3900x high temps

C

ChronoDetector

2[H]4U
Joined
Apr 1, 2008
Messages
2,783
Just recently bought a 3900x and so far it is good.

However, I'm rather concerned about the temperatures, I'm using a 360 Thermaltake CPU cooler, the Thermaltake Water 3.0 RGB cooler to be precise and so far I'm getting rather high temperatures. I haven't even overclocked it yet and it goes up to 90C at full load, idle temperatures are around 30-40C. Is this normal for a 3900x or did I overlook something when I installed my CPU cooler?
 
ok, same issue here with 3950x, however use hwinfo64 to read both tdie temps. I was using coretemp prior and whatever sensor its reading is not correct. report back with what you find
 
depends on what you're doing to get "full load" certain micro-benchmarks that use avx will certainly push the temps there on AIO coolers ... hitting 80C is fairly normal. Custom loops will even hit 80C.

You can undervolt a bit, but that will impact your boost speeds a bit. Though if you're hitting high temps, you're already impacting your boost temps so you may end up with a net speed increase.
 
Too much or too little thermal paste?

Use a carbon pad. I did and it was wonderful!
 
Also - use the AMD utility to read temps. from what I've read / understand, other sensor apps tend to read temps on the high side.
 
What is "full load"? if it's Prime 95 Small FFT stress test with AVX then it's normal. If it's encoding a video with handbrake then it's not.
 
Yeah what are you using to push full load? 90 degrees is perfectly normal if running Prime95. Heck I have tried several different coolers both air and AIO. The one I am using now has kept it for the most part below the 95 degree max but unless you are running a custom loop you are not going to keep this Processor cooler than 90 at full load for an extended period of time. My AIO water temp raises almost 10 degrees C after about 30-40 minutes of Prime95 and the temps on the processor are around 80-90 for most of the standard blend test and if I do small FFT forget about it they hit 95. This is normal for these chips. This is with the chip running between 3.9 and 4.15Ghz. All temps according to Ryzen Master and HWInfo64. Air cooling causes the chip to throttle even more at full sustained load to keep it at 95.
 
Off and on subject, my 3960x hasnt even hit 70c under full load yet. I applied with the tip of my finger uneven coating and rather thin of Noctua paste from brown tube.

I've hit 100% cpu load for hours on end running h265 conversions.

When I mount a CPU I always tighten screws in a criss cross pattern then back them off a little, wiggle and wobble the waterblock, the. Tighten to completion.

I've never had a CPU not mount well using this method.

Again Inhighly highly recommend using a carbon pad instead of goop.

But I didnt have a carbon pad big enough for threadripper.

I need to load this CPU with prime 95 avx and see what it can really out out temp wise. Ill.do it this evening.
 
tangoseal said:
Off and on subject, my 3960x hasnt even hit 70c under full load yet. I applied with the tip of my finger uneven coating and rather thin of Noctua paste from brown tube.

I've hit 100% cpu load for hours on end running h265 conversions.

When I mount a CPU I always tighten screws in a criss cross pattern then back them off a little, wiggle and wobble the waterblock, the. Tighten to completion.

I've never had a CPU not mount well using this method.

Again Inhighly highly recommend using a carbon pad instead of goop.

But I didnt have a carbon pad big enough for threadripper.

I need to load this CPU with prime 95 avx and see what it can really out out temp wise. Ill.do it this evening.
Click to expand...

Honestly on this last cooler install I do not see it hit 95 unless I am running Prime95 on small FFT with AVX. I hit it on my previous Noctua NH-D15 and a Corsair H100i Pro but not the EVGA unless I do the smalls.
 
tangoseal said:
Off and on subject, my 3960x hasnt even hit 70c under full load yet. I applied with the tip of my finger uneven coating and rather thin of Noctua paste from brown tube.

I've hit 100% cpu load for hours on end running h265 conversions.

When I mount a CPU I always tighten screws in a criss cross pattern then back them off a little, wiggle and wobble the waterblock, the. Tighten to completion.

