Iron cores and hatred, I want an opinion.

   2102   2   0
User Avatar
Member
1 posts
Joined: 8月 2021
Offline
Hello. I am interested in your opinion about the Houdini system requirements.
I have heard that Houdini makes good use of multithreading in processors, and open cl can also speed up many simulations using a video card, but I began to doubt this very much.
The fact is that for 8 years I have been dreaming of building a computer that would cope well with simulations. But updating the system every time, I do not notice any increase in this regard at all.
I understand that there is no computer for calculating simulations in real time and you can't do without caching, but I would like to at least double the speed of calculations.
Let's say I had a computer:
Intel i7 8700k (4.7 hz, 6 cores)
GTX 1060ti (6gb)
16gb ram (3200mg)
hdd
I created several dozen scenes using different simulations for tests and started updating the components.
To begin with, I bought a video card, since it is needed for rendering and many other tasks – RTX 3080. I tested it, the tests showed that there is no gain from it at all in the simulations. Although open cl. The result is no different from the gtx 1060ti.
I started looking for a bottleneck.
Next, I updated the hhd, on an ssd m2, (its speed is 10 times higher). I expected that when reading from the disk, the simulations would load faster. In fact, nothing has changed again.
Added 64GB of RAM. I expected that at least this would help in heavy scenes. In very severe cases, this added an increase of 10-15% at best.
Then I changed the processor to Ryzen 9 5950x(4.6 hz, 12cores)(currently the top)
Here I already wanted to see all the beauty of 12 cores. But after running through the tests, I noticed only an increase in some types of simulations by 30%, and in some, to my surprise (like calculating pure code on VEX), on the contrary, FPS fell by 53%, that is, in the dry balance, I received almost no increase.
As a result, I updated the computer for $ 4083. And in the best case, I received an increase of 30 percent, and in some calculations, it began to work even worse
Of course, Redshift now renders perfectly, but damn, what kind of computer is needed for these simulations? Is it really possible that only supercomputers can cope with this and the creation of effects is still the fate of only large companies? Tell me, what is wrong with this assembly? And which specific part is overwhelmingly the most important for simulations? Where is the bottleneck?
Ryzen 9 5950x (4.6Hz 16 core)
GTX 3080
64gb ram(3200mg)
SSD 1ТБ(Samsung 970 EVO)
Did I correctly draw the conclusions that everything is counted on one core, so the number of cores in general almost does not play a role? And how to deal with this in an era where everyone is chasing the number of cores, and the ghz of the last 10 years has hardly grown at all?
User Avatar
Member
238 posts
Joined: 3月 2013
Offline
You've got some fundamental issues in how you think a Computer works, and how simulation and rendering works.

Let's take a pyro smoke sim. The more cores you have, the faster it will simulate, no question, this is easy to
prove. Now if you have a powerful GPU, that is only going to benefit your pyro sim IF all the nodes in your setup
for the sim are openCL capable. If they are, great. You can flick over to openCL and your sim will potentially be
even quicker than your CPU one. But you will hit GPU mem limits very quickly, so GPU sims are not a great idea for huge
situations. The speed of your hard disk will come into play purely for writing the cache data to disk, which happens in the
background while simming. Unless your sim is generating huge caches you won't see much improvement at all, as 50mb being written
to disk on an old spinning drive Vs SSD isn't going to be that different in terms of speed.
Where your hard disk will come into play more, is if you are pulling in heavy collision geometry, or some heavy velocity fields
that you have stored on disk, and want them to interact with your pyro sim. In that situation the sim will be waiting for the data
to be read in, so if it's huge, an SSD will totally make a difference.
Ram wise, it's been many years since the speed of ram was even a remote bottleneck to things. All that really matters with ram is that you have as much as you can get. The amount will do nothing speedwise, it will only enable you to sim larger more memory hungry simulations.

There are plenty of examples online of the very real differences in sim times between upgraded systems, so your assertion is wrong.

If I had to pick, I would say core count, ram, and depending on how big the data you generate or pull in is, I'd add the hard disk speed as well.

GPU will only benefit you in situations that take advantage of it, and if your sim/render fits into VRAM.


Cheers

L
I'm not lying, I'm writing fiction with my mouth.
User Avatar
Member
648 posts
Joined: 7月 2005
Offline
PVK
I understand that there is no computer for calculating simulations in real time
with the right skills its possible...

1 core fluid sim by John Lin last year:
https://www.youtube.com/watch?v=4Y58Pg9tpSo [www.youtube.com]

and this one by Alex Mcleod... 10 years ago:
https://www.flickr.com/photos/25502060@N03/5407756540/in/album-72157610018291727/

Not useful for your work though, maybe reconsider doing sims if its too frustrating.
  • Quick Links