Jump to content

Pyro Drag Race | Houdini Graphics Card Comparison


ninjismo

Recommended Posts

Hello!

 

I haven't seen a resource that benchmarks the calculation speed of graphics cards in Houdini so I thought I'd post this. I hope that people can participate and post their scores using their gaming/workstation graphics cards so other people can get a rough idea of the performance of each graphics card in Houdini before they decide to buy/upgrade anything. This should only take a few minutes -

 

 

1. Start Houdini 

2. Use the following units under your Hip File Options - 

Unit Length  - 1 

Unit Mass - 1 

3. Frame Range 1-240 @ 24fps

4. Add a sphere at origin

5. Goto the Pyro FX shelf tool and select Explosion. Select the sphere. 

6. Disable Cache Simulation in your DOP and use the following settings - 

 

Pyro Container Settings

Division size - 0.075

Size - 7.5, 10, 7.5

Center - 0, 4, 0 

 

Resize Container Settings

Padding 0.3

Subtract Threshold 0.2 

 

9. Select the Pyrosolver. In the Advanced tab, turn on OpenCL

10. Save the file and restart your computer

11. Use a stopwatch or http://www.online-stopwatch.com/ to keep track of the time 

12. Open the file, start the sim and the stopwatch timer

13. Record the time after 240 frames.

14. Run the same sim again but with OpenCL disabled

15. Post the exact times with your PC Specs, Houdini version and OS in the following format (using my results as an example) -

 

 

Houdini Version - 14.0.201.13 

OS - Windows 7 Ultimate x64

 

GFX - Zotac 770GTX Amp! Edition (2GB) / Driver - 347.09

Time - 1:37.222

 

CPU - Intel i7 3770K @ stock 

Time - 2:09.503 

 

Other - ASRock z77 Pro3 Motherboard, 32 GB Corsair Vengeance RAM

 

 

Thank you! 

 

PS - I've attached a hip file with the above settings using Houdini 14 that you can run right away. Thanks for the suggestion, Skybar. 

OpenCL_TestFile.hipnc

Edited by ninjismo
Link to comment
Share on other sites

I would like to know how often graficcards are used in production at all, I switched to cpu open cl because it uses all my memory and is not limited to grafik ram, it is not as fast as a grafik card but has no limits.

Well the answer is probably never in a production environment for the final sim. To get a fair understanding of how the sim will look before I start churning out high-res versions of it, I always use the graphics card OpenCL version first even with its limitations of RAM because of the number of iterations I can see in a given amount of time, considering how small my production deadlines are. I have done this on 3 productions so far. It is very helpful for me in personal projects as well. And now with graphics cards coming out with 8GB and 16GB of RAM, it will be interesting to see how they perform on the OpenCL test. The main thing here is to check the speed, hence the name of this thread. 

Link to comment
Share on other sites

Well the answer is probably never in a production environment for the final sim. To get a fair understanding of how the sim will look before I start churning out high-res versions of it, I always use the graphics card OpenCL version first even with its limitations of RAM because of the number of iterations I can see in a given amount of time, considering how small my production deadlines are. I have done this on 3 productions so far. It is very helpful for me in personal projects as well. And now with graphics cards coming out with 8GB and 16GB of RAM, it will be interesting to see how they perform on the OpenCL test. The main thing here is to check the speed, hence the name of this thread. 

thanks

Link to comment
Share on other sites

  • 3 weeks later...

Wouldn't it just be easier to upload an example file we can run the benchmark with. And then use the Perfomance Monitor in houdini to get the times. This is just tedious and is wide open for errors. If you want to get some sort of reliable results that is, maybe that isn't the purpose.

  • Like 1
Link to comment
Share on other sites

My 2 cents: To get a good view of OpenCL performance you really need to run small, medium and large tests.

 

Be very aware of the platforms; if the Gtx 980 runs really well on Linux, yet to test, it currently doesn't run well on OsX, where as the AMD 7950 runs really well on OsX 10+ but quite bad and very bad on OsX 10.9.5  / <10.9.3/4 . On Linux Mint it's great.

