Author Topic: Threadripper & Ryzen only builds (3rd Gen starts on page 50)  (Read 520428 times)

2020-03-22, 20:05:03
Reply #1005

Leonardo Restrepo

  • Primary Certified Instructor
  • Active Users
  • ***
  • Posts: 15
    • View Profile
All good even as is. My personal notes:

- I would go with nVidia graphic for the single reason so that you can use Optix Denoiser during Interactive rendering in Corona. No issue with 5700 cards (apart from the driver shanenigans) but unfortunatately too many features depend on CUDA acceleration which is exclusively on nVidia for now.
- CMK64GX4M4B3200C16 With this kit (from B to C revision) it's unclear if you'll get B-die chips or Hynix, one works worse on Threadripper, and while you won't have issue with 64GB total memory, you might in future if you buy another kit and want to run them at 3200/CL16. Based on Amazon reviews, this particular kit is 90perc. Hynix, and people can't run it at 3200 on X570. X570/TRX40 have similar memory compatibility. I have personal experience where revision with Hynix wouldn't run at 3000 on my 3990X, but the b-die did up to 3200 without issue.

Which brings us to eternal memory question: How to even know which memory is made of highly-binned dies and which one not? There is limited online list with Samsung b-die kits, and perhaps even Micron Rev-e. Buying from brands that primarily source them (Crucial brand belongs to Micron, so their mid-to-high end modules are Rev-e automatically, while G-Skill built their reputation from sourcing b-die from Samsung mainly. Brands like Corsair buy absolutely everything, so you can never quite know and have better chance with their higher tiers but otherwise it's lottery).
I don't suggest wasting money on expensive memory, but usually the price difference isn't drastic enough right now to not search a little bit.



Hello Juraj,

Speaking about memory kits maybe you would answer this question pretty easy.

Some parts for my new 3990x workstation are on the way and I do have this memory from my 2990wx build

https://www.alza.cz/g-skill-16gb-kit-ddr4-3200mhz-cl16-trident-z-rgb-d4948815.htm

I have 64gb on my 2990wx and I have had many issues, Now moving these rams kits into the 3990x using Zenith Xtreme 2 alpha, do you think I will have ram issues too?

if the answer is yes, this, for example, this ram kit would give me better stability?

https://www.alza.cz/hyperx-64gb-kit-ddr4-3200mhz-cl16-rgb-fury-series-d5658326.htm

Thanks in advance for sharing so much valuable info!

2020-03-22, 20:16:07
Reply #1006

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4743
    • View Profile
    • studio website
Hi,

from head, I know nothing about any particular memory kit. You can just google it and hope that someone who has it answered it at some point. But..

- 3200 CL16 for 8&16GB modules, are 90perc. lower end Hynix, 10perc. lower end B-die. Most high-end kits were 3000/3200 C14.
- The linked HyperX is probably the same, it wouldn't be upgrade.

Zenith Alpha is now the best board for TRX40 in regards to memory compatibility, everyone tests their kits on it.
Keep the memory you have, and if it doesn't run stable & error free at XMP profile, then you can either lower the speed or sell and buy another. But you won't know until you try.

If you decide to sell it and buy another, wait for the upcoming 32GB 3600 CL16 modules from Micron and Samsung and use what you have until then. That is going to be the best memory for Zen2 chips.
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!

2020-03-22, 20:28:47
Reply #1007

frame9

  • Active Users
  • **
  • Posts: 27
    • View Profile
Hello Juraj

Thank you so much for your time.

I'll then go with the
Asus GeForce RTX 2060 SUPER TURBO-RTX2060S-8G-EVO
or
GIGABYTE GeForce RTX 2060 SUPER GAMING OC 3X 8G

Would you go with the Asus because of the Blower design?

