Author Topic: Threadripper & Ryzen only builds (3rd Gen starts on page 50)  (Read 188935 times)

2020-01-07, 16:42:59
Reply #750

dfcorona

  • Active Users
  • **
  • Posts: 172
    • View Profile
I am 50/50 on whether to build 3970X or 3990X also.  Will wait and see more info on the 3990X, but so far the 3970X is a nice option. Need to see the 3990X in real world situations.

2020-01-07, 16:57:22
Reply #751

Juraj Talcik

  • Moderator
  • Active Users
  • ***
  • Posts: 4050
    • View Profile
    • studio website
I am 50/50 on whether to build 3970X or 3990X also.  Will wait and see more info on the 3990X, but so far the 3970X is a nice option. Need to see the 3990X in real world situations.

Definitely, hope we get to see a lot of useful benchmarks and not 15 pages of Tomb Rider in 1920x1080p (ehm...Anandtech and every youtuber) :- ).

What's your go-to benchmark website?

I don't have any myself. I'll just skim-read everything that will come and extrapolate from it something. I am happy for everyone who will run Corona/Vray/Blender but also SPEC and various latency based benchmarks.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2020-01-07, 19:08:27
Reply #752

JoeVallard

  • Active Users
  • **
  • Posts: 129
    • View Profile
    • Joe Vallard
Definitely, hope we get to see a lot of useful benchmarks and not 15 pages of Tomb Rider in 1920x1080p (ehm...Anandtech and every youtuber) :- ).

This thing is a production beast! Heres 50 benchmarks where 99% of the software is video editing and a Cinebench R20 score.

I'd like the see the rendering benchmarks switch over to time based scores like Corona and Blender our because seeing some random 25,399 score is cool and all. But considering render time is what were interested in... maybe vray/cinebench use a score base cause its a big numbers thing that gamers like.

Would also like to see the benchmarks do a bench on a resolution you would actual use and not something under 720p.


Sidenote Cinebench R20 was just released last year on March 5th because modern cpu's were too fast to get consistent results on R15. Here we are less than a year later and we know the 3970x is already doing it in around ~16 seconds.

How many workstations do you have Juraj? Just ran a cinebench r20 with my 2950x and the 3rd gen threadrippers are such a performance increase but so is the price.

Price per core of Ryzen 3000s. The 3990x actually is cheaper per core, not but much but its also not a step up in price per core which you see throughout the stack.
3600x - $33.16
3700x - $41.12
3900x - $41.58
3950x - $46.87
3960x - $58.29
3970x - $62.46
3990x - $62.34
« Last Edit: 2020-01-07, 19:57:10 by JoeVallard »

2020-01-07, 20:13:14
Reply #753

twoheads

  • Active Users
  • **
  • Posts: 224
    • View Profile
Two questions for Juraj,

1- do you think Noctua NH-U14S TR4 SP3 will handle 3970x properly? Long runs under full load? If not can you suggest any other solution except watercooling?

2- is there any reasonable and cheaper competition for BenQ PD3200U (for max2016)

TH


2020-01-07, 20:26:25
Reply #754

Michael Arch-Viz

  • Active Users
  • **
  • Posts: 63
    • View Profile
Two questions for Juraj,

1- do you think Noctua NH-U14S TR4 SP3 will handle 3970x properly? Long runs under full load? If not can you suggest any other solution except watercooling?

TH

i have the same question. I'm afraid of the water system because of the leak :D

2020-01-07, 20:29:36
Reply #755

dfcorona

  • Active Users
  • **
  • Posts: 172
    • View Profile
Two questions for Juraj,

1- do you think Noctua NH-U14S TR4 SP3 will handle 3970x properly? Long runs under full load? If not can you suggest any other solution except watercooling?

TH

i have the same question. I'm afraid of the water system because of the leak :D

Just go AIO, with newer solutions they seem the way to go.  I'm running one on a 1950x, great cooling. Waiting to upgrade to possibly the 3990x then will be looking for AIO again.

Wonder why they went with 3990x and not 3990wx, wonder if there is something else possible coming, higher clock or TRX80, WRX80.

2020-01-07, 20:38:03
Reply #756

vlado

  • Chaos Group Team
  • Active Users
  • ****
  • Posts: 36
    • View Profile
Quote
I'd like the see the rendering benchmarks switch over to time based scores like Corona and Blender our because seeing some random 25,399 score is cool and all. But considering render time is what were interested in... maybe vray/cinebench use a score base cause its a big numbers thing that gamers like.

We use scores because they are more accurate - the current V-Ray benchmark runs for 1 minute and we count how many samples were computed in that time. Our previous version was time-based, but with the latest CPUs that time got down to a few seconds and in that case the statistical variation from different runs became quite large compared to the rendering time itself - which meant that the benchmark became less precise and less useful. With a time-based benchmark it was difficult to encompass all the different CPU configurations out there while still making sure that the benchmark completes in a reasonable amount of time. Switching to a score-based benchmark makes it more accurate and a bit more future-proof.

