Hi,

I want to build a new PC. Can anybody help which would be good?

Intel 13th Gen Raptor Lake Core i9 13900K 3.00GHz-5.80GHz
or AMD Ryzen 9 7950X 4.5GHz-5.7GHz

Intel has 24 cores but it is 16 physical core and 8 performance core. So, here I would get 16core*3.00GHz = 48 speed for Houdini sim?

Though AMD Ryzen 9 has 16 core, its base speed is 4.5. Should I a get about 16core*4.5 = 72 Speed for Houdini Sim?

I am not good in express, Just want to know what would be good choice.

Also please suggest relevant motherboard, Ram for the budget friendly build.

Thanks in Advanced.

I have the cpu 13900k, In pyro tab Sparse pyro fireball shelf tool, without any changes 50 frame simulated in 13.94 sec
128GB DDR4 3600 RAM

you can ask any 7950x guys to do the same lets see how much time it taking...

9900KF 39sec, just for reference

23 sec with 5950x ( 2x32 3600mhz ram)

would be interesting to know results with 7950x and now with new 9950x

12.02 Sec with 7950x 64GB DDR5 5200 MHz - Also I should note that I have undervolted my 7950x a bit (a negative 15 Magnitude on my BIOS curve optimizer)

AMD yes!

27sec on a Threadripper 3970X. But the sim cannot utilize all cores at 100%, system load is between 50-70%. Fewer but faster cores are an increasingly valid choice for small to medium sized sims I feel.

Threadripper 3970X and 5950x don't do better?!
this takes around 24 seconds on my entry level mac mini M2 with 16GB Ram
I am about to build a machine with the 7950X so I would also be very interested in sim performance numbers of that chip

24s on M1 Max 32MB as per performance monitor on the fireball_simulation object.

It doesn't appear to max out any of the cores but it does use all performance cores to around 80%.

GPU not used at all. OpenCL not supported for sparse solving.

Really curious about M1 Ultra or M2 Ultra.

7950x hands down. It is the best bang for the buck chip out there. You are also future proofing at the
newer chips will use the same socket.

L

lewis_T
7950x hands down. It is the best bang for the buck chip out there. You are also future proofing at the
newer chips will use the same socket.

L

why go with last gen, price?

If you need to build a machine right now without delay a 7950X might be a better bet as its proven beyond doubt. Certainly better value. If you can wait a little bit the 9950X might get sorted out with regards to its lackluster performance gains -or even regression in some areas- over the Zen 4 architecture.

5950X, no PBO, with 128GB Ram and 2x4090s

Keen to see this on the new 9950X

Does Houdini take advantage of AVX-512?
If I remember correctly, RenderMan uses AVX 512 acceleration.

Antti1999
Does Houdini take advantage of AVX-512?
If I remember correctly, RenderMan uses AVX 512 acceleration.

Houdini itself doesn't, however some of the libraries that we use can take advantage of it.

johnmather

Antti1999
Does Houdini take advantage of AVX-512?
If I remember correctly, RenderMan uses AVX 512 acceleration.

Houdini itself doesn't, however some of the libraries that we use can take advantage of it.

Just out of curiosity, how much of a performance uplift would that translate to and where? Sim times? Remeshing? VEX?

johnmather
er correctly, RenderMan uses AVX 512 acceleration.
Houdini itself doesn't, however some of the libraries that we use can take advantage of it.

Does Intel Embree use it if present, as part of Karma XPU? (and will it block use of it when an AMD CPU is detected >)

jsmack
Does Intel Embree use it if present, as part of Karma XPU? (and will it block use of it when an AMD CPU is detected >)

Intel Embree does use it, and no its not blocked for AMD. AMD (and apple) both implement/emulate the AVX instruction set, so works fine

But is there a noticeable real world difference with AVX-512?
Karma CPU (as most CPU render engines, except Corona) is extremely slow compared to XPU, so it seems more practical to invest in 4x 4090s rather than 2x 64-core Threadrippers.
Of course, if you're working in a large studio with a massive render farm, CPU rendering still makes sense.

In this day and age, 24GB of VRAM feels like a joke. It seems like Nvidia is laughing at us. I really hope they release 48GB versions of the 5090 at a reasonable price.

If I open Photoshop and Substance Painter at the same time, half of the VRAM is already gone. We’re paying so much for these GPUs, and they’re still not delivering...
We just pretend they’re awesome, but honestly, they’re only great for gaming. Someone on Nvidia’s marketing team needs to address this, 24GB is ridiculously low, and no, Quadros are way too overpriced just for an extra 24GB.

Speaking about CPU differences..try spawning a regular box, then a vdb from polygons, change it to density and set it to 0.0003 so you get 3333x voxels.

I have two machines and one is a Intel Xeon W-2245 it handles it no problem in 14 sec or so never breaking 64gb RAM used while physically having 128GB. The other machine is a Ryzen 5950x it has 64GB RAM, which when doing the same thing runs it right into the 64GB RAM limit and fails to ever convert the box to a 3333x density volume and I have to kill the process.

While doing simple things you wont realy notice any difference, but I do wonder if when pushing the system if there is something benficial for Xeon W, or perhaps even regular Intels. Anyone care to do the same test and say how it goes?