Author Topic: Threadripper & Ryzen only builds (3rd Gen starts on page 50)  (Read 200975 times)

2020-03-30, 10:35:15
Reply #1035

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4079
    • View Profile
    • studio website
thanks for your time :)

Last question on the power supply, 1000 or 1200w ? MAX 2 video cards.

Go 1200W for peace of mind. 1K is plenty for 3990X + Top GPU (both have capacity to consume 300W each, but will usually not run at same time that way) and will even cover the 2X GPU.
But PSU still benefit from higher capacity, running at lower load the Seasonic can for example run passively.

On the Quadro RTX 5000, I think if I was going to go for a card like that I would consider the RTX Titan instead. They are similar in price and the RTX Titan has 24gb of vram and is faster. I think the only advantage the Quadro RTX 5000 has (if you even need it) over the RTX Titan is ECC memory.

No contest, Titan RTX is the ultimate GPU right now, it's equiv to RTX 8000 Quadro, minus the memory on which it's same with 6000. But RTX 5000 can be snatched under 1500 Euro (And I bought all three of mine for 1300 each), while I was unable to buy Titan for anywhere less than 2700 Euro in Europe. I tried... for past 5 months ;- ) I have scripts that notify me of prices. 2000 Euro was my limit. It's not worth more.

Checked just in case eBay and all e-shop through Geizhals: https://geizhals.eu/nvidia-titan-rtx-900-1g150-2500-000-a1939621.html
2700 Euro. That's not worth the price esp. at this point. They're cheaper in US, but VAT + Customs will bring it back to same price as here.

The RTX 5000 is still overpriced at 1300-1500 Euro, I do suggest people to buy used 1080ti for 300 bucks, and Quadro is only valuable to people who need more than 11GB Vram. I have such specific needs, but was surprised to see it here :- ).
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2020-03-30, 18:23:37
Reply #1036

dfcorona

  • Active Users
  • **
  • Posts: 172
    • View Profile
Wow Juraj you are finding some great prices. Over here in the USA I cannot find the RTX 5000 under $2,150. Because of that I went with the 2080 ti, really good performance but for a pretty hefty price, payed $1,250 for it.  It is the Hydro AIO version, but way overpriced. I know the 1080 ti is better buy, I have one of those also but the 2080 ti is faster for GPU rendering and probably a little faster for Viewport. But if you don't use GPU rendering I would absolutely recommend the 1080 ti instead like you said.

2020-03-30, 19:05:14
Reply #1037

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4079
    • View Profile
    • studio website
There are periods of time where it's hard to source things for good price, but enterprise hardware usually goes for heavy discount because no one would pay such price individually in retail or from auction.

I don't think anyone buys Quadros for themselves in retail, 99perc. of them come in form of prebuilt workstation from Dell/IBM/etc.. and when they're not needed the sys admins pilfer them onto eBay and that way you end up with 50perc. discounts on brand-new enterprise hardware, mainly SSDs, but less often also GPUs, CPUs,...

For GPU rendering the RTX 2080ti makes plenty of sense, for pure Corona & Max, the different between RTX Pascals and Maxwell is very tiny, but that's because 3dsMax can't utilize much of the performance properly and the slowness comes mainly from object properties (like modifiers), not even their visual complexity often.
In meantime you can always enjoy Doom Eternal in high-detail :- ).

BTW RTX Titan is very peculiar, I think very few were sold in Europe and all the listings I've seen were in US as well. So brand-new or auction, this card uniquely still goes for MSRP here, making it very tough buy.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2020-03-31, 18:10:28
Reply #1038

pirillino

  • Active Users
  • **
  • Posts: 5
    • View Profile
Hi, I went from this build:

AMD Threadripper 3970x
MB Asrock Taichi TRX40
RAM G Skill DDR4 F4-4000C18Q-32GVK, Ripjaws V (4x8Gb 4000 Mhz CL18)

all is working wery well (also with PBO +200) for small hi quality object.

