acc3d

More observations about OpenCL performance in Houdini 14.0.291: I tried running the same pyro smoke simulation (1500 frames, no caching, straight to disk) on the following hardware. Results are in minutes:seconds format: Software mode: 29:20 - Intel i7-5960X CPU @ 4.3GHz OpenCL Mode: 7:20 - GeForce GTX 660 2GB (OpenCL 1.2 CUDA) 7:19 - Radeon R2 290 4GB (OpenCL 2.0 AMD-APP, 1642.5) 8:30 - Intel i7-5960X CPU @ 4.3GHz (OpenCL 1.2, Build 57) I was really surprised that the Radeon wasn't any faster than the GeForce. The other surprise was how fast the CPU is with the Intel OpenCL driver. For now, I'm going to use the CPU/OpenCL and wait for the next gen graphics cards to be released. Edit: There's also a big discrepancy in the look of results from the OpenCL vs software mode simulations. The OpenCL-generated ones don't seem to dissipate density properly. Can anyone else confirm this?

There's also an MSI 8GB Radeon R9 290X for around $380 USD. I've been looking for a good GPU for OpenCL acceleration of pyro simulations. Rumor has it AMD is about to announce the R3 series of GPUs this quarter though...might be wise to wait a bit.

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.

True, but it's MUCH slower to simulate non-packed. In any case, kgmcnamara submitted this bug to SESI a while ago. Is there any way for us to inquire about it?

Great, thanks for confirming my assumption!

Does anybody know if a DOP Network automatically generate sub-integer frames if you set sub steps higher than 1? For example, if I pass in some geometry from SOPs that's saved on disk in integer frames, does DOPs automatically interpolate subframes, of would I need to use a Timeblend to generate subframes.

I just tested, and the bug does NOT appear to be fixed. I will try to follow up with SESI and see what's going on. It may actually be a bug in Bullet, not Houdini...

About 2 weeks ago, I got an email from SESI that said they're looking into the issue. I've been watching the Houdini Journal and so far there is no fix:

Pawel, I just submitted a bug report to SESI with your example file. We will see what they have to say.

I don't think this is a display/OpenGL bug, because the re-topologizing appears to affect the simulation. When the topology changes, the objects jitter because they're trying to resovle new collisions with the ground plane. Also, the re-topologizing is visible in wireframe mode too. I posted a file last week that demonstrates the same problem: http://forums.odforce.net/topic/16254-testing-bullet-why-is-everything-poppingblowing-up/page-2 -AC

Hi cubiccube, I've been trying to troubleshoot some annoying jitter in my own Bullet sim and came across your example file above. The problem with my sim seems to be thin pieces combined with what I believe is a bug in the 'RBD Packed Object' OTL. With the 'show guide geometry' turned on and 'display geometry' off, you can see that it's regenerating every frame (use smooth or flat wire shaded). Simulate long enough to the pieces settle, then you'll see the guide geometry change topology. Is this a bug in SideFX's OTL? Can somebody look at the attached file and tell me if I've made a mistake somewhere? Thanks, AC bullet_packed_rbd_retopologize_01.hipnc

Hi guys, this is great to hear about the Surface Pro! I've been trying to decide between a MacBook Air and a Surface Pro 2 (sick of carrying my 11 lb Dell M6600 around). Can anybody comment on how well H13 runs on an Air? I'm leaning towards the Surface because you can detach the keyboard and use it as a tablet, but they sure are pricey :/