Build: 3970x, dual 2080ti, 8TB m.2 RAID = Render Monster

Indeed, doesn't take long with the noctua cooler, really a custom loop seems to be the only option with these.
maybe with prime but with real world usage (in my case rendering in cinema4d) the noctua does fine (if you dont overclock).
i managed already to bring it down to 72/73C max by repositioning fans and i will replace them with more and better ones the next days.
 
maybe with prime but with real world usage (in my case rendering in cinema4d) the noctua does fine (if you dont overclock).
i managed already to bring it down to 72/73C max by repositioning fans and i will replace them with more and better ones the next days.

Only takes like 5 minutes of running the tracker in davinci resolve and mine throttles.
 
Only takes like 5 minutes of running the tracker in davinci resolve and mine throttles.
hm, i doubt that davinci uses more resources than cinema4d, at least not in that amount. to me that looks like another problem...
 
Last edited:
maybe with prime but with real world usage (in my case rendering in cinema4d) the noctua does fine (if you dont overclock).
i managed already to bring it down to 72/73C max by repositioning fans and i will replace them with more and better ones the next days.

And? Did you run the prime test?
 
hm, i doubt that davinci uses more resources than cinema4d, at least not in that amount. to me that looks like another problem...

Most resolve features don't use that much, but the tracker uses 100% of the cpu and heats it up more than a continuous cinebench loop.

It can somewhat handle it if I leave the case open.
 
Don't scratch your head on getting the fastest storage. Sata SSD are good enough and keeps you out on plenty of Nvme and compatibility trouble. I recommend Raid 1 with 2 different brands of SSD. The speed of the rendering is not anywhere affected. If you want to accelerate these for some other tasks, use an NVME Optane and Primocache over the Radi1 and disable delayed writes from the cache (RAM or optane over SSD).
This is the only trustworthy good way.
 
thesmokingman :
just ran prime95 twice now (blend test) for about 20-30mins and i got a phase (maybe 3-4min) it peaked at 92C, but mostly it was under 80C.
all just stock here, asus zenith extreme 2, 3970x, noctua NH-U14S TR4-SP3 standard, mounted with conductonaut.
only two 120 case fans pushing the air out (underpressure). it is not even a great case (NZXT 510)

mikeo :
compared with this, your temps seem strange.

i ordered some nocuta fans with better performance / noise values than the two that came w the case
i will try a 2nd fan on the cooler, 2 fans out, 2 fans in. atm my goal is not to oc, but to get it as silent as possible at load / when rendering.
atm even when idle the fans always spin up and down a little... sounds like a neat summer breeze .-)
 
Last edited:
thesmokingman :
just ran prime95 twice now (blend test) for about 20-30mins and i got a phase (maybe 3-4min) it peaked at 92C, but mostly it was under 80C.
all just stock here, asus zenith extreme 2, 3970x, noctua NH-U14S TR4-SP3 standard, mounted with conductonaut.
only two 120 case fans pushing the air out (underpressure). it is not even a great case (NZXT 510)

mikeo :
compared with this, your temps seem strange.

i ordered some nocuta fans with better performance / noise values than the two that came w the case
i will try a 2nd fan on the cooler, 2 fans out, 2 fans in. atm my goal is not to oc, but to get it as silent as possible at load / when rendering.
atm even when idle the fans always spin up and down a little... sounds like a neat summer breeze .-)

Edit: nevermind with it set to default settings my board all core boosts mine to 4.2 then slowly drops to 4.1 when running prime95, so not quite stock. Also prime95 blend only shows 129w in ryzenmaster and it hovers around 60c, doesn't seem to really stress it like resolve tracking 4k60 footage does even though both show 100% cpu usage. Also resolve taxing the 2080ti at the same time probably doesn't help with case cooling. Looking forward to getting both parts in the 3x360 water loop when the parts arrive. Right now the case has 7 noctua fans in addition to 2 on the u14s.
 
thesmokingman :
just ran prime95 twice now (blend test) for about 20-30mins and i got a phase (maybe 3-4min) it peaked at 92C, but mostly it was under 80C.
all just stock here, asus zenith extreme 2, 3970x, noctua NH-U14S TR4-SP3 standard, mounted with conductonaut.
only two 120 case fans pushing the air out (underpressure). it is not even a great case (NZXT 510)

You only ran it for 30 minutes so you didn't give it time to heatsoak, yet it hit 92c anyways. That should reinforce the fact that you are already pushing it cooling wise. I would be wary of a long sustained load.

https://hardforum.com/threads/build...render-monster.1990145/page-2#post-1044427819
 
Last edited:
Posting some leftover pics with comments.

