AMD Ryzen 9 3950X - vcore and thermal

[GamerGuy]

I applied the 1usmus Power Plan on my Ryzen rig which has Win10 v1909, as well as the latest BIOS F10c for my Aorus Xtreme. This is with a Thermaltake Water 3.0 360 AIO....a good 280 AIO and higher should be pretty good for the 3950X. The below screenshot was taken after playing about 20 mins of UT3 and about 1 hour of Metro Exodus.

 

[Zach_01]

If only we could see the perf# numbers...

I have also 1usmus's power plan along with the BIOS settings that requiered. Since that I saw light/middle/gaming loads stay on the reffered top performance cores, and besides that windows scheduler is trying to keep things mostly in 1 CCX of the 1 and only CCD.
I'm also on F10c and 1909 but the above findings I noticed, was the same with F10a and 1903.
The only difference with BIOS change F10a --> F10c was that PrecisionBoost wasnt working the same. I had it on advanced and eventually now is just enabled to work the same.

This power plan was targeting of course the 2CCD/4CCX CPUs like yours and 3950X. There should be working better.

 

[csendesmark]

After 36min of Premiere with around ~50% utilization:

Later will check with a little stress test when it runs ~100%

 

[Zach_01]

Man are you having those values magnified... why really?

Did you change the order of the columns from current/min/max/avg to current/max/?/?

 

[csendesmark]

Well, my next 4K display will be a larger one

 

[Zach_01]

This is what a 24" 1920x1200 fits... you dont need a 4K for that...

 

[Deleted member 185158]

Well, my next 4K display will be a larger one

Looks good to me!! My eyes aren't as good as they used to be!!

Put a fan on it yet?? (curious to see if it would help!!)

 

[oxrufiioxo]

Your temps do look high for a 50% load. The review units were running at around 1.2v on all core workloads yours seems to be running much higher than that.

 

[Zach_01]

Yes, the temps are a bit high but the vcore I dont know... Its possible once in full load all cores may drop frequency further (4.2 -> 4.1 or even 3.9) and voltage from 1.3+ to 1.2~1.15. And that consumption of ~140W will remain about the same.
This is how ZEN2 works...

EDIT: 45 minutes later

I just did a quick Cb R20 test with 6 and 12 threads...

6 threads: Avg speed ~4050MHz, Avg voltage 1.37V, Avg total package draw 68W, score 2250
12 threads: Avg speed ~3950MHz, Avg voltage 1.30V, Avg total package draw 76W, score 3500

 

[oxrufiioxo]

Yes, the temps are a bit high but the vcore I dont know... Its possible once in full load all cores may drop frequency further (4.2 -> 4.1 or even 3.9) and voltage from 1.3+ to 1.2~1.15. And that consumption of ~140W will remain about the same.
This is how ZEN2 works...

one of his CCD is significantly hotter also and looks like the culprit of his bad temps maybe its a mounting issue.... at 50% ish load for a couple hours my cpu maxes in the 50s his shouldn't be 20-30C higher even on his cooler.

My CPU is more comparable than yours to his and even under 100% load for hours my temps don't get that high with the same ish voltage. 1.32-1.35 also both my CCD are typically 2-3C from each other regardless of workload.

 

[Zach_01]

I edited my previous post, did you see it? with the Cb test.

one of his CCD is significantly hotter also and looks like the culprit of his bad temps maybe its a mounting issue.... at 50% ish load for a couple hours my cpu maxes in the 50s his shouldn't be 20-30C higher even on his cooler.

Of course that is happening he is running only 50% load... All the load is mostly on 1 of the 2 CCDs. Look at the individual core power SMU from 0 to 7. Its ~13W per core and then look the other cores 8-15.
Maybe he doesnt have good case air flow...

My CPU is more comparable than yours to his and even under 100% load for hours my temps don't get that high with the same ish voltage. 1.32-1.35 also both my CCD are typically 2-3C from each other regardless of workload.

He is having 25% more cores than you to each CCD. With the same voltage and speed it will produce more heat. Are you running 4.2GHz at 50% load?

 

[oxrufiioxo]

I edited my previous post, did you see it? with the Cb test.


Of course that is happening he is running only 50% load... All the load is mostly on 1 of the 2 CCDs. Look at the individual core power SMU from 0 to 7. Its ~13W per core and then look the other cores 8-15.
Maybe he doesnt have good case air flow...