I've never had a CPU not mount well using this method.

Again Inhighly highly recommend using a carbon pad instead of goop.

But I didnt have a carbon pad big enough for threadripper.

I need to load this CPU with prime 95 avx and see what it can really out out temp wise. Ill.do it this evening.
Click to expand...

There's probably a number of reasons for this. TR undoubtedly uses much better binned silicon than Ryzen, particularly the 3900x. We see a big clue to support that with the 3950x vs 3900x Where the 3950x was pulling less power than the 3900x in spite of it having 4 more cores and a few MB more cache. TR likely has silicon quality at LEAST as good as the 3950x.

The giant heatspreader is also probably more efficient at transferring the heat away from the dies.
 
3960x avx p95 full load only pb2 no pb0 full custom loop with 7x120 worth of rads

THis is about 2 mins worth. No need to run it for an hour. I cant, in my wildest dreams ever heat soak my 7x120 worth of radiator slots lol Not even with a 700 watt load

upload_2019-12-3_3-8-18.png
 
I'm using CPU-Z bench to stress my CPU, and it goes up to 90c and above which is rather high.

Running Prime95 only maxes put at around 85c.
 
Prime95 running default settings after 30 minutes with an H100i

h100i-prime95.jpg


IMG_3899.JPG
 
ChronoDetector said:
I'm using CPU-Z bench to stress my CPU, and it goes up to 90c and above which is rather high.

Running Prime95 only maxes put at around 85c.
Click to expand...

In before someone tells you CPU-Z benchmark or stress test is not a legitimate test of your CPU...

I'm guessing you were using the beta with AVX? Were you using a Prime95 with AVX support?
 
Something to look for ... I just picked up my 3900x and I noticed that there was what felt like a left-over burr from manufacture or an anomaly from nickel plating all along the edge on two adjacent sides. It would catch on your fingernail and felt like those edges needed be 'broken' with sandpaper or a file. The blade of a fresh razor over light confirmed a high spot on those 2 edges for sure. Lightly sanding those two edges fixed it and definitely decreased the gap any TIM would have to bridge.

Stands to reason this could be on any similar IHS really.
 
  • Like
Reactions: N4CR
like this
tangoseal said:
3960x avx p95 full load only pb2 no pb0 full custom loop with 7x120 worth of rads

THis is about 2 mins worth.
Click to expand...

The first 5 minutes or so is a self test. You gotta run it for at least 10 minutes before it starts doing real work.
 
Been running RandomX (mining Monero) for over 72hours, 100% CPU load, 3900x and CPU temp is 82c. Seems to be higher then what Aida64 stress test gives which was around 76c which was similar to Prime95. 360mm radiator all EK stuff.
 
thesmokingman said:
The first 5 minutes or so is a self test. You gotta run it for at least 10 minutes before it starts doing real work.
Click to expand...

I had absolutely no idea it ran a self test. Now I HAVE to run it again ha ha and see what the temps really are.
 
tangoseal said:
I had absolutely no idea it ran a self test. Now I HAVE to run it again ha ha and see what the temps really are.
Click to expand...

In the first 5 mins temps dont go over 65c. After that, temps will go past 70 then 80... then ya know its doing real work.

Also, settings make a huge difference. Run blend if you wanna hammer it and really find out.
 
I know that my computer rebooted on a Prime 95 blend test and stopped in the BIOS with an over temp warning. This is in spite of HWInfo showing temps in the low 70s when it happened. Maybe I looked away and it rocketed to 95? I don't think that happened though. I had another test lock up when I was testing 4.3 GHz. It was at 85 C when the computer stopped responding. I had to poke it in the eye to turn it off. Corsair H100i platinum SE cooler with push pull fans.
 
Well ill run it for 20 mins. Cpus aren't going to keep climbing slowly in temp with a peak maximal workload.