Vray bench cpu+gpu, in sli. 48K and change on the cpu is pretty darn good imo
20191219_152510.jpg

R20, cpu at 4.2 with stock memory timings vs tweaked
20191219_152619.jpg

Memory timings... using safe settings. Fyi, Dram Calculator does not work on Threadripper 3, so swap yer ram to a Ryzen system to run it. If yer using same memory modules, u can just use these too.
20191221_194807.jpg
20191221_194817.jpg
 
  • Like
Reactions: mikeo
like this
Btw, I didn't take any screens in games that I remember, but this rig maxed every title we tested. 60fps was pegged in everything. I did disable sync in Dying Light just to see how high it'd go and framerates were well over 120fps everywhere in DL at least.
 
  • Like
Reactions: mikeo
like this
You only ran it for 30 minutes so you didn't give it time to heatsoak, yet it hit 92c anyways. That should reinforce the fact that you are already pushing it cooling wise. I would be wary of a long sustained load.
... to me the test behaviour was a bit strange... it buildt up till ±80C and then after ca 10mins suddenly raised to ±90C for 1-2 minutes and then fell back to 80C again and stayed there for the next 20mins without any change ... cinema4d doesnt raise the temps that much, but i already ordered fans to get temps and noise down a bit...
 
... to me the test behaviour was a bit strange... it buildt up till ±80C and then after ca 10mins suddenly raised to ±90C for 1-2 minutes and then fell back to 80C again and stayed there for the next 20mins without any change ... cinema4d doesnt raise the temps that much, but i already ordered fans to get temps and noise down a bit...

P95 avx blend runs thru all the tests that p95 can run, that's why temps fluctuate because each test is simulating a different load. The important thing is that blend tests everything, system ram/imc included. The max temp you see in blend is more likely what you will see in a balls out 4 hour+ render session. That gives it time to heatsoak your cooling. These quick runs can be very misleading because you don't know what your cooling can "really" handle when everything is against it.
 
ok tx, i give it a longer try when i got my fans.

but just out of interest... would you say custom watercooling with a big external radiator (ie. 9x9) is a good way to get a (very) silent system?
 
Last edited:
ok tx, i give it a longer try when i got my fans.

but i have another question... would you think custom watercooling with an external 9x9 radiator is a good/clean way to get a (very) silent system?

External has definite benefits but it also brings its own set of problems. The pump, will it be run manual or pwm? If pwm, then you need a signal to the mb. How will you power the pump? External power or run a molex/etc to the pc? How far away will your hose runs be? A box of Primo LRT is only 10ft, so if you run it 10ft away you need two boxes of tubing. And you need QDCs! And along with those QDCs I recommend using proper bulkhead fittings with an external setup.

Technicalities aside, I have run two Phobya 1080 Xtremes in the past. They do offer a crapton of cooling. Each one increases the thermal load a ton since its soo much damn water. And yes, you can stick it anywhere well anywhere with a decent supply of fresh air. It is an easy way to get massive cooling if your case doesn't permit big rads. There's just a lot of small things to plan out.

https://hardwarecanucks.com/forum/t...eview-4x-1080-monster-radiator-roundup.60386/

There's another review out there that compares the Phobya 45mm to the MORA. The MORA is better but its a small margin.
 
Last edited:
tx for this... !
my thought was actually to maybe have it right next/behind the workstation and that the big surface
with steady running slow and silent fans would be quieter than case internal radiators/fans. (with a more limited surface)
 
optimus-threadripper-3-coverage_2000x.jpg

Gotta say, I'm very interested in this massive cold plate that is designed for a full 8 chiplet setup, aka 3990x. Their foundation block supposedly bests the next best block by 6c or more.



**I just ordered the AM4 block to test, wee...
 
Last edited:
I have a pile of EK parts still waiting for watercool to ship the heatkiller I ordered while they were on vacation.
 
