Question Ryzen 3950X Overclocking/undervolting benchmarking results spreadsheet: Temps and CB20 scores. Thoughts?

[tiberium87]

Hi all,

I am new to the overclocking world, and tried to document the CPU temperatures and Cinebench 20 scores as a function of different combinations of clock frequency caps and Vcore caps. The results are summarized in the attached images (that I am still in the process of filling in, but there is already a lot of info that might be useful for other 3950X users!). I would love to hear your thoughts.

Specifically:
1. Do you think my CPU is running too hot at given frequencies and voltages--and so should I check my fan setup / thermal paste (I did not do them myself, had a company do them for me)
2. Any other suggestions? Would love to hear what people with experience think about these stats.

Bonus question:
3. Do the starred processors (On Ryzen master) actually perform better than the other processors? I am thinking of testing different Vcaps on different processors to see what I can get there in terms of performance boost. For example starred processors cap at 4.2GHz, while unstarred at 4.15GHz, and variations of this.

Best,
tiberium87

 

[AnandThenMan]

What cooler? If you're using a spoon as a heatsink your temps are incredible if you have an expensive 360 CLC then they are kinda high.

 

[tiberium87]

Hello AnandTheMan,

Thanks for the response, good to know that this is running hot and that I need to check this. I am usinga 360 mm Radiator, 120mm Fans x 3. So yes, a liquid cooler that I think is CLC / AIO.

I guess I could try to talk to the manufacturer and ask them if they can check the cooling setup (I don't want to open the chassis myself until I know that doing that will not void the warranty).

Best,
Tiberium87

 

[AnandThenMan]

First thing to realize is the 3950X runs hot or more specifically it has a lot of heat in a very small area this is hard to cool. I'd say anything around 85C is fine on that processor. But at times it looks like you are getting close to thermal throttling for a custom cooled loop I'd be disappointed.

 

[tiberium87]

Alright so I think I am good with just sticking to something like 4.05GHz at 1.2V --> 72C.

Actually somehow when I run MATLAB the actual temps the CPU gets to are always 4 - 8 C lower than the Cinebench benchmarking temps. My standard workflow is running a full 16CPU ("worker", in MATLAB parlance) load 24/7 for days or weeks on end (though this PC is new, so far I have only run the code for a day or so straight). So at 4.05GHz/1.2V, instead of 72C, my MATLAB code gives me a cool 68C. Basically, I imagine if I want to run something 24/7 forever, probably good for CPU longevity if I try not to exceed high 60s/low70s.

I do still want to get to the bottom of why my CLC is underperforming. Will talk to the manufacturer (and report back, but they are slow to respond rn because of the whole pandemic situation).

Thanks Anand.

 

[AnandThenMan]

Maybe the CLC isn't under performing, at a given liquid volume, airflow etc. it can only cool so much. On OC'ing I've learned to accept not going for the extra 5% the heat/temp cost is too high. Your CPU is a beast even at stock. BTW yes the star marked core is the one that will boost the highest, although for me I don't care much about single core boost most any video/rendering work I do hits all cores.

 

[tiberium87]

Yeah, that is my thinking too. I am getting 10-15C lower (!) for a very affordable 5-8% performance hit.

And yup, same here actually. Transient single core boosts serve almost no purpose for me.. Am just curious about the nature of this beast.

 

[AnandThenMan]

Yeah, that is my thinking too. I am getting 10-15C lower (!) for a very affordable 5-8% performance hit.

Same deal for me I go from about 100watts max to 135 (according to Ryzen Master) that's going from 4.25 to 4.4-ish. Temps go way up as well 72 @4.25 to 86 @4.4.

 

[thor23]

Yeah, that is my thinking too. I am getting 10-15C lower (!) for a very affordable 5-8% performance hit.

And yup, same here actually. Transient single core boosts serve almost no purpose for me.. Am just curious about the nature of this beast.

You should try CCD overclocking, the first CCD is usually 100mhz+ faster than the 2nd CCD. Your temps look fine to me for water cooler, my budgetish air cooler lets my 3600 nearly hit 80c in aida64 at 1.2v, you're probably pulling around 2x as much power than me at 1.2v so the temps look fine.

 

[tiberium87]

Hey, so I posted a similar thread on Tom's hardware, and a couple of folks there had some very deep and insightful comments. I would suggest having a look. Looks like 1.2V are not necessarily safe on the 39xx chips. See the thread.

Also, looks like the eco mode stock is doing exactly what I need. See there for an in depth discussion!

TomsHardwareThread