 

AMD also runs very well on smaller tests but Nvidia picks up greatly on larger tests. etc. 

Link to comment
Share on other sites

Well the answer is probably never in a production environment for the final sim. To get a fair understanding of how the sim will look before I start churning out high-res versions of it, I always use the graphics card OpenCL version first even with its limitations of RAM because of the number of iterations I can see in a given amount of time, considering how small my production deadlines are. I have done this on 3 productions so far. It is very helpful for me in personal projects as well. And now with graphics cards coming out with 8GB and 16GB of RAM, it will be interesting to see how they perform on the OpenCL test. The main thing here is to check the speed, hence the name of this thread. 

 

I would have to disagree with this one.

At ilm they have the best tool ive ever used for fluid simulation, fire smoke and explosions. Its called plume and the entire sim as well as rendering is done on a gpu.

Its so fast that we barely ever used to save cache files, only when particle advection was necessary. All the fluid sims you see on transformers and most of ilm movies in the past couple of years were done on a gpu with no more than 4gb of ram. Its a CUDA gpu solver if i remember correctly, you can actually find some links on it.

Its such a stripped down solver which makes it so fast and so easy to iterate.

 

Here is a tech video on it.

 

Sidefx is doing a great job moving dop nodes to be executed on the gpu, it does make a difference.

Link to comment
Share on other sites

huh? - this is Houdini not ILM's custom tool.

 

Sorry didnt quite make myself clear, thought this is about general GPU usage in simulation.

 

I was only saying that GPU simulation in production is being used and widely at certain places, and that houdinis opencl is on the right track to utilize the gpu in the production. Even tho the 4gb of gpu memory is a bottleneck sometimes, it does improve the performance quite a lot, and now in h14 with the grain solver and intel OpenCL it does look very promising.

 

The only problem ive experienced so far in production was the memory limit of the GPU.

Edited by nosferatu_037
Link to comment
Share on other sites

Also, if you want to test pure OpenCL paths, just turning on OpenCL on a shelf-created explosion doesn't quite do it because caching is on by default (which copies GPU memory to CPU memory), and there are (or were) some nodes/noises that weren't OCL ready as well.

 

There is a "pure" OpenCL example (which jlait setup for the H12 teaser video) in the docs: http://www.sidefx.com/docs/houdini14.0/examples/nodes/dop/smokeobject/OpenCL

Link to comment
Share on other sites

  • 1 month later...
Houdini Version - 14.0.291 

OS - Windows 7 Pro x64

 

MSI GTX 970 Gaming G4 (4GB) / Driver - 347.52

Time - 58.5 s

 

CPU - Intel i7 4790K @ 4.5 Ghz

Time - 1:38.5 m

 

Other - MSI Z97 Gaming Motherboard, 32Gb Corsair Vengeance RAM

Link to comment
Share on other sites

  • 1 month later...
Houdini Version - 14.0.291

OS - Windows 8.1 Pro x64

 

Machine specs:

CPU - Intel i7 5960X

GPU - MSI GTX 660 2GB / Driver 350.12

Motherboard - Gigabyte x99 G1-Gaming

RAM - 32Gb G.Skill DDR4

 

CPU Only: 1m, 18s

OpenCL: 1m, 1s

 

Seems like the GeForce 970/980 don't improve much on my old 660, which is strange. Guess I won't be dropping $500 USD for a GTX 980...

 

Also I didn't restart my machine or close any running applications, so my scores could be better.

Link to comment
Share on other sites

  • 3 weeks later...

I'd like to see a production example for a benchmark because all of the quick tests I've done show the CPU and GPU have pretty much the same performance in this very specific use case.

 

With OpenCL: 56.06 s

Without OpenCL: 59.93 s

 

The GPU is a GeForce Titan X and the CPU is dual Xeon E5-2630 V3.

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...