Jump to content


  • Content count

  • Donations

    0.00 CAD 
  • Joined

  • Last visited

  • Days Won


Everything posted by malexander

  1. Linux Mint 18 - Crash on wake

    On Linux, the Nvidia driver has no way of notifying the application that a suspend/resume cycle has nuked various GL objects. We attempt to do a test for this in Houdini when a redraw hasn't occurred in a while, and while the test did seem to work for some earlier driver versions and Linux distributions, it's not a sure thing. Nvidia has recognized this issue and added a GL_NV_robustness_video_memory_purge extension, but I've yet to see that pop up in a driver. Once it does I will see if it can be integrated into Houdini.
  2. Feedback on new Custom PC build

    Viewport rendering.
  3. Feedback on new Custom PC build

    Honestly, I'd trim back to a GTX 1080 and use the extra $$ to buy 64GB. Dual 1080s would work just as well as dual Titans, if you need that sort of thing. Houdini can only use 1 GPU for rendering and 1 for Compute, so even 1080 (rendering) + Titan (compute) would work.
  4. MocapBiped1 Capture Weights (Huh?)

    The bone weights were likely painted on, instead of using the bone capture regions.
  5. Xeon Max Turbo Frequency and Houdini

    Besides the turbo and base frequencies, there are several speed steps in between. When running multithreaded you'll sometimes get a higher clock than the base, either for short bursts or if some cores are idle. However you can generally assume that a heavy threaded task using 100% CPU will run at the base frequency. A well-threaded task will benefit more from the extra cores than the slight clock boost from a lower-core part, though.
  6. Xeon Max Turbo Frequency and Houdini

    A good portion of Houdini (and most DCC apps) is still single threaded, especially UI, GL drawing, and "managerial" type work. The Turbo frequency should kick in for these cases.
  7. Pascal's 16 bit compute

    Maxwell and Kepler didn't have fp16 compute, the new pascal-based Telsa card (GP100) was the first to introduce it. The reason cited for fp16 compute was for "deep learning" applications. It basically allowed for twice the throughput of FP32 operations. The pascal-based Quadros and GEForces have the same 1/32 FP64 rate as their Maxwell predecessors, so at least in that aspect they're the same. As for fp16 compute, you wouldn't be able to run it on Maxwell or Kepler cards - or at least it'd be FP32-emulated. It would have been nice to have since HDR color is a good candidate for fp16 compute, but fp32 will do I guess (just not at double rate that full fp16 support would have had). The Radeon Pros are based off the same Polaris 10 architecture as the Radeon 480, putting them right about at the GEForce 970-980 mark in terms of CL performance.
  8. Pascal's 16 bit compute

    Just a quick update on the non-Tesla Pascal cards: The GEForce 1080, 1070, 1060 and the new pascal-based Titan X (no "GTX" to differentiate from the Maxwell-based GTX TitanX) and the new Quadro P series. FP16 compute on these cards is horribly crippled, even worse than FP64. For every shader module of 128 FP32 units, there are 4 FP64 units (1/32) and a single vec2 FP16 unit (1/64 rate) on the GP102 and GP104 that powers the GEForces, Quadros, and Titan X. Basically they are only there for debugging programs to run on the Pascal-based Teslas.
  9. CHOP Spike Cleanup Recipe?

    Use the Filter CHOP with the filter set to "Despike".
  10. Is a single point "geometry"?

    Curves also need some width given to them as well, otherwise they'd be in the same boat as points - invisible
  11. Is a single point "geometry"?

    I just think of a point as a location in space which you can position other geometry at (an instance, a dot, sprite, etc). If you don't place any other geometry there, it can't be rendered. So on its own, it's nothing but a placeholder.
  12. Houdini OpenGL 3.3 error.

    What driver version is installed? Did you try any others?
  13. Houdini OpenGL 3.3 error.

    A long shot, but do you have the environment variable HOUDINI_OGL_MAX_GL_VERSION or HOUDINI_OGL_CORE_PROFILE set somewhere?
  14. OpenGL VR shaders

    That sort of display is more appropriate for the OpenGL ROP and Flipbooks. In the viewport, it makes selecting things rather challenging
  15. OpenGL VR shaders

    The main barrier is performance, the lens shaders themselves are pretty simple. Maintaining 90fps at all times when we don't have any control over what is displayed, is the main problem. I suspect we'd have to severely restrict how things are displayed when in any sort of "VR" mode (shaded only, normal quality lighting, no visualizers, etc).

    VDB is actually a third party lib from Dreamworks. I guess they love their acronyms too Volume: Dynamic Blocked is what I believe it stands for.
  17. smooth in viewport

    You probably pressed `+` to turn on subdivision. You can do when fencer suggests above or press `-` to turn it off.
  18. Houdini 16 Wishlist

    Either Image > Cache All within MPlay, or mplay -C <files> from the command line.
  19. synchronizing two mplay (flipbook) sessions

    If you're rendering or flipbooking, click the "New Sequence" button in the upper right corner (Film-like icon). If you're loading files, use File > Merge for the second sequence, or mplay sequenceA\$F.pic sequenceB\$F.pic from the command line. Then View > Layout > 2 Columns.
  20. PC build for 3000€

    That's a pretty sweet build!
  21. Houdini 16 Wishlist

    Just guessing, but probably because integers are very straightforward to convert to strings, whereas FP numbers have quite a few more options - #significant digits, scientific notation, decimal digits, etc. Once you've added those options to ftoa(), sprintf() doesn't look like a bad alternative.
  22. Houdini 16 Wishlist

    What basically happens is that the viewport object that encapsulates the node's viewport data becomes de-synchronized from the node, and doesn't receive notifications anymore. Then it just sort of sticks around forever.
  23. High pass filter in COPs

    The Sharpen COP also does this with "Show Edges Only" enabled, though it centers around zero instead of 0.5. It also has a noise filter.
  24. Houdini 16 Wishlist

    Yeah, it's not a GPU cache problem. It's the scene manager which is getting confused, probably by a network change, because the rules for what should be displayed and what shouldn't are ridiculously complicated (once you factor in all the different visibility flags, subnets, modes, nettypes, ways to change visibility, etc). That's also the reason that its very difficult to get reproducible steps.