Question I want to upgrade my CPU from i7-8700 for the soft body physic computation of more than 3 people

[ruinousdeity]

Hi there, I'm a totally newbie just knew and bought the software from the steam.
So this is my first post here.
I am doing some mass data analysis, statistical modeling, and AI studying for a living.
Basically I use my desktop for the aforementioned tasks, so I am now running "an old i7-8700 CPU with an AMD RX6950XT graphic card."
Cpus were not so important for my previous jobs.

Of course, doing so will not make the software run very very smoothly, especially with the soft body physic computation.
And there are some demo YT videos
Sorry for the Chinese system UI because I am form Taiwan. I guess you guys can understand the values.
You can see the physical calculation effects look great but the frames and sounds are lagging and skipping stitches.

The same scene on my i7-1260P laptop+ AMD VEGA56
Still low FPS but the time frame and sound are correct

And this evening I also changed the setting in my desktop.
Same low FPS, but much better. I do know the reason of laggy frames is my i7-8700 can't match the RX-6950XT.

There is only 1 person dancing with the soft body physic computation,
but I want to try to make more than 3~5 people (the limits will be 15 I guess) with time line control and the soft body physic computation.
So, which CPU can afford this amount of computation?

Thanks for you guys' reply.

 

[HolySchmidt]

None, you are insane

I own a 13900K and it's the first CPU I own (after a 9700K and a 12700K) that makes it possible to run scenes with 3 people (male/female/female) pretty smoothly (around 50fps, 12700K was limited to around 25fps, same scene). It also depends on male or female. Females do have way more physics to calculate in VaM. So if you wanna create a scene with 5 females, forget it. No CPU can handle this right now.
I know there's a way to render a scene, but I don't have no knowledge about this.

If you wanna increase your fps big time, go to the "glute physics" tab from a female and disable the "soft physics". This should increase your fps massively. Also, the hairstyle is a really big performance killer. There's so much different hairstyles, play around with it and have a look at your performance.

The problem with VaM 1.X is: it only uses a specific amount of your CPU cores, not all. It's limited by the engine version of Unity. VaM 2.X will open the gates but no one knows when it will be released. But if you want the best right NOW, get a 13900KS or wait until February when AMD launches their new 7000X3D CPUs. We then will see how they perform. Maybe better, maybe not. A 13700K for example should be a big upgrade for your sytem and your physics time.

//EDIT
And yeah, like @DJ said below my post: you want all the single core performance you can get. The more, the better. And believe me: a 13900K is not enough, but it's a pretty good upgrade from a 12700K.

 

[DJ]

The version of Unity Vam uses forces most body physics calculations onto a single thread. So for maximum performance, you should be looking for a CPU that has the greatest possible single thread performance you can afford.

Hair and clothing computation is on the GPU.

That said, 3-5 people will be pushing even the very fastest available components to their limits. 15 is simply not possible.

Check out the Benchmark results discussion. You'll be able to estimate how much faster various systems are.

Looking at this cpu single thread performance chart, I see that your cpu scores 2656, and the top of the line i9 13900KF scores 4700.

So at a rough estimate, you should expect that upgrading to that CPU would improve your framerate to about 1.76 times what you see now.

 

[ruinousdeity]

None, you are insane
..

Thanks for your reply.
I think I should be pretty normal. LOL
I will reply both of you at same post.

The version of Unity Vam uses forces most body physics calculations onto a single thread. So for maximum performance, you should be looking for a CPU that has the greatest possible single thread performance you can afford.

Thanks for your reply.
I do know VAM has a main threading control and assign tasks, but I don't know the soft body physics calculations are part of the main threading or just the task be assigned.
As you can see VAM always takes about 55~70% of CPU usage with a i7-8700.

With a i7-1260P, VAM still takes about 55~70% of CPU usage.

An i7-8700 is a 6 physical cores CPU.
An i7-1260P is a 12 physical cores CPU(4 P-cores+8 E-cores) while amount of computation of the single core and total cores is about 1.5X compared with the i7-8700.

Based on the above factors, I consider though VAM run an single thread, it support multi-physical-cores with acceptable efficiencies.
But VAM is BAD with logical cores technologies like Intel Hyperthreading (Not sure of AMD SMT, because their structure layers are different.)

So I guess certainly the single core performance is very important for VAM, but perhaps not such kind of important compared to the total-cores-performance?
I knew the benchmark thread, but most were about GPUs performance not CPUs.
https://hub.virtamate.com/threads/benchmark-result-discussion.13131/page-25

Actually I run a small office for previous task, so yes I do have a XEON server.
I know the main use of VAM is for EROS.
But in fact the human body is very complex especially female body. As I remember it is called the "inhomogeneous structure."
Anyway, I think VAM do have a great job on soft body physical calculations, maybe I can discovery some different use way.

So, even if 1 or multiple "Server CPUS", I'd like to give it a shot as an investment.

As the list of CPUs above, are there any CPUs matched my computing requirements?

Thanks for all you guys' replies.

 

[DJ]

Typically, if the CPU is the bottleneck, you'd see one or two cores very near 100%, and the others with much less activity. To be clear, Vam is multithreaded, but that main thread is the issue.
If none of the cores are maxing out, the bottleneck is your GPU. The GPU deals with hair and clothing simulation, so try removing hair & clothing until one of your CPU cores is at 100%. That will show you what kind of framerate your CPU is capable of delivering.

 

[ruinousdeity]

Typically, if the CPU is the bottleneck, you'd see one or two cores very near 100%, and the others with much less activity.

It did happen to my i7-1260P+VEGA56 GPU, didn't it?

I used to work on total motherboard planning and development, all for intel Core & XEON CPUs.
So i can explain you misunderstood something and "why software often has difficulty reaching 100% utilization in environments using INTEL CPUs."

1st, a software supports multi-core CPUs or multi-CPU environments, it does not mean that it is a multi-threaded software; these are actually two very different things.
They look quite different, right ?

2nd, Intel's Hyper-Threading technology allows for each physical core to be divided into two virtual cores, which can process multiple threads simultaneously. However, this can also lead to contention for shared resources such as cache and memory bandwidth. This can result in reduced performance for software that is not optimized for Hyper-Threading, or for software that requires a large amount of resources from a single thread.

Additionally, Intel's multi-core architecture and Hyper-Threading rely heavily on software-level control to achieve optimal performance, rather than being built into the hardware itself.

But as far as I know, on the other hand, AMD SMT allows each physical core to process multiple threads at the same time. AMD's SMT is similar to Intel's Hyper-Threading, but it is built into the hardware itself and is available on a wider range of processors.

Basically, this knowledge requires experience in motherboard or CPU hardware design and development, and is not commonly known by most people or software designers and engineers. The fact that in an Intel CPU environment, multi-core and multi-threaded software must specifically support Intel's hyper-threading instruction set is not widely known.

As a supplement, the reason for the 100% usage of the RX6950XT in the previous screenshot is because at that time, I had not loaded any scene, I was only in the scene list. Because the CPU was not performing any calculations and the GPU was not performing any MSAA, and there was no limit on the FPS, the GPU usage was usually very high.


So, the CPU is actually the only factor causing the performance bottleneck here.

Thanks for your reply.