For a render monster you need Windows 7 pro or ultimate. It's still faster than Windows 10, on Blender. Even if Windows 10 is catching up.
Or Linux ! But you may not find the best graphics drivers on Linux.
 
here some noctua fans received and with the first setup my 3970x is not cooler but louder ! ouch .-)
i get the feeling that a custom watercooling is mandatory to run a 3970x with acceptable noise :-/

mikeo:
did you get your parts ? it would be nice if you could report your experience and temperatures / noise after you built your loop ! :)

thesmokingman :
how silent/noisy is your system?
do you have some advices to get the system as silent as possible?
 
Last edited:
  • Like
Reactions: mikeo
like this
thesmokingman :
how silent/noisy is your system?
do you have some advices to get the system as silent as possible?

The rig in this build is completely silent except when it is doing extended max workload. And then the fan noise is not obtrusive. If you've ever heard Corsair ML120 fans at max speed then you'll know, its there but not annoying and can fade into background.

Silence, is all about having more radiator surface area than your heat production by a factor of at least 3 to 1 and then some padding. That means 3x120mm per block plus one or two. Thus if I were to be pragmatic about it, a 3970x and two 2080ti would equal 1080mm of rad surface and change. However we only went dual 480mm which is 120mm down from theoretical ideal not including padding. The cpu is not the biggest producer in this loop, its the two 2080ti overclocked. From idle to a 4 hour render session, water delta is 10c.
 
Also, I should do a good explanation of how to monitor temps on Ryzen and TR. First, avoid apps that do time based polling, for ex. afterburner, hwi64. About Hwi64, by default it is set to poll however we can change it to monitor based on clock speed by disabling polling in settings. Once that's done, setup hwinfo64 to use RTSS for overlay. I disable all monitoring in AB btw and only use hwi64 overlay to rtss. Also, since you now have the luxury to choose and rename your sensors, choose to monitor die average temp instead. This way you're not always freaking out at the transient peaks over individual cores which most apps show by default. The setup below is with average cpu temp...

20191221_001517.jpg

20191220_234152.jpg

Tutorial on hwi64 and rtss overlay





First read this to understand the flaw in poll based monitoring:

https://www.techpowerup.com/257312/...age-exaggerated-a-case-of-the-observer-effect

"We have determined that many popular monitoring tools are quite aggressive in how they monitor the behavior of a core. Some of them wake every core in the system for 20 ms, and do this as often as every 200 ms. From the perspective of the processor firmware, this is interpreted as a workload that's asking for sustained performance from the core(s). The firmware is designed to respond to such a pattern by boosting: higher clocks, higher voltages," stated Robert Hallock, AMD's head of technical marketing for processors. "So, if you're sitting there staring at your monitoring tool, the tool is constantly instructing all the cores to wake up and boost. This will keep the clock-speeds high, and the corresponding voltages will be elevated to support that boost. This is a classic case of observer effect: you're expecting the tool to give valid data, but it's actually producing invalid data by virtue of how it's measuring," he added.

Zen also brings a different way of monitoring cpu clocks

Effective clock vs instant (discrete) clock | HWiNFO Forum
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 worked quite well in the past, but is not longer sufficient. Over the years CPUs have become very dynamic components that can change their operating parameters hundreds of times per second depending on several factors including workload amount, temperature limits, thermal/VR current and power limits, turbo ratios, dynamic TDPs, etc. While this method still represents actual clock values and ratios reported match defined P-States, it has become insufficient to provide a good overview of CPU dynamics especially when parameters are fluctuating with a much higher frequency than any software is able to capture. Another disadvantage is that cores in modern CPUs that have no workload are being suspended (lower C-States). In such case when software attempts to poll their status, it will wake them up briefly and thus the clock obtained doesn't respect the sleeping state.

Hence a new approach needs to be used called the Effective clock...
 
Last edited:
20200110_224452.jpg

If you don't have a loop filter, I suggest ya get one with some QDCs. This is from when the first fill up. Look at all that gunk.
 
here some noctua fans received and with the first setup my 3970x is not cooler but louder ! ouch .-)
i get the feeling that a custom watercooling is mandatory to run a 3970x with acceptable noise :-/

mikeo:
did you get your parts ? it would be nice if you could report your experience and temperatures / noise after you built your loop ! :)

thesmokingman :
how silent/noisy is your system?
do you have some advices to get the system as silent as possible?