Hm. Ok.
Does this mean I only run into issues if I want to upgrade to 128GB?
Unfortunately the memory support for the ASUS Prime TRX40-Pro doesn't give me a lot of options.
The only option for RAM with Micron Chip (if that's what you suggested) would be the Kingston HyperX Fury RGB DIMM Kit 64GB, DDR4-3200, CL16-18-18 (HX432C16FB3AK4/64)
But I think it's a little to big for the Noctua and the socket support is 1,2,4 (which means they are not tested for all 8 slots if i understand correctly) 

Thank you again,
Tobias
« Last Edit: 2020-03-22, 20:36:56 by frame9 »

2020-03-22, 20:42:25
Reply #1008

Leonardo Restrepo

  • Primary Certified Instructor
  • Active Users
  • ***
  • Posts: 15
    • View Profile
Hi,

from head, I know nothing about any particular memory kit. You can just google it and hope that someone who has it answered it at some point. But..

- 3200 CL16 for 8&16GB modules, are 90perc. lower end Hynix, 10perc. lower end B-die. Most high-end kits were 3000/3200 C14.
- The linked HyperX is probably the same, it wouldn't be upgrade.

Zenith Alpha is now the best board for TRX40 in regards to memory compatibility, everyone tests their kits on it.
Keep the memory you have, and if it doesn't run stable & error free at XMP profile, then you can either lower the speed or sell and buy another. But you won't know until you try.

If you decide to sell it and buy another, wait for the upcoming 32GB 3600 CL16 modules from Micron and Samsung and use what you have until then. That is going to be the best memory for Zen2 chips.

Thanks for the info!


I will do the respective tests when the shipping company brings me the parts (they are delayed !! xD)

2020-03-22, 21:13:46
Reply #1009

michaltimko

  • Active Users
  • **
  • Posts: 305
  • Coronaut (c)
    • View Profile
My 3990 build is running now. I wasnt able to boot/post, fans spinning at 100% speed and not even gpu signal. I thought its memory issue but i tried to flash it from usb drive and that helped.

Corona benchmark 17s, this cpu is monster. Whats blew my mind was FPS counter when i tried BF5. I have 1080ti and im getting 170-200fps on High settings. How is that even possible that 64 HEDT CPU with such low base clock is destroying some "gaming" cpus from intel. This is incredible.
Coronaut!(c)2011

Supporting Corona in commercial projects since pre-alpha

2020-03-22, 21:46:55
Reply #1010

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4743
    • View Profile
    • studio website


Unfortunately QVL lists are very limited. But all memory will work. It just might not work at its designed profile. Threadrippers put more stress on the memory, so kit that would run no problem at 3200+ on Intel, might only run at 3000.
The higher-end the memory, the better chance it will run.

But I can't give any concrete advice on which memory to buy. It's thousands of options.. there are threads on memory compatibility on Reddit and Overclock.net forums for those who have endless time to sink.. which no one has.

Any older Samsung B-Die kits are high quality (usually validated as 3000/c14,  3200/c14) or the newer Micron Rev-e (usually as 3600/c16). These were mainly sold as Trident series from G.Skill or Ballistics series from Crucial.
I can't give any concrete suggestion, I don't own any particular kit from it.

G.Skill for example has their own QVL list for TRX40 boards, here is link for Asus:
https://www.gskill.com/configurator?page=1&cls=1529635169&manufacturer=1524725352&chipset=1574737122&model=1574737611&adSearch2=Memory_Type%C2%A7DDR4,Tested_Speed%C2%A73200MHz,Tested_Speed%C2%A73600MHz,

Quote
Would you go with the Asus because of the Blower design?

The blower is always louder...in games that utilize it 100perc. For work it's equally silent. I am fan of blowers because you don't need such good airflow in case (which requires either more open/mesh type case, or higher speed of case fans, both which contribute to overall noise).
No particular suggestion as for model.

(My secret/no guarantees given is to buy used 1080ti for its 11GB Vram which can now go under <350 Euro second-hand, and wait for 3xxx Ampere series.)


Quote
(they are delayed !! xD)

Better wait for PC parts than food & toilet paper :- ).
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!