Now I'm sold the CPU and I'm planning to buy the 3990x, yes I know that I should increase the memory...
Do you think that I could have trouble if I buy another same kit, running the 3990x with 8x8Gb?
Could I have some frequency or CL problems ?
I would avoid to sell also the actual memory....
Thanks

2020-03-31, 22:02:17
Reply #1039

Leonardo Restrepo

  • Primary Certified Instructor
  • Active Users
  • ***
  • Posts: 13
    • View Profile
Hello Guys, after few days of having a couple of 3990x here are my thoughts and doubts that I want to share with you.
First here are the most important specs of the 2 builds in order to have an idea of the experience and performance.

Build 01
3990x
4 slot- HyperX 64GB KIT DDR4 3200MHz CL16 RGB FURY series (HX432C16FB3AK4 / 64)
Zenith extreme II alpha
ADATA XPG GAMMIX S50 1TB SSD

Build 02
3990x
8 slot -x4 G.SKILL 16GB KIT DDR4 3200MHz CL16 Trident From RGB (F4-3200C16D-16GTZR)
Zenith extreme II alpha
Samsung 970 PRO 1TB

1. first of all, the performance of this processor it's just amazing, solid 18 seconds in both  in the corona benchmark, slightly faster in the Build 02
/1/ in the other hands when I do the cinebench r20 I am having 24-24.5 as a score and I have seen some guys here have reported 25k,  build 02 is slightly faster

2. If I speak about memory here is the tricky part , after running several tests where I tried first the stock ram speeds, 2133 for build 2 and 2400 for build 1, it seems that in certain scenarios the stock speeds are just going faster than the 3200 MHz that both of the memory kits provided (3200 MHz)

-I have seen using stock memory speeds while doing rendering around 3500-3600mhz on all cores , looks weird right?
-when I switch to the 3200 mhz speed in both setups (and making sure the timing in both kits ends in 38) I am having while rendering around 3100-3300 mhz

3.Ryzen master Wierd behavior. this is really confusing since I see in the ryzen master a peak speed without doing anything around 800-1500 mhz , while in the task manager I see around 3.9-4.2 ghz, some of you are experiencing the same'?, I don't feel the builds slow or laggy just this is confusing and want to know why the ryzen master is behaving in that way.

4. Temps are just insanely amazing, never above 70 degrees while doing heavy renderings even in a pretty low airflow spot.


Let me know if you have experienced the same issues and also if you are using the same motherboard what are your speeds/Memory kits in order to see the differences and possible issues with the ram , also looking the best ram kit for this setup, (of course, as juraj suggested there are some really good ram kits coming to the market very soon)

Best regards
Leonardo


2020-03-31, 23:33:25
Reply #1040

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4079
    • View Profile
    • studio website
Hey, hope that everyone is still safe and well, still have work and enjoying their speedy chips :- ). So..

Quote from: Leonardo Restrepo
2. If I speak about memory here is the tricky part , after running several tests where I tried first the stock ram speeds, 2133 for build 2 and 2400 for build 1, it seems that in certain scenarios the stock speeds are just going faster than the 3200 MHz that both of the memory kits provided (3200 MHz)

-I have seen using stock memory speeds while doing rendering around 3500-3600mhz on all cores , looks weird right?
-when I switch to the 3200 mhz speed in both setups (and making sure the timing in both kits ends in 38) I am having while rendering around 3100-3300 mhz

Read my answer about memory to Michal from previous page, this is due to shared power constraint. I also write solution if you want to have both.

Quote from: Juraj Talcik
You can also just try PBO for moment (or keep it if you'll find the temps still good). Up the PPT limit to 300-320W, not sure the other two limits (TDC, EDC) are needed to modify for such small increase (but their ratio is basically 3:2:2 +/- ).
While the IO does eat up little bit of multi-core performance (but imho only perceptible in short benchmarks, anything past few minutes equalize), the single-core stays the same.

Zen2 chiplets, can eat up to 70+ W each +/-, so if 3990W was allowed to scale as well as 3960/3970X, it would be 500W chip. SoC only eats 20W at stock, but if faster memory makes it consume even slightly bit more (let's say 30W), it power starves the 8 dies, something that wouldn't be noticeable otherwise. 3970X would go from 65W to 63W, but 3990W will go from 32 to 30 per chiplet die. I am making these numbers up, I don't know what they exactly are, but the difference is that the chip will run slightly lower all-core boosts unless compensated for the additional SoC consumption. And this will only become perceptible difference on 3990X.