It's pretty quiet unless I use all the cores at once, still playing with fan profiles.

Noctua cooler with the 4.15 to 4.275 all core OC... cinebench loop after 5 min would get close to 95C and sometimes throttle, now with the same OC doesn't go above 65C. The video card seems to stay below 36C as well. The dual D5 pumps and 3x360 rads with noctua fans can definitely handle a 3970x.

First custom watercooling loop BTW, have always done AIOs in the past.
 

Attachments

  • 20200117_231724.jpg
    20200117_231724.jpg
    237.5 KB · Views: 0
  • 20200117_191636.jpg
    20200117_191636.jpg
    194.8 KB · Views: 0
  • 20200117_191630.jpg
    20200117_191630.jpg
    169.1 KB · Views: 0
mikeo : thanks for your update...

i guess the water cooling catches all those small peaks, that make fans go up and down all the time
but how is it about noise/loudness in the long run under sustained load (ie 30mins of rendering)

at the moment i took another direction... i m throttling the 3970x using the ryzen master app
you can go down to 180W and still loose only 10% of performance (cinebench r20 is now 15500)
at the same time the temps go down 5-10 degrees and the fans are way more silent,
also under sustained load.

for infos check this:
https://www.computerbase.de/2019-12...erformancerating-fuer-anwendungen-single-core
 
by the way... i get an error when opening ryzen master app now very often "Ryzen Master unable to initialize."
i think it happens if i change anything with my bios.. like change fan curves for example
does anybody know how to get rid of that?

ps. if you google you just get "reinstall chipset drivers" or "reset cmos"
i managed to get it to start again, but the error came back.. no idea what is the reason causing it
 
Last edited:
The only real option is to reinstall RM.
i think i tried that, too... you have to uninstall, then install again... but i'm quite sure it did not work...
what worked was to revert the changes i made in the bios.

but would be kind of odd, if ryzen master breaks on every bios change

by the way would your recommend the amd chipset drivers or the asus motherboard chipset drivers?
i think i had first the asus, then later also installed the amd ones ... (without uninstall first)
maybe i must uninstall all of these and re-install only one
 
I tend to go to the chip makers site for drivers. MB drivers are always months behind. Regarding uninstall, It does make it cleaner and makes it so the right driver is used.
 
Only takes like 5 minutes of running the tracker in davinci resolve and mine throttles.

Your really cooling challeneged then.

I can run a planar tracker in DR16 and I barely hit 50c.

I also find as fast as the 24core is, its still molasses compared to my 2080ti in exporting dr16 jobs

Where I found stellar performance was all yhe stuff you do in edit, sound, and color. Thats where threadripper absolutely shines

thesmokingman I use Amd direct as well for chipset drivers. The latest and greatest.

indexofrefraction dont use fan curves man especially if your water cooling. Water can hold 4.184 joules of energy per deg centigrade per cubic centimeter. It has a MASSIVE capacity to absorb heat. Let the water do the work. The fans will just cool the water down in due time at a static speed. Water doesnt react to spikes and dips. It has a slow ramp up and slow ramp down.

Water cooling and curves are what beginners do who do not fully understand water and its physical properties. That is NOT a stab at you. Im not calling you a beginner. I have no idea what your expertise is.
 
Last edited:
  • Like
Reactions: mikeo
like this
Your really cooling challeneged then.

I can run a planar tracker in DR16 and I barely hit 50c.

Yeah the noctua cooler can't really handle a 3970x, average temps are 30c cooler now under load with the new water loop.
 
Posting some leftover pics with comments.

Vray bench cpu+gpu, in sli. 48K and change on the cpu is pretty darn good imo
View attachment 212586

R20, cpu at 4.2 with stock memory timings vs tweaked
View attachment 212587

Memory timings... using safe settings. Fyi, Dram Calculator does not work on Threadripper 3, so swap yer ram to a Ryzen system to run it. If yer using same memory modules, u can just use these too.
View attachment 212588
View attachment 212589

Here is mine with process priority set to high in Process Lasso

upload_2020-1-18_18-33-10.png
 
Back
Top