2020-03-22, 23:24:28
Reply #1011

frame9

  • Active Users
  • **
  • Posts: 27
    • View Profile
Alright

Thank you again for your input!
I will do some more research on the RAM

Have a good evening and take care,
Tobias

2020-03-23, 10:42:40
Reply #1012

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12711
  • Marcin
    • View Profile
3900x Update:
Part 1
Part 2
Part 3:
- slapped 2 extra fans in the case (one blowing from the top onto the CPU radiator, one on the front panel)
- connected the Noctua through the supplied low-noise adaptors
- enabled PBO in BIOS
- overclocked Infinity Fabric (from the stock 1800 to 1833 1867 preset)
- overclocked RAM (from the stock 3600 to 3733)
- enabled global C-state in BIOS (aka "cool'n'quiet" depending on the mobo manufacturer) - now I am getting lower temps, the system is more quiet, and most importantly the cores do go to sleep in Ryzen Master (they never went to sleep before)
With the above fixes I am now getting about 42C at idle and ~70-75C at stress. Corona bench time 1:09. :)
« Last Edit: 2020-03-23, 16:48:29 by maru »
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2020-03-23, 12:27:31
Reply #1013

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4743
    • View Profile
    • studio website
You don't have to keep Memory and IF strictly in parity (1:1), but I would consider it highly preferable. If 1866 as IF isn't stable, then I wouldn't overclock the memory up to 3733.

You went full Reddit-mode into tweaking last 0.005 perc. improvement :- ). It's not worth it for the eventual crash (it will inevitably come at worst moment).
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!

2020-03-23, 12:33:15
Reply #1014

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12711
  • Marcin
    • View Profile
You don't have to keep Memory and IF strictly in parity (1:1), but I would consider it highly preferable. If 1866 as IF isn't stable, then I wouldn't overclock the memory up to 3733.
You went full Reddit-mode into tweaking last 0.005 perc. improvement :- ). It's not worth it for the eventual crash (it will inevitably come at worst moment).

I read here (Polish warning https://www.benchmark.pl/testy_i_recenzje/goodram-irdm-pro-ddr4-test-pamieci-ram/strona/31723.html) that RAM @3733 is the most beneficial for Ryzen 3000. Then IF survived the slight overclock, so why not. :) I am not getting any crashes so far.
It's a bit like tweaking render settings. I reach a satisfying state, but then think "why not push it a little bit further" and change one more setting. Of course I had to reset BIOS many times already because it wouldn't boot, but the current seems to be super stable.
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2020-03-23, 12:58:54
Reply #1015

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4743
    • View Profile
    • studio website
The difference between 3200 and 3733 is...almost nothing outside of high-fps at fullHD in games like GTA V.

But the ratio between IF and Memory should be at 1:1, otherwise it defaults asynchronous, which in your case will because your IF (1833) is less than your memory (1866 since DDR is double MT/s for MHz).
So either make sure your IF is also 1867, or I would take the memory back.

With Zen2, asychronous divider doesn't affect stability anymore, but it lessens the latency which is the main benefit of...faster memory. This can be seen when they tested various exotic kits like 5000 MHz at which point, the divider is obviously asynchronous.
And the performance was not only not better, it was often worse.

Very few people can run IF at 1867 (or 1900), it requires highly binned sillicon, which is why it's easier to achieve on 3950X (which has all dies selected) than the lesser models (when only one or two will be binned), but it's still ultimately pointless.

You can look at these graphs, this is the same extreme 5000 MHz kit, run at various speeds (and IF dividers, the ratio is only 1:1 up to 3600).
Tested at:

DDR4-3200 18-26-26-46 1T (FCLK 1600 MHz)
DDR4-3600 18-26-26-46 1T (FCLK 1800 MHz)
DDR4-5000 18-26-26-46 1T (FCLK 1800 MHz)




I chosen a regular Workload (Blender), which doesn't care about memory speed, and GTA V, which is like the only game where you can perceive benefit. Neither benefits from faster memory past 3600.
The test above unfortunately didn't test ultra low timings (like 3200-3600 CL 14), but the timings do surprisingly less than frequency speed for Zen2. And it's only the frequency that affects the FLCK.