Best regards,
Vlado

2020-01-07, 20:54:49
Reply #757

Juraj Talcik

  • Moderator
  • Active Users
  • ***
  • Posts: 4050
    • View Profile
    • studio website
1- do you think Noctua NH-U14S TR4 SP3 will handle 3970x properly? Long runs under full load? If not can you suggest any other solution except watercooling?

There is nothing else you can use anyway. All other TR4 dedicated coolers (which are those whose block fully covers the heat spreader and thus all dies underneath, basically, it sits on top of the whole chip) are less or more inferior.
Those are BeQuiet! DarkPro TR4, ThermalRight SilverArrow TR4, CoolerMaster WraithRipper and most recent Arctic Freezer 50 TR4. At best they will give you the same temperature, but mostly somewhat worse.

It's actually easier to cool 3970X than weaker 3960X, maybe because of the die layout. It will still run rather hot, easily in 80C in common room temperaturs (25C ambient). It is what it is. I suggest really good case airflow and preferably blower GPUs.

Now to water cooling, little bit long-winded:
- It was possible to water cool 1st generation because they were low-TDP chips. 1950X can be cooled easily by anything on market. This is not good proof for watercooling.
- It was possible to water cool 2nd generation top chip (2990WX) just "so-so" with most AIOs because they aren't made for Threadrippers, they didn't cover the dies properly (they all use the same Asetec components, just slap their logo on it).
  The only exception was Enermax LiqTech which went through 4 revisions and still is very unrealiable product that suffered leaks, corrotions or just faulty pump.
- With 3rd generation, the dies under heatspreader are more close together, so with interesting twist of fate, it's once again possible to "sort-of" cool them with common AIOs, but they will not give you better performance than Noctua unless it's the Enermax LiqTech which still remains unrealiable product best to be avoided.

Personally, I am not fan of AIOs, imho they are trashy products for most part that use thin (28mm) radiators, poor CPU blocks, often use aluminium instead of copper, use loud fans,etc.. They are for "most-part" reliable, but air cooling is 100perc. reliable. When they don't offer any performance advantage, I don't see much use for them at all.

I would either go with Noctua, or build custom loop from quality parts.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2020-01-07, 21:43:29
Reply #758

twoheads

  • Active Users
  • **
  • Posts: 224
    • View Profile
Thanks Juraj, :)

Can hardly wait for news from danio1011 about his setup.

2020-01-07, 21:52:31
Reply #759

JoeVallard

  • Active Users
  • **
  • Posts: 129
    • View Profile
    • Joe Vallard
Quote
I'd like the see the rendering benchmarks switch over to time based scores like Corona and Blender our because seeing some random 25,399 score is cool and all. But considering render time is what were interested in... maybe vray/cinebench use a score base cause its a big numbers thing that gamers like.

We use scores because they are more accurate - the current V-Ray benchmark runs for 1 minute and we count how many samples were computed in that time. Our previous version was time-based, but with the latest CPUs that time got down to a few seconds and in that case the statistical variation from different runs became quite large compared to the rendering time itself - which meant that the benchmark became less precise and less useful. With a time-based benchmark it was difficult to encompass all the different CPU configurations out there while still making sure that the benchmark completes in a reasonable amount of time. Switching to a score-based benchmark makes it more accurate and a bit more future-proof.

Best regards,
Vlado

Ty for this info this make a lot of sense to do it like this! I figured it was similar to Cinebench and was still a score based on time, but rendering for a minute regardless of cpu speed really does future proof cpu scores and elimate the large variation you may get from renders going to fast.


I would either go with Noctua, or build custom loop from quality parts.

Odd that its either $80 or $1000+ for cooling solution. Nothing really in between. My Noctua is providing good cooling and barely can hear it, for my 2950x, OC to 4.0GHz. Stays in the low 70c with the ambient temps being around 21-23c.  Looked into a custom loop just because when I have nothing playing music/video the video card can get rather noisy.

2020-01-07, 22:02:46
Reply #760

Juraj Talcik

  • Moderator
  • Active Users
  • ***
  • Posts: 4050
    • View Profile
    • studio website
Talk about devil heh.. apparently this was shown on CES as "concept" https://www.anandtech.com/show/15303/ces-2020-asus-showcases-concept-420mm-aio-for-trx40

Quote
Odd that its either $80 or $1000+ for cooling solution.

