Author Topic: Threadripper Builds  (Read 27515 times)

2018-12-10, 18:36:17
Reply #240

philipb

  • Active Users
  • **
  • Posts: 16
    • View Profile

Are there any better cooling solutions i should be considering?

It's been mentioned here a few times... The Enermax Liqhtec tr4 gen.2

You can see the Gen.1 performance advantage here. The issue was that gen.1 had all kind of problems that would lead to failure...
https://www.gamersnexus.net/guides/3089-threadripper-cooler-comparison-full-coverage-liquid-vs-air

Supposedly, the current (there were maybe even problems with the first batch of gen.2) Liqtech TR4's have all the issues resolved. I'm taking the gamble as the performance is clearly good. I have a gen.1 280 that is still working fine after 9months and its good stuff. But I will RMA it for a new Gen.2 just to be safe.

2018-12-10, 19:13:09
Reply #241

Vuk

  • Active Users
  • **
  • Posts: 42
    • View Profile
I am in the same spot as you are. Got a 280 working fine nearly a year now on the 1950x but I am not sure if you can RMA the cooler if its working ok now or they are taking it RMA's without any questions?

2018-12-11, 10:36:55
Reply #242

jpjapers

  • Active Users
  • **
  • Posts: 1120
    • View Profile

Are there any better cooling solutions i should be considering?

It's been mentioned here a few times... The Enermax Liqhtec tr4 gen.2

You can see the Gen.1 performance advantage here. The issue was that gen.1 had all kind of problems that would lead to failure...
https://www.gamersnexus.net/guides/3089-threadripper-cooler-comparison-full-coverage-liquid-vs-air

Supposedly, the current (there were maybe even problems with the first batch of gen.2) Liqtech TR4's have all the issues resolved. I'm taking the gamble as the performance is clearly good. I have a gen.1 280 that is still working fine after 9months and its good stuff. But I will RMA it for a new Gen.2 just to be safe.

Is it overkill if i dont plan on overclocking?

2018-12-11, 11:54:28
Reply #243

Juraj Talcik

  • Moderator
  • Active Users
  • ***
  • Posts: 3486
  • Tinkering away
    • View Profile
    • studio website
There is no such thing as overkill cooling. If you don't use the performance, you'll trade it in for silence.

And Enermax is of course doing some good old marketing, it's 500W rated...as long as you let the fans run at 2000rpm. It's not overkill in true sense at all, it's just much better solution than other available.
I would probably trust it right now.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik    My Behance portfolio
lysfaere.com   Something new

2018-12-12, 20:32:54
Reply #244

philipb

  • Active Users
  • **
  • Posts: 16
    • View Profile
Is it overkill if i dont plan on overclocking?

Juraj said it...but think of it this way.

Its not really overclocking anymore....these new CPUs are capable of automatic all core turbo boost. Overclocking used to be a project in itself, now its a feature.

  If you don't use it you are just leaving performance on the table, rotting fruit on the vine. The real question is, how much do you want to spend on cooling, and that breaks down into a performance/noise/cost trade-off.

In other words....just do it!

2018-12-13, 07:37:30
Reply #245

ivabox

  • Active Users
  • **
  • Posts: 26
    • View Profile
    • ivaBOX
So after reading tons of pages for this processor, I want to ask one last time:) Do you guys recommend 2990WX over 7980xe. In terms of performance I am sold, everything looks fine, but I am really afraid of that ram problem, we work with really huge scenes, generally it uses all of the 128GB of ram, so what would you guys suggest? (Also we couldnt not find 7980xe available right now in Turkey, but AMD is ready to ship)

2018-12-13, 11:46:03
Reply #246

Juraj Talcik

  • Moderator
  • Active Users
  • ***
  • Posts: 3486
  • Tinkering away
    • View Profile
    • studio website
Hi,

I have both. My personal suggestion is 2990WX being superior as overall package. Now regards to the "ram problem":

- What is it that concerns you ? The fact only half of dies are connected to memory directly ? This affects only single-threaded performance and is alleviated by recent Ryzen software which basically does automatical project lasso, it shuffles processes like 3dsMax and Photoshop into those physical cores that have direct memory access. With PBO, you keep high single-core performance on par with i9, so applications run on the same smoothness.