I've ran prime 95 for 16ish hours once because I forgot it was running. This was on my 7820x. It just went to 85c and stayed ther the entire time. it wasn't like 789c because it grows like a nuclear reaction. I think some people overdo it by saying I ran it for 2 hours to see what my temps were. That's just wasting electricity.

Thier gonna make whatever heat they're gonna make by that point. No way prime 95 will ever saturate the loop I built. It can handle 3 GPus and a Threadripper easy. I know because I used to run quad crossfire 290x gpus years ago on the same setup. Push pulling Scythe Ultra Kazes in each radiator slot x7 slots.

Also some video work loads I do in 4k absolutely hammer my cpu doing hevc. CPU has never hit 71c to this day been hammering my 3960x since black friday.
 
  • Like
Reactions: N4CR
like this
Custom loops with lotta rads won't have an issue. It's the guys with AIO setups that will be pushed to their limits.
 
Look at that graph line. Its just flat .....

Prime 95 small FFT with AVX2 enabled

upload_2019-12-3_17-22-0.png
 
  • Like
Reactions: N4CR
like this
tangoseal said:
Look at that graph line. Its just flat .....

Prime 95 small FFT with AVX2 enabled

View attachment 204294
Click to expand...
Looks very good!
Can you run the RandomX benchmark? Here is quick walk through if you want to try it out, Windows version:
  • Download from github Benchmark V1.1.7
  • Type "command prompt" in search bar which will open up menu
    • Command Prompt Icon right click and run as administrator
  • In Command Prompt CD to the directory you placed RandomX benchmark
    • I usually just Cntl C (copy) address in File Viewer address bar and stick that into the console
      • In console:
      • CD /D "cntr+V path from file explorer"
  • In the console:
    • type in or copy paste:
      • .\ benchmark.exe --mine --largePages --threads 48 --nonces 800000 --init 24 --jit
    • When completed it will tell you the hashes, my 3900x is 13351.9 hashes. Wondering if the 3960x will double or better
  • To see what the parameters are, in console type:
    • .\benchmark.exe --help
EDIT: I could not get V1.1.7 to work, has VCRUNTIME140.dll is missing which is a 2015 Microsoft C++ redistributor. When I try to download from Microsoft it says it is already installed. So if you have this problem and still want to continue, V1.1.1 seems to work well without this issue.
 
Last edited:
noko said:
Looks very good!
Can you run the RandomX benchmark? Here is quick walk through if you want to try it out, Windows version:
  • Download from github Benchmark V1.1.7
  • Type "command prompt" in search bar which will open up menu
    • Command Prompt Icon right click and run as administrator
  • In Command Prompt CD to the directory you placed RandomX benchmark
    • I usually just Cntl C (copy) address in File Viewer address bar and stick that into the console
  • In the console:
    • type in or copy paste:
      • .\ benchmark.exe --mine --largePages --threads 48 --nonces 800000 --init 24 --jit
    • When completed it will tell you the hashes, my 3900x is 13351.9 hashes. Wondering if the 3960x will double or better
  • To see what the parameters are, in console type:
    • .\benchmark.exe --help
EDIT: I could not get V1.1.7 to work, has VCRUNTIME140.dll is missing which is a 2015 Microsoft C++ redistributor. When I try to download from Microsoft it says it is already installed. So if you have this problem and still want to continue, V1.1.1 seems to work well without this issue.
Click to expand...
Yes I will in a couple hours
 
Here is with just pb2 and I am running a slight -0.006 negative volt offset in bios
upload_2019-12-3_23-41-35.png


rebooting to turn PBO on and remove undervolt and see what it is then.
 
oops
tangoseal said:
Here is with just pb2 and I am running a slight -0.006 negative volt offset in bios
View attachment 204402

rebooting to turn PBO on and remove undervolt and see what it is then.
Click to expand...
oops try this:
  • .\ benchmark.exe --mine --largePages --threads 48 --nonces 800000 --init 48 --jit
Looks like you were limited to 24 vice 48 threads. Still with 24 threads, same as 3900x it was way faster. Thanks for testing. Forgot to change --init 24 to 48 which initializes the dataset with the number of threads.

