Can Daz Studio use a 128 threaded CPU to render?

RexRed

I am considering the 64 core 128 AMD Threadripper, I am wondering if DAZ Studio is optimized to use all 128 threads?.

Another question, with this addon program FarmIt, would I be able to set the render to use 128 threads and a graphics card in one PC and also farm out the same job to a networked computer with a graphics card and mulpitple cores?

In the render settings there is a "Bridge Beta" would I really need FarmIt? Couldn't I use BridgeBeta instead to render over my own network?

 

 

Dim Reaper

There were some posts about this a couple of months ago.  From what I remember, Windows itself will not allow 128 threads to run in the same program, unless you are running a workstation version.

https://www.techradar.com/uk/news/amd-threadripper-3990x-128-thread-monster-is-too-much-for-windows-10-pro-to-handle 

Robinson

Dim Reaper said:

There were some posts about this a couple of months ago.  From what I remember, Windows itself will not allow 128 threads to run in the same program, unless you are running a workstation version.

https://www.techradar.com/uk/news/amd-threadripper-3990x-128-thread-monster-is-too-much-for-windows-10-pro-to-handle 

Windows Home can't but Windows Pro can.  They patched it sometime after that issue arose.

RexRed

Robinson said:

Dim Reaper said:

There were some posts about this a couple of months ago.  From what I remember, Windows itself will not allow 128 threads to run in the same program, unless you are running a workstation version.

https://www.techradar.com/uk/news/amd-threadripper-3990x-128-thread-monster-is-too-much-for-windows-10-pro-to-handle 

Windows Home can't but Windows Pro can.  They patched it sometime after that issue arose.

Windows Pro upgrade is 100 bucks from Windows Home, Windows Pro "Workstation version" is a 200 dollars upgrade.

Does Windows Pro (non workstation) handle 128 cores? Can Daz handle 128 cores?

I think Bryce has a limit of 4 cores.

I am not sure where I got the 99 core limit from, I read it on some program's limitations.

RexRed

I am thinking that a 128 thread threadripper would still not be faster than 1 1080ti.

 

 

Paintbox

GPU will always win with rendering, since it is very good at the kind of calculations you want with rendering. So while its great to have 128 threads, especially if you do compute intensive tasks beside rendering, it is indeed no match for a GPU solution. Check the benchmark thread.

TheMysteryIsThePoint

RexRed said:

I am thinking that a 128 thread threadripper would still not be faster than 1 1080ti.

We did some tests on @RayDAnt's benchmark topic. A 32 threadTR is about 6 times slower than a single 2080ti. Assuming all other things equal and generously assuming linear scaling, you're most certainly right.

But that's kind of like that funny line from the movie JFK when the main character says his dog is not very intelligent because he could beat it two games out of three at chess... not having to worry AT ALL about scene size and still getting probably 2060 like performance is kind of attractive. And if your scene doesn't fit on a Titan, then even that $4K CPU is a lot cheaper than the alternative GPUs onto which it WOULD fit.

A part of me is wanting to say "Yeah, man, do it!" just because I want to see  how it turns out. :)

And my obligatory Linux/Blender plug: Neither Linux nor Blender have any such arbitrary limits. And even if they did, you could just change them to suit your needs and re-compile.

 

Paintbox

TheMysteryIsThePoint said:

RexRed said:

I am thinking that a 128 thread threadripper would still not be faster than 1 1080ti.

We did some tests on @RayDAnt's benchmark topic. A 32 threadTR is about 6 times slower than a single 2080ti. Assuming all other things equal and generously assuming linear scaling, you're most certainly right.

But does it still scale with scene complexity? So indoor vs outdoor , hdri vs single headlamp, 100 mb scene vs 1000mb scene etc...

 

TheMysteryIsThePoint

Sorry, I only tried it on one scene, and only because I was asked to.

kenshaw011267

Paintbox said:

TheMysteryIsThePoint said:

RexRed said:

I am thinking that a 128 thread threadripper would still not be faster than 1 1080ti.

We did some tests on @RayDAnt's benchmark topic. A 32 threadTR is about 6 times slower than a single 2080ti. Assuming all other things equal and generously assuming linear scaling, you're most certainly right.

But does it still scale with scene complexity? So indoor vs outdoor , hdri vs single headlamp, 100 mb scene vs 1000mb scene etc...

 

The way iRay does CPU rendering it uses every thread available. Whether it scales nearly linearly is something I don't know but it should be close.

I have tested a dual socket 7742 system in DS and ran into what I think is a NUMA issue but DS and iRay ran fine. I never timed the renders because I was trying to diagnose why one CPU was being used and the other wasn't.

JamesJAB

Paintbox said:

TheMysteryIsThePoint said:

RexRed said:

I am thinking that a 128 thread threadripper would still not be faster than 1 1080ti.

We did some tests on @RayDAnt's benchmark topic. A 32 threadTR is about 6 times slower than a single 2080ti. Assuming all other things equal and generously assuming linear scaling, you're most certainly right.

But does it still scale with scene complexity? So indoor vs outdoor , hdri vs single headlamp, 100 mb scene vs 1000mb scene etc...

 

Back a few years ago when I was rocking a dual 6 core Xeon setup (12 cores, 24 threads) I did an experiment disabling threads in the task manager.  Iray scaled very linear with a consistant itteration speed per thread.

I was a quick 100 itteration render test using the at the time current Iray benchmark scene:

--24 Threads
31.29 Sec (7.50 Sec per itteration / thread)
--20 Threads
36.80 Sec (7.36 Sec per itteration / thread)
--16 Threads
44.90 Sec (7.19 Sec per itteration / thread)
--12 Threads
59.10 Sec (7.09 Sec per itteration / thread)
--8 Threads
90.81 Sec (7.26 Sec per itteration / thread)
--4 Threads
190.45 Sec (7.16 Sec per itteration / thread)

RexRed

I have 12 cores 24 threads (non-overclocked) I9 and using that compared to 1 x 1080ti there is no contest the 1080ti is much faster.

2000 x 2000 only a sphere in the scene takes 8 seconds 1080ti

2000 x 2000 only a sphere in the scene the i9 takes 1 minute and 30 seconds 

My graphics card is 11 times faster than my CPU and I paid 1000 dollars for my CPU and 1000 for my graphics card.

Turning on the I9 slows down the 1080ti a tiny bit

When I turn on my second 1080ti it takes 5 seconds to render

So, my conclusion, I am not going to spend 4,000 dollars for a 64 core chip that even with 128 threads may not render as fast as one 1080ti, I would be better off getting another 44 lane 12 core I9 PC and putting three titans or better in it.

I suppose it would speed up the render having 128 threads it might be equivalent to another graphics card, but the cost is really high.

Once again, I would be better off getting a third computer with more titans in it than going with a 128 threadripper.

Then just network link them all.

JamesJAB

Honestly if I was looking at the posibility of spending $4000 on a rendering upgrade for my computer, I would be looking into adding a Titan RTX because of the 24GB of VRAM on the card.
First off there will be a huge rendering speed improvement because of the RTX architecture, then there is the 24GB of VRAM for rendering larger/more complex scenes, and finally ther eis the price.  At $2500 that leaves you with enough budget for a pretty beefy upgrade for the rest of the computer.

Chumly

Is the Threadripper your only AMD Consideration?

Have you looked at the AMD Ryzen 9 3950X ($750) or the AMD Ryzen 9 3800X ($459).  Lots of threads and PCIE 4.0