- Or maybe the fact that affinity fabric within dies together affects the overall performance even multithreaded due to latency. This is very application specific but even with this latency delay, Corona is still faster, and the other workloads where this manifests, like file compression, etc.. are usually very small times regardless. In practical terms, this is more about not using the potential of the CPU to the fullest within certain applications, but still come ahead of the i9.

Now the benefits of 2990WX:

- You can choose from two great boards (MSI MEG and Aorus Xtreme). For x299, there is only recently introduced MSI MEG. Otherwise you're stuck with bunch of 8-phased VRMs for most part.
- In future, next or the following year, you would still be able to upgrade to 48 or even possible 64-core Threadripper. I would say though, that this will only be possible on MSI MEG (or otherwise necessitize X499).
- Excellent thermals. Short of VRMs, for which you have better boards, 2990WX runs much cooler than i9 at comparative clocks. Yes i9 can be clocked slightly higher, but due to AVX instructions, it becomes hell-fire.
- Thus every 2990WX can be overclocked, even on air cooler ! But for i9, you need to win silicon lottery and get a good chip, delid the heatspreader and replace the paste, and then live with terrible accoustics ;- ).

I only kept my i9 because what I am gonna waste time&money now trying to sell it.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik    My Behance portfolio
lysfaere.com   Something new

2018-12-13, 14:30:06
Reply #247

ivabox

  • Active Users
  • **
  • Posts: 26
    • View Profile
    • ivaBOX
Thanks for the answer,

What we read from some of the reviews, people claim that AMD crashes much more than intel. Of course speed and performance is important but if the scene crashes a lot, there wont be any meaning. However as you do not comment on it, and also bios updates will cover most of it (as with other threadrippers before)there wont be any problem I guess. thanks again.

2018-12-13, 15:03:30
Reply #248

Juraj Talcik

  • Moderator
  • Active Users
  • ***
  • Posts: 3486
  • Tinkering away
    • View Profile
    • studio website
This is first time I ever hear about CPU architecture (AMD in this case) being the cause of crashing. That is some serious nonsense. People with stupid builds, cheap underpowered PSUs, odd idiosyncratic settings in bios, overclocked CPUs beyond VRM capacity, not enough voltage, million stuff where people are only to blame themselves.
Even the people who come to Corona forum with the "Only Corona crashes my PC to blue screen  !!", well no shit, you probably didn't stress your PC before enough with full AVX load (on top of GPU utilization). Your build is to blame, not the software.

Early bios versions are always hassle, but it's purely MB manufacturer's fault, and fully unrelated to CPU.


People also claim the earth is flat, sometimes it's better not to read the internet proletariat with their collective wisdom. I am pretty sure this rumor is from gaming forums because those are most guilty of building crappy PCs and strange settings.


None of my PCs crashed in years. They all are almost 24/7 at 100perc., the workstations at 1000+ Watt each.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik    My Behance portfolio
lysfaere.com   Something new

2018-12-14, 07:22:51
Reply #249

philipb

  • Active Users
  • **
  • Posts: 16
    • View Profile
VRM thermal issues... Time to rethink the ASUS ROG STRIX. Should have listened to Juraj.

I have received my Enermax TR4 Gen.2 360. This was an upgrade from my gen.1 280. It let me push the PBO to overclock a good bit further. However, after putting everything in the case (R6) I can see that the thermals on the VRM will be unsustainable.

Testing with prime-95 blend (which goes up and down in terms of stress) the VRM temps are touching 100C, with a 120mm fan 2 inches away blasting at 100%, shit. Running blend the temps get to 103, and are often in the high 80s to low 90s. FTT testing takes the temps to over 100C within 10 min, again with the same fan setup...crazy! Running corona the temp levels out at 94-95C around 30 min, don't think that's sustainable, is it? And its, loud, of course.

I'm attaching a graph, it shows a longer Blend run in prime 95, and then a short FFT run after a period to let everything cool off. It shows the problem very clearly. The performance is great, but the VRM is a disaster.

Juraj, I'm curious what the VRM thermals are like on the MEG? Now that I have a better AIO the VRM are showing themselves to be the weak link. I can still return the ASUS, and given the cash I already have in this system it seems a few hundred more for stability and maximized performance would be worth the tradeoff.

Damnit, I guess its time to tear this thing apart for now.