Now if you want to use the benchmark for temperature testing and stress testing add another 0 or more to the nonces like from 800000 to like 8000000 or more. It will peg your CPU to 100% and generate a lot of heat.
 
Last edited:
Something is not right with those commands.

I do not know if it has to do with ram or what? I am only running 16GB dual channel right now on the 3960x. I need to wait a bit more before ordering more ram. Have gifts to buy for Christmas and all right now.

I got the same score on 24, 48, and 8 threads... see screenshot

upload_2019-12-4_3-26-27.png
 
I would think it was memory starved since algorithm is anti ASIC driven meaning a lot of memory access read/writes. You actually have half the memory bandwidth per core. Interesting results and lower then I expected. If you open up task manager and go to logical processors you will see what cores/threads are being used. If 48 threads then all cores/threads will be pegged at 100%, at least on the 3900x that is how it works. I take it Version 1.1.7 didn't work. It may also be the benchmark is buggy for Threadripper 3. Last test was 8 threads and 48 VMs. See if first test commands does indeed give 48 threads of work to the 3960x via task manager if you would.

Here is submitted benchmark results from a number of different CPU's and GPU's
https://monerobenchmarks.info/

If really interested, I can walk you through actually mining using RandomX, even for testing, if machine is unstable this will bring it out. It pushes the cores, temperatures and memory pretty hard. If so, then let me know and I will open new thread in the Crypto section.
 
Last edited:
noko said:
I would think it was memory starved since algorithm is anti ASIC driven meaning a lot of memory access read/writes. You actually have half the memory bandwidth per core. Interesting results and lower then I expected. If you open up task manager and go to logical processors you will see what cores/threads are being used. If 48 threads then all cores/threads will be pegged at 100%, at least on the 3900x that is how it works. I take it Version 1.1.7 didn't work. It may also be the benchmark is buggy for Threadripper 3. Last test was 8 threads and 48 VMs. See if first test commands does indeed give 48 threads of work to the 3960x via task manager if you would.

Here is submitted benchmark results from a number of different CPU's and GPU's
https://monerobenchmarks.info/

If really interested, I can walk you through actually mining using RandomX, even for testing, if machine is unstable this will bring it out. It pushes the cores, temperatures and memory pretty hard. If so, then let me know and I will open new thread in the Crypto section.
Click to expand...

Let me hold off until I get more ram. I do alot of davinci resolve work and need 16 more minimum for 4k work using Fusion.

I had a crap ton of ram that I sold. Now I regret overselling it ha
 
Last edited:
  • Like
Reactions: N4CR
like this
tangoseal said:
Let me hold off until I get more ram. I do alot of davinci resolve work and need 16 more minimum for 4k work using Fusion.

I had a crap ton of ram that I sold. Now I regret overselling it ha
Click to expand...
Ok sounds good. You might have to turn on VMs in the bios.
 
tangoseal said:
Too much or too little thermal paste?

Use a carbon pad. I did and it was wonderful!
Click to expand...

Looking into carbon pads...specifically the one made by IC...is it true that they can be inferior to high end paste? Other than that claim, which I am trying to find out more about, I can't find any reason NOT to use those pads...
 
knox1711 said:
Looking into carbon pads...specifically the one made by IC...is it true that they can be inferior to high end paste? Other than that claim, which I am trying to find out more about, I can't find any reason NOT to use those pads...
Click to expand...

I'm almost Gaurantee that the people calling them inferior have never tried them or is a complete troll.

I bought the IC diamond brand one and it must be mounted perfectly centered and tightly bound and your temps are like 1 or 2 c higher at peak over the best paste. When I sold my 3900x I pulled my waterblockand my chip was sparkly clean and I can reuse the carbon pad over and over and over and over. No alcohol cleanup, no paste staining your fingers and your equipment etc....
 
  • Like
Reactions: noko
like this
You must log in or register to reply here.
Back
Top