He is having 25% more cores than you to each CCD. With the same voltage and speed it will produce more heat. Are you running 4.2GHz at 50% load?

typically 4.2-4.3 at 50% or less and 4.0-4.1 100% all core..... Reviews showed the 3950X to run cooler than the 3900X regardless of load due to it being better binned. This is why OP is mostly confused by his temps his seem to be worse than even the boxed cooler with similar workloads on a 3900X


honestly until more people get this cpu at retail it's going to be really hard to gauge how Retail samples behave.
AMD could have sent golden samples to reviewers and slightly higher than 3900X may be the norm

launched slightly after R20 started

 

[Zach_01]

I agree with you 100% that he is running hot. No doubt... yours having the same 140~145W reading and it is 10~15C lower. But you are on AIO. Don’t forget also that reviewers do the tests straight to ambient ~22C.

Also the difference in the 2CCDs the OP is seeing is normal as the hottest one produces on avg 13W per core and the coolest only 4~5W per core. This explains the almost 20C delta.

 

[oxrufiioxo]

I agree with you 100% that he is running hot. No doubt... yours having the same 140~145W reading and it is 10~15C lower. But you are on AIO. Don’t forget also that reviewers do the tests straight to ambient ~22C.

Also the difference in the 2CCDs the OP is seeing is normal as the hottest one produces on avg 13W per core and the coolest only 4~5W per core. This explains the almost 20C delta.

Yeah, I'm not disagreeing with you either your points are all valid.... Just going by reviews his seems to be running hot even for his cooler I could be absolutely wrong and his cooler may just not be good enough.

Also his VRM is likely running significantly hotter than ours even during modest workloads contributing to the heat around the socket.
every review ran the chip in an extremely overbuilt x570 board that I have read.

Regardless going high end Noctua or similar is probably his best bet assuming it fits in his case..... the 690 III isn't all that great for radiators.

 

[Zach_01]

Yes, could be just the cooler... I don’t really know how it performs normally.

 

[Deleted member 185158]

Yes, could be just the cooler... I don’t really know how it performs normally.

Pretty much the only thing that's cooling the cpu, so I agree with you guys.
He could try to space the fan off the cooler to eliminate the dead zone at the center of the fan.
Push pull fan set up if the case has the space.
Fan near the VRM just for added measures.
Could also try to lower ambient temps 5-10c also. That helps quite a bit sometimes.
All of the above, or build a nice custom water loop and run it nice and cool..... (AIO I guess if you're broke after buying the cpu....)

 

[csendesmark]

Your temps do look high for a 50% load. The review units were running at around 1.2v on all core workloads yours seems to be running much higher than that.

I learned that HWInfo can't measure correctly the vcore on these processor, because it heightening the average vcore measurements,
when a Ryzen 3000 CPU sending any core to sleep, hwinfo keeps showing the last reported vcore voltage, incorrectly
Also checked against the Ryzen Master tool which is usually showing a lower value across the chip.

 

[Zach_01]

What version of HWiNFO are you using? There have been some changes in the last few versions (6.13.3955beta and after)

FEATURE - Effective clock vs instant (discrete) clock

It has become a common practice for several years to report instant (discrete) clock values for CPUs. This method is based on knowledge of the actual bus clock (BCLK) and sampling of core ratios at specific time points. The resulting clock is then a simple result of ratio * BCLK. Such approach...

www.hwinfo.com

Also from the day I got the R5 3600 and install HWiNFO (6.10.xxxx) I noticed that if the PC stay long enough at idle state, the minimum individual core voltage reading is 0.2V

 

[csendesmark]

the recommended version - always the latest stable available

Also from the day I got the R5 3600 and install HWiNFO (6.10.xxxx) I noticed that if the PC stay long enough at idle state, the minimum individual core voltage reading is 0.2V

Those V 0,200-s are totally possible,
what I say is: when the cores are running with an elevated voltage like 1,3~1,4V then getting a sleep command the HWInfo will not refresh the value and keep calculating the avg vcore with that "ghost" value
While Ryzen Master can detect sleeping cores

 

[Zach_01]

Yes, Ryzen master uses different method... The last implementations of HWiNFO though respect the sleeping mode. Thus, the effective core clock.