There actually is but not for TR socket. You can buy pump+block combo like this (https://www.aquatuning.us/water-cooling/cpu-water-blocks/cpu-water-blocks/23937/alphacool-eisbaer-lt-solo-black) and add your own radiator and fans.
It will be 200-250 +/- Euros but pretty great inbetween. You can go with 420 or 480 40-50mm thick radiator with Noctua fans and it will blow every AIO on market off the table but still cheaper then full loop.

If only there was such pump+block for TR4 ;- ).
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2020-01-07, 22:28:08
Reply #761

twoheads

  • Active Users
  • **
  • Posts: 224
    • View Profile
GIGABYTE TRX40 AORUS XTREME is really worth extra money compared to ASUS ROG ZENITH II EXTREME?

2020-01-07, 22:32:40
Reply #762

dfcorona

  • Active Users
  • **
  • Posts: 172
    • View Profile
We'll have to see what else is coming.  IceGiant's ProSiphon looks really good but still in preorder.  A while back they spoke of a 3990x Black Series that comes with it's own Water cooler packaged. There might be much more to come and maybe even Noctua will have something to show. I cannot speak for Gen 2 Threadrippers cause I usually skip a Gen unless something ground breaking happens.  Also went the GPU route for a while. But AIO has worked good in my workstations for years.  Overclocked good with 65 degree readings. The thing I always hated about Noctua was the space it took up, impossible to work around without taking out of the case first.  I see that Dr. Lisa Su is saying they will most likely get better than 25k in cinebench after optimizations, with them and there partners.

2020-01-07, 22:35:39
Reply #763

michaltimko

  • Active Users
  • **
  • Posts: 295
  • Coronaut (c)
    • View Profile
I see 3990 more like node/work horse where i can send some stuff and let it work. For every day use / occasional gaming / multitasking, 3970 looks like a better choice , mainly for having much higher base clock. Even on 2990, you can feel that 3.0 in some scenarios where its not as snappy as it should be. Correct me if im wrong.
Coronaut!(c)2011

Supporting Corona in commercial projects since pre-alpha

2020-01-07, 23:06:04
Reply #764

Juraj Talcik

  • Moderator
  • Active Users
  • ***
  • Posts: 4050
    • View Profile
    • studio website
The base is bit lower than I expected, but I wouldn't worry about it too much. If the boost works just as well as it does on 3950X, it never even gets near the base. (from all tests I ran on 3950X which has 3.5 GHz base, it's always at 4 to 4.1 when rendering all-core, and 4.2-4.55 when up to 4 cores are utilized, so it covers almost all the workstations tasks. Even during rendering, it will still boost up single-core for Photoshop action for example, it works really nicely).

But otherwise looks like everyone here is on the same ship, 3970X is simply far too good. Let's hope the benchmarks will come sooner than 7th Feb.

GIGABYTE TRX40 AORUS XTREME is really worth extra money compared to ASUS ROG ZENITH II EXTREME?

Not sure if anyone can categorically tell. Aorus went extra mile to really go over the top with the build and features with price that follows (particularly the Intel dual-nic), but they both have fantastic VRM, heatsinks and shit ton of m.2 slots.

One thing I can tell, is that Asus 'looks' better than it actually is, while I really enjoy the ROG aesthetics (minus rgb) and just today received Rampage Omega for my multi-media/gaming PC where I'll slot the i9, there are so many complaints about the lack of working bios, various design faults,etc.. Did you guys know that almost half of those uber Dominus boards for Xeon 3175X were faulty and couldn't boot due to misplaced conductors? Asus lives on brand-name recognition, but quality control is not what it used to be.

For Zenith:
- E-ATX fits almost everywhere, in majority of cases, not just mid-towers, even some small towers!. It's just wider.
- Easier access and better cooling of m.2 drives due to Asus's proprietary "DIMM.2" addon card that slots next to memory. This is really nifty thing.
- Lot more popular, brands test their memory on it to make sure it work. So G.Skill has kits that were tested and guaranteed to work on Zenith II, but not on Aorus. This gives you a peace of mind when buying 1000+ Euro memory.
- If you only ever plan to add dual GPUs, the layout is sort of better.

For Aorus:
- If you ever planned on Quad-GPU rendering, then this is the board for it.
- Intel 10GBe is of much higher quality than Aquantia, both on hardware and driver level. It's not dramatic difference but if you want the best...also comes with two 10GBe nics so you can go 20GBe straight on.
- Passive cooled VRM. TRX40, just like X570 has already tiny and annoying chipset fan. No need to have more of them on board.
- Sturdier build. Both boards have backplate, but Aorus is full-cover and more heatsink connected, might even actually help with temperatures, not just aesthetics.
- General consensus on UEFI is that Gigabyte is really trying hard at the moment and might be the best player right now when it comes to care.

You can't go wrong with either. And if you don't need 10GBe, you don't even need to go with these. More affordable Aorus Master and Asus Strix are also good boards with same VRM.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika