x99 5960x Overclocking Question

[tyga909]

Hi to all

My first post here. I have a question about overclocking. So I overclocked my 5960x to 4.0Ghz using XMP 3000Mhz settings with BCLK and CPU strap at 125, CPU core offset of .11, Cache V offset at .05, Sys V offset at .22 and Dram V at 1.35 as per specification. So upon learning that its near impossible trying to run my RAM at 3K, i bumped it down to 2750 which was the next best thing and stability achieved with the tweaking mentioned above. I did stress tests using SuperPi(Quick), OCCT and OCCT LinPack - stable for 1.5 hrs.

Then when I got to overclocking my CPU, got it bumped upto 3.8Ghz with only CPU core offset but at 4Ghz I had to bump the cache voltage up too, I just thought I should because the stock cache V was a specific amount below the stock CPU V, so to match that difference I bumped it up - result stable. My idle temps of all cores come down to 34C average and 42C for the package. Also, I am using the balanced power option setting for my PC for low V at idle and whatever is required when under load.

Now, I heard this thing about reaching the 'Sweet Spot' V when overclocking that helps keep temps and everything optimum, I have referred to overclocking charts with voltages and temps listed, which match; so what I am trying to ask for is how do I get to know whether I have reached that sweet spot? Does increasing the voltage inevitably add more heat as should be or help me reach the coveted 'sweet spot' leading me to have lower temps? Or are the settings that I have good enough as it is?

I would really appreciate it if anyone could review the above settings and let me know if what I have done is good enough. This is my FIRST PC build and well FIRST overclocking experience. I have read extensively about all these things and have applied whatever I was able to understand. System specs in profile drop down.

Thank You

[Arne_Saknussemm]

Hi tyga909 🙂

Thing is the "sweet spot" is different for each CPU so it's not really possible to check out other people's results other than to get a vague idea of the average.

The only way to find your sweet spot is to experiment. when you satrt OCing the CPU you can get a 100MHz increase in performance for relatively small increase in voltage....and this continues quite often linearly as you go on up...and yes each increase in voltage/current will mean more heat.

At some point this linear increase stops and the next 100MHz costs you a big jump in voltage....just below this point will be your sweet spot...it's just a matter of plotting out this graph for yourself...

Often these processors liek 1.15 to 1.2v and give good performance with good temps...but as I say each is different so...

These CPUs can very often run 3000MHz ram on the 100BCLK

If you set the RAM up manually (frequency, primary timings, and voltage...try 1.35 or 1.36..1.37) on 100BCLK you might find yourself able to run...experiment with VCCSA too...up and down...sometimes it's not just a question of increasing here...there is a sweet spot for this too.

My 4.4 settings here might help in terms of settings in digi+ section etc. ....voltages your own of course
https://rog.asus.com/forum/showthread.php?51063-The-Haswell-E-Overclocking-Thread-and-Overclocking-G...

[Janne-71]

Arne Saknussemm wrote:
Hi tyga909 🙂

Thing is the "sweet spot" is different for each CPU so it's not really possible to check out other people's results other than to get a vague idea of the average.

The only way to find your sweet spot is to experiment. when you satrt OCing the CPU you can get a 100MHz increase in performance for relatively small increase in voltage....and this continues quite often linearly as you go on up...and yes each increase in voltage/current will mean more heat.

At some point this linear increase stops and the next 100MHz costs you a big jump in voltage....just below this point will be your sweet spot...it's just a matter of plotting out this graph for yourself...

Often these processors liek 1.15 to 1.2v and give good performance with good temps...but as I say each is different so...

These CPUs can very often run 3000MHz ram on the 100BCLK

If you set the RAM up manually (frequency, primary timings, and voltage...try 1.35 or 1.36..1.37) on 100BCLK you might find yourself able to run...experiment with VCCSA too...up and down...sometimes it's not just a question of increasing here...there is a sweet spot for this too.