But the architecture still does benefit from that faster memory and increased infinity fabric. And if the loss makes it seems like tradeback, upping the PPT limit should solve it. I wanted to do such tests when my waterblock arrives, but 20+ W shouldn't make much of dent for 3990X, the CPU cools exceptionally well.

This is also partly the reason behind different Cinebench scores. But these are also influenced by power profile (both in bios and in Windows). That will affect what frequency you see in idle and boosts. Ryzen Master is simple tool, and Windows is innacurate, so you can get precise data of minimum, maximum, average,etc.. from HWiNfo.


Quote from: pirillino
Do you think that I could have trouble if I buy another same kit, running the 3990x with 8x8Gb?
Could I have some frequency or CL problems ?
I would avoid to sell also the actual memory....

You can buy the same kit again. What will happen is that it will not run at the XMP profile, but you shouldn't do it at the moment either because your frequency is too high for Zen2 chips.

Zen chips benefit from parity between memory frequency (half of MT/s because DDR is double-rate,) and Infinity Fabric frequency. Highest it can go on auto is about 3733, after that, the ratio will go from 1:1 to arbitrary because the infinity fabric cannot clock so high.
The highest officially supported FLCK (Infinity Fabric) clock speed is 1800 MHz, which correlates to 3600 MT/s memory. Anything above is extreme overclocking, and that is taxing and unstable on 3990X.

You will simply lower the memory to something reasonable like 3600/CL16. That should be still stable with 8GB high-end B-Die memory kits like you have.

Quote from: Leonardo Restrepo
as juraj suggested there are some really good ram kits coming to the market very soon

Yeah about that... I don't think anything is coming to market very soon anymore. I would focus on stocking a bit of pasta and toilet paper than waiting for memory :- ).




« Last Edit: 2020-03-31, 23:36:55 by Juraj Talcik »
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2020-04-03, 13:48:56
Reply #1041

giona4

  • Primary Certified Instructor
  • Active Users
  • ***
  • Posts: 132
    • View Profile
    • Giona Andreani Architectural Visualisation
There's anyone that updated the BIOS of an Asus Rog Strix X570-E Gaming to the latest one (1407)?
I'm always a bit scared when updating the bios.. I noticed that on my board there is a quite old 1005 bios.

On that mainboard there is a Ryzen 9 3950X and 128GB Corsair Vengeance LPX 3200Mhz.

2020-04-03, 21:11:49
Reply #1042

twoheads

  • Active Users
  • **
  • Posts: 225
    • View Profile
Possibly stupid question,

So I bought four 140mm noctua NF-A14 PWM's to replace stock fans in meshify S2 and maybe add extra one on top of the case (not sure yet) + bought extra fan for NH-U14S TR4-SP3 just in case. Fractal's stock fans are 3 pin, noctua's are 4 pin. There are only three 4pin slots on case's PWM fan hub. Should I return all PWM noctua's and buy NF-A14 FLX? Where should I plug cpu cooling fans and where the rest of the fans?  Also I'm super worried if NH-U14S TR4-SP3 wil fit in meshify.....

thanks
TH


Update: both cpu NF-A15 PWM's plugged directly to MOBO's CPU_FAN via Y cable and rest of the fans plugged to HUB's 4 pin, is this the way to go?
« Last Edit: 2020-04-03, 21:45:46 by twoheads »

2020-04-04, 00:09:15
Reply #1043

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4079
    • View Profile
    • studio website
both cpu NF-A15 PWM's plugged directly to MOBO's CPU_FAN via Y cable and rest of the fans plugged to HUB's 4 pin, is this the way to go?

It doesn't matter at all, you don't even have to use the case hub if you have enough plugs on your motherboard. Most high-end boards have lot of PWM headers like (CPU, CPU OPT(ional), Case 1, Case 2, PUMP, etc..) you can use directly or any combination with hubs. But your suggestion is pretty good as it would let you control the fans that should spin the same (like all intake) with single header.
I usually do all intake fans (3x) on one header (through hub or splitter), rear(1x) exhaust separately (since I want the exhaust to ramp up more than I want intake due to ratio) and CPU separately.

