted I have not done overnight testing on this, but I have run quite a few stress tests and benchmarks so far and all seems fantastic.

I just did some SCS Benchmark Project tests, and I will upload those in the correct thread in a moment. Including a XDCAM EX render in 12 seconds.

OldSmoke wrote on 2/10/2016, 8:15 PM

If you are custom overclocking your motherboard beyond this, you are spending your life re-testing things Intel and the board manufacturer has spent thousands of man hours arriving at a stable solution.

Not quite true. What MB manufactures and Intel do/did is to make sure that the final system is stable with a great variety of components in it, even the not so good ones. Motherboards like the Asus Rampage series is designed for gamers and overclockers with very special features for it.

BruceUSA wrote on 2/10/2016, 10:05 PM

Very nice 12s XDCam rendering. 5960X with Fury x GPU equal impressive combo for video editing/rendering.

doublehamm wrote on 2/19/2016, 5:04 PM

OldSmoke, What kind of cooling system do you use? Currently I can do 4.5GHz with 1.27 volts (manually) but still gets a bit warm (mid 70s) so I have gone down to 4.4GHz.

Next I started playing with the cache speed. I have increased it from 3000 to 3800 and seem to need at least 1.315 volts to get it stable, but then it starts to run warm again. I got about another 18% increase in render times with this setting, but I know I will not keep it like that unless I can keep everything cool. I am sure I could go even higher, but I stopped at 3800 for now.

OldSmoke wrote on 2/19/2016, 6:43 PM

I use a custom loop; 2x 120mm on a 120x240 radiator and an additional 1x 120mm radiator... my case doesn't have space for a single bigger radiator so I had to split it. I currently use a pump/reservoir for 5.25 bay from XSPC and a XSPC CPU block but I am changing to a stronger EK pump/reservoir soon.

doublehamm wrote on 2/26/2016, 6:00 PM

I have currently settled for 4.4GHz, 1.25V (Manual), 3900 Uncore Clock,. This serves me well with my current cooling setup of the H100i. This keeps all my temps below 70C while rendering the toughest of my videos. The temps do hit the mid 70s if I do a stress test, but in reality, the most stress I will be putting on the CPU is rendering, and I did a 10 hour render and kept temps below 70.

Also as far as rendering, this is only about 2% slower than when I had the CPU running at 4.6GHz @ 1.38V and a 4600 uncore clock. For rendering I don't believe the temps were the worst in the world (only 1 core went over 80C and only for very brief spikes, with average in the mid 70s over 10 hours).

I was getting some kick arse benchmark scores with this setting (4.6x4600) using the ASUS RealBench app and made the first page on the ROG website rankings - but this really was pushing some of the cores in the upper 80s, and I believe one core did hit 90C briefly on spikes. So the geek in me wants to get a custom loop here soon (if I can remove my anxiety for doing so) to see what I can do. This setting still seemed remarkably stable - just HOT!

So far I have resisted the urge to try for 4.7GHz as I really do not want to hurt my CPU which I seemed to have received a good sample in the CPU Silicon lottery - but I am confident it is capable. I have been very happy with the results and seem to be able to get everything running stable with less than average voltage.

My RAM is now the G.Skill 64GB and that I have running at the XMP profile of 2400MHz, and not one issue with it running at this speed. In fact I had it a step above that for a couple days, and was running stable there as well. The first set I had of Corsair RAM was only posting 48GB/64GB once I started upping its rated frequency to 3000, and I did not see 64GB again until I got it below 2700. I know you do not like running memory above what is rated for the CPU, but so far I have had zero issues with the G.Skill.

EDIT: Should I worry about the CPU Package temps? I see mixed thoughts on this. They seem to run about 5C higher than any of my cores, which does seem odd as most places I have looked at show the package temps on the lower side of the core max temps. Maybe it is just a faulty sensor?