I've looked up your Polish link, and he tested 2x single-rank 8GB modules at 1.4-1.45V. That really isn't a scenario that can be run for work since that will be stable (well..."stable" like stable) only at that configuration but most importably it doesn't scale.
Running something like 4x dual-rank 16GB modules at 1.4+ Voltage is nothing possible for Zen2 unless you want to burn down your house :- ).
« Last Edit: 2020-03-23, 13:19:47 by Juraj Talcik »
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!

2020-03-23, 14:45:09
Reply #1016

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12711
  • Marcin
    • View Profile
Thanks for all the info. I am aware of the 1:1 RAM to IF ratio, but thought it wasn't an issue to use higher RAM clock than IF clock.
Just for clarification regarding the Polish link - I am using exactly the same kind of RAM they are using, and I have 2 x 16GB running at dual.

Update: sorry for all the confusion, I just checked and I am already running IF at 1867MHz and RAM at 3733.
« Last Edit: 2020-03-23, 16:24:09 by maru »
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2020-03-23, 16:47:06
Reply #1017

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4743
    • View Profile
    • studio website
Update: sorry for all the confusion, I just checked and I am already running IF at 1867MHz and RAM at 3733.

Ok, than you're all set and won a slight sillicon lottery :- ). Nice! Check the memory temps on HWiNfo to see if they are reasonable and if yes, I guess that's a stable setup for now.
Are you the min:maxing/munchkin type in RPGs :- )?

Disclaimer for anyone else reading this: None of this is necessary, it has barely perceptible <sub 1perc. benefit. It's mostly for good feelz.
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!

2020-03-26, 15:58:42
Reply #1018

dfcorona

  • Active Users
  • **
  • Posts: 290
    • View Profile
Having some weird issues, My CB score which was a solid 25k on almost every run is now around 24,700 average.  Very rarely spike to 25k on a rare run.  But bounces around 24,600 to 24,800. Not sure why.  Noticed something was wrong when I ran the Corona benchmark and am getting around 19-20sec.  pretty sure I was around 17sec.  Is anyone else finding there system has slowed, trying to figure out if it could be something I installed or something running in background. Possibly a windows update. I am running avast, but have been since the beginning.  I turned off it's real-time protection, no difference.

2020-03-26, 17:34:55
Reply #1019

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4743
    • View Profile
    • studio website
Having some weird issues, My CB score which was a solid 25k on almost every run is now around 24,700 average.  Very rarely spike to 25k on a rare run.  But bounces around 24,600 to 24,800. Not sure why.  Noticed something was wrong when I ran the Corona benchmark and am getting around 19-20sec.  pretty sure I was around 17sec.  Is anyone else finding there system has slowed, trying to figure out if it could be something I installed or something running in background. Possibly a windows update. I am running avast, but have been since the beginning.  I turned off it's real-time protection, no difference.

Well, the past few recent W10 updates did have security mitigations, so anything is possible. Benchmarks for reviews are also done on fresh installation for the sole reason to avoid any process interference. Eventually all those Adobe, Office, etc.. background processes will take some super tiny toll on scheduler. Your ambient room temperature probably isn't stable across all the time.

The CB20 sounds ok, I almost never hit 25k on stock. The +/- 500 is well within its standard measuring, it's still too short for 3990X. With Corona benchmark, I will oscilate between 17-18 second for 90perc. of time. The rest will jump to 20 second.
That can be absolutely anything, it's way too short of benchmark, the turbos and scheduling can influence far too much as well.

If either benchmark was 10 minutes long, I am confident the scores would be lot closer.

I do highly suggest to uninstall the Avast malware. There is no reason to use anything outside of the native Windows protections (or additional hardware firewall for companies), and if someone absolutely must, than NOD by Eset is the only reputable at this point.

I really wouldn't stress this.
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!