« Last Edit: 2018-12-14, 10:13:03 by philipb »

2018-12-14, 14:08:15
Reply #250

Juraj Talcik

  • Moderator
  • Active Users
  • ***
  • Posts: 3486
  • Tinkering away
    • View Profile
    • studio website
Don't panic :- ). Common mosfets (even average quality ones) are rated for 120C. But lot of circuits around can start to deteroriate at roughly 105C. I see you often pass this peak.

VRM is often poorly cooled by even the best airflow, which is why people come up with all sorts of hacks by trying to point a fan directly onto the VRM. It makes a bit of chaos in overall airflow but can help the temps up to 20C.

Regarding my VRM temps I think I would have to measure it again at different voltages. I really didn't touch that PC at all since I gave it to Veronika. I think they never went above 80.

Definitely try to get some fan pointed at the VRMs. People with Aorus often do so and they have 10 phases for the CPU (versus 8), better heatsink, and most people in this thread under-volted.

Btw, if the noise from Enermax feels annoying, you can swap the native fans for Noctua Sterrox (NF-A12x25). I know it's not budget solution but they're state of art in terms of airflow/accoustics.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik    My Behance portfolio
lysfaere.com   Something new

2018-12-17, 22:31:01
Reply #251

philipb

  • Active Users
  • **
  • Posts: 16
    • View Profile
Don't panic :- ). Common mosfets (even average quality ones) are rated for 120C. But lot of circuits around can start to deteroriate at roughly 105C. I see you often pass this peak.

VRM is often poorly cooled by even the best airflow, which is why people come up with all sorts of hacks by trying to point a fan directly onto the VRM. It makes a bit of chaos in overall airflow but can help the temps up to 20C.

Regarding my VRM temps I think I would have to measure it again at different voltages. I really didn't touch that PC at all since I gave it to Veronika. I think they never went above 80.

Definitely try to get some fan pointed at the VRMs. People with Aorus often do so and they have 10 phases for the CPU (versus 8), better heatsink, and most people in this thread under-volted.

Btw, if the noise from Enermax feels annoying, you can swap the native fans for Noctua Sterrox (NF-A12x25). I know it's not budget solution but they're state of art in terms of airflow/accoustics.

Okay, yes I have calmed down.

Running Corona the VRM seems to stay under 95C with a fan pointed at them. If Vray also stays under 100C (haven't tested yet) then I will likely keep the ROG STRIX board. Im also having an issue where the PC wont wake from sleep using AUS PBO, but if I use Ryzen-master it does. Ill see if i can get that fixed. If not i might consider another board.

Yes, Ill definitely change over to Noctua fans, as these Enermax ones are loud. That being said the Enermax gen.2 seems to be very good. Hopefully, they have worked out all the problems and it won't quickly fail like the Gen.1 versions.

2019-01-03, 18:08:18
Reply #252

michaltimko

  • Active Users
  • **
  • Posts: 272
  • Coronaut (c)
    • View Profile

2019-01-03, 23:43:52
Reply #253

philipb

  • Active Users
  • **
  • Posts: 16
    • View Profile
https://www.hardocp.com/news/2019/01/03/amd_ryzen_threadripper_2990wx_performance_regressions_linked_to_windows_bug

https://bitsum.com/portfolio/coreprio/

For slovak / czech speakers https://www.svethardware.cz/chaby-vykon-threadripperu-2990wx-byl-identifikovan-jako-dusledek-chyby-v-kernelu-windows/48337

Seems to slow down Corona benchmark by 2 seconds for me. Doesn't appear to have any impact on Vray benchmark. May still be a good idea to run, either way, to ensure that dynamic local mode is running as a permanent service. For 7-zip there is apparently some benefit.

Anyone else loses a second or two with corona bench?
« Last Edit: 2019-01-04, 00:09:34 by philipb »

2019-01-04, 11:41:06
Reply #254

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 8783
  • Marcin
    • View Profile
<slightly offtopic> looks like some Threadripper prices dropped recently (at least in Europe), which makes them even more affordable</slightly offtopic>
https://geizhals.eu/amd-ryzen-threadripper-1950x-yd195xa8aewof-a1664849.html
https://geizhals.eu/amd-ryzen-threadripper-1920x-yd192xa8aewof-a1664904.html