My 4.4 settings here might help in terms of settings in digi+ section etc. ....voltages your own of course
https://rog.asus.com/forum/showthread.php?51063-The-Haswell-E-Overclocking-Thread-and-Overclocking-G...

Nice description, i suppose i did the right thing instinctively when i left my 6600k to 4.6ghz when reaching 4.7 would have needed too much additional voltage. It was a clearly the sweet spot of my cpu(4.6).

[Chino]

tyga909 wrote:


Now, I heard this thing about reaching the 'Sweet Spot' V when overclocking that helps keep temps and everything optimum, I have referred to overclocking charts with voltages and temps listed, which match; so what I am trying to ask for is how do I get to know whether I have reached that sweet spot?

What I do is graph each frequency with the corresponding voltage necessary to achieve stability. Normally your CPU voltage scales in a linear pattern as you start increasing your CPU frequency. Eventually you will get to a point where it needs a hefty increase in voltage for an extra 100MHz. That’s the point where diminishing returns start to kick in.

[Gobe]

I tend to look at voltage with my 5960X. I can stabilize mine at 4.7 GHz, but I have to mess with a chiller and high speed AC fans to pull it off and I'm pushing toward 1.5 volts. This is far from practical for a routine overclock... It's good for getting stable benchmarks to make my friends envious, but if it won't stabilize on my CM Nepton 240M, it's not stable for my day to day usage..

4.6 GHz stabilizes at 1.36 volts with max CPU temperature around 90°C on a one hour Real Bench run.
4.5 GHz stabilizes at 1.25 volts with max CPU temperature around 78°C on a one hour Real Bench run.

Guess which one I run routinely? If you said 4.5 GHz/1.25 volts, you got it. I give up a single bin for a 12°C temperature drop, much lower power, and much less fan noise.

[Arne_Saknussemm]

Gobe wrote:
I give up a single bin for a 12°C temperature drop, much lower power, and much less fan noise.

Exactly! :cool:

[Arne_Saknussemm]

Cheers Janne-71!

Yes it's not to hard to find....what's hard sometimes, is admitting to yourself that maybe that sweet spot is not as high as you'd like it to be 😮 but the CPU will be telling you... just have to listen...

[Gobe]

Arne Saknussemm wrote:
Cheers Janne-71!

Yes it's not to hard to find....what's hard sometimes, is admitting to yourself that maybe that sweet spot is not as high as you'd like it to be 😮 but the CPU will be telling you... just have to listen...

Accepting the limitations of a new chunk of silicon is always hard... we all want to win the lottery just once. I tried like hell to surpass your 220K Real Bench result, but I had to finally admit that it wasn't going to happen with my half-assed graphic setup. Did you buy that 5960X pre-binned to get 4.8 GHz? Outstanding! Did it take cooling heroics or do you get 4.8 GHz on an everyday setup?

[Edit: Just looked at your system specs, looks like your everyday setup involves cooling heroics!]

[Arne_Saknussemm]

Gobe wrote:
Accepting the limitations of a new chunk of silicon is always hard

Believe me I was talking from personal experience...:o

No, actually got this CPU from Intal RMA when a chip died on me...pure luck

Cooling heroics LOL....I have a radiator hanging off my loop that I dunk in iced water....but no exotic phase or LN2

By the way your image editing is way above mine....I need to take a look at that!;)

[Gobe]

Arne Saknussemm wrote:
By the way your image editing is way above mine....I need to take a look at that!;)

I'd hazard a guess that the image editing benchmark uses some writes to disk along the way. Samsung 950 Pro NVMe might be the secret sauce on this one. I also suspect that once the fastest DRAM, CPU, and cache frequencies are found, there might be gains to be had from backing off on the cache a bit to find a performance sweet spot between the three.

Also, and this is very important, I find that when I run the benchmark for the first time after starting, the image editing score is always significantly lower than starting the benchmark, letting image editing run, bumping the mouse after image editing completes to stop the benchmark, and restarting the benchmark. Don't know why, but this is big for maxing image editing.