Regarding the CPU's fans, I highly suggest not to use two fans on thin heatsinks like U14s. I've recently read why it works so poorly (I tried it myself for two hours before I put it away). There is very little pressure needed to pass through this heatsink (unlike D15 or U14a) so the second fan will end up pulling the pushing fan a lot creating that annoying oscillating sound. Multiple fans in row don't improve airflow, just the pressure so they don't really make sense here (and the pressure starts to improve at higher RPMs, so also not situation with A15s).

(Apparently this can be alleviated by having the pushing (first fan) rotate lot less (lets say -200 RPM compared to pull) to avoid this issue. You would then have to connect the fans to CPU and CPU OPT header to control them separately.
But ultimately this is not necessary for the fact the rear exhaust fan is already doing enough pull. )


I know a lot of youtubers added that second fan in their videos but.. I doubt they did any test other than because it looks better and they thought it might work better in theory.

It will absolutely fit though, Meshify S2 has 185 mm CPU cooler clearance. Every case that can fit 140mm fan above rear IO has massive space for the cooler :- ).

There's anyone that updated the BIOS of an Asus Rog Strix X570-E Gaming to the latest one (1407)?
I'm always a bit scared when updating the bios.. I noticed that on my board there is a quite old 1005 bios.


Nothing to worry about, you can always use USB flashback to bring back earlier version automatically if something happens during update process. But this bios came out month ago, it should be pretty safe.
I would try it as well, but the 3950X is with our modeler in remote quarantine and she's not a PC girl so that will have to wait :- ).

Also... if you don't have any issues.. you technically never have to upgrade bios. At some point, chipset drivers might necessitate that but the current ones don't.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2020-04-04, 00:29:03
Reply #1044

twoheads

  • Active Users
  • **
  • Posts: 225
    • View Profile
Thanks Juraj,

  I'm gonna test it anyway but from the top of my head I would pick this configuration for start:

2x intake (one header, closed basement)  there is no reason to add third intake fan and blow air directly on PSU right?
1x fan on cpu cooler
1x exhaust fan

1x exhaust fan on the roof, closer to the end of the case optionally.

What do you think?

2020-04-04, 02:04:56
Reply #1045

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4079
    • View Profile
    • studio website
Ultimately, it doesn't make too much of difference how you place them but yes, this is the most common configuration.

The roof exhaust, should only be positioned at the rear if it should help CPU temperatures, although it does only very little if there already is plenty of positive intake.
Roof exhaust in such configuration mainly helps with GPU temperatures if non-blower cards are used (and can slightly worsen CPU + memory cooling).

I am personally fan of using full frontal intake (the bottom third fan doesn't just blow at PSU, but also towards the GPU) as you can run 3 intake fans slightly slower than 2.
It's the cleanest solution to have front-to-back airflow without complicating it with roof.

No reason to pontificate on it, it will work however you place the fans and orient them.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2020-04-04, 02:17:21
Reply #1046

dfcorona

  • Active Users
  • **
  • Posts: 172
    • View Profile
More fans the merrier, my case has 18 fans in it, Lol. heat always rises so I always have a top fan, but only at the very back or else you screw up your air flow.

2020-04-04, 17:00:37
Reply #1047

Oleg_Kuchmin

  • Active Users
  • **
  • Posts: 9
    • View Profile
Hi guys.

Could you please give any advice for cooling solution for Threadripper 3990X, do you have trouble with temps?
What will be the most effective solution, air or AIO?

I'd really appreciate for your help.
 

2020-04-04, 17:57:22
Reply #1048

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4079
    • View Profile
    • studio website
Could you please give any advice for cooling solution for Threadripper 3990X, do you have trouble with temps?
What will be the most effective solution, air or AIO?

Noctua NH-U14s TR4, works fantastically, great temperatures. 3990X is very easy to cool (regardless of how counter-intuitive it sounds). At least 5-6 people in this thread including me runs that setup.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2020-04-04, 19:09:04
Reply #1049

Oleg_Kuchmin

  • Active Users
  • **
  • Posts: 9
    • View Profile
Juraj,
Many thanks for your advice. I appreciate it.

What can you say about overclocking?
I saw your post earlier in which you have 3,4 GHz oc for all cores. What the temps do you have with Noctua?
Or may be I misremember.