VMD-L Mailing List

From: John Stone (johns_at_ks.uiuc.edu)
Date: Fri May 29 2020 - 16:46:33 CDT

Hi,
  I've posted the first round of VMD 1.9.4 alpha 42 builds, which has
OptiX 6.5 and the code path that no longer tries to workaround the old
broken drivers. The Linux build was done on CentOS 6.x and is a bit of
an experiment due the bleeding edge libraries it is built on.
                                                                               
If anyone encounters any trouble, I'll likely make a second Linux build
using CentOS 7 or CentOS 8, but I'm planning to hold off on that unless
it is needed, as my plan for CentOS 8 and Ubuntu 18/20 distro users is
to do builds on CUDA 11 whenever it posts.
                                                                               
Best,
  John

On Wed, May 27, 2020 at 04:26:16PM -0500, John Stone wrote:
> Hi,
> Quick answer since I'm amidst zoom sessions:
> Yes, the older alpha builds had to not free some memory, or they would
> crash due to a driver bug, but this led to a huge memory leak, which also
> causes a crash, but not until you render a bunch of frames. The new
> test build I will be posting no longer does that, since the drivers
> were finally fixed, so you'll get the expected behavior.
>
> I'm finishing integrating new code and patches, and hope to have
> a new test build before the end of the week.
>
> Best,
> John
>
>
> On Wed, May 27, 2020 at 12:04:45PM -0700, Lorenzo Casalino wrote:
> > Thank you John,
> > I updated the driver to 440.82 and that fixed the problem!
> > Now I was able to use 1.9.4a38 RTX, with significant performance
> > improvement.
> > However, while rendering a movie, it always fails after a few frames and
> > similarly to a31.
> > This doesn't happen with other vmd (nonRTX) like 1.9.3, 1.9.4a27 or
> > 1.9.4a35.
> > I am rendering a scene with 240,000 atoms, with a transparent surface, new
> > cartoons and a few graphics elements. I am using a user-defined procedure
> > and running using a tcl script (text mode).
> > It seems that when it reaches a view point and starts proceeding to the
> > next one it crashes with the following error (vmd 1.9.4a38):
> > ERROR) OptiXRenderer) ERROR: Unknown error (Details: Function "RTresult
> > _rtContextLaunch2D(RTcontext, unsigned int, RTsize, RTsize)" caught
> > exception: Assertion failed: "!m_enteredFromAPI : Memory manager already
> > entered from API", file: <internal>, line: 1103) (OptiXRenderer.C:2621
> > OptiXRenderer) Error during AS generation. Rendering aborted.
> > I tried with a31 and crashes at the same point with this error:
> > ERROR) OptiXRenderer) ERROR: Memory allocation failed (Details: Function
> > "RTresult _rtContextLaunch2D(RTcontext, unsigned int, RTsize, RTsize)"
> > caught exception: Out of memory) (OptiXRenderer.C:2597
> > OptiXRenderer) Error during AS generation. Rendering aborted.
> > I have a 24 GB RTX GPU so I believe it can handle the scene pretty easily
> > memory-wise.
> > Thank you,
> > Best
> > Lorenzo
> > - - - - - - - - - - - - -
> > Lorenzo Casalino, PhD
> > Post-doctoral research scholar - Amaro Lab
> > University of Califonia, San Diego
> > 3210 Urey Hall
> > 9500 Gilman Dr, La Jolla, CA 92093-0340
> > [1]https://amarolab.ucsd.edu/
> >
> > On May 25, 2020, at 9:16 PM, John Stone <[2]johns_at_ks.uiuc.edu> wrote:
> > Hi,
> > The most likely cause of your problem is the out-of-date NVIDIA driver.
> > Unlike prior versions of VMD, VMD 1.9.4 is compiled against a new
> > generation
> > of OptiX wherein a large fraction of the OptiX code lives as part of the
> > driver rather than in the liboptix.so.* shared libraries. This change
> > has the direct consequence that if you run an older driver, you're
> > also running an older version of the OptiX code. The latest VMD 1.9.4
> > test builds I'm making are compiled against OptiX 6.5, and the minimum
> > driver version required is 436.02, which is quite a bit newer than
> > your current driver.
> >
> > It gets more interesting however. Since a large part of the core
> > OptiX implementation now lives in the driver, another issue is that
> > bugs in OptiX now get fixed as part of driver releases. This means
> > that it is much more important not only to have a new-enough driver,
> > but also one that is "stable". In my mind, people using VMD 1.9.4 with
> > OptiX should be using NVIDIA driver version 440.44 or later, as some
> > issues VMD previously encountered memory management were cured in the
> > later drivers.
> >
> > So, I'd go for version 440.44, or later. We have a bunch of machines
> > running 440.44 with RTX cards, and they have been pretty stable for us.
> >
> > There will soon be a new set of drivers coming out with the
> > release CUDA 11, I'm hopefull that the OptiX components in these new
> > drivers will be good and solid, in which case that will be what I advise
> > most people to use going forward (445.xx driver series).
> >
> > I expect to post new VMD 1.9.4 test binaries shortly, you'll also want
> > to use these rather than the older a38 build.
> >
> > Best regards,
> > John Stone
> >
> > On Mon, May 25, 2020 at 01:28:53PM -0700, Lorenzo Casalino wrote:
> >
> > Hi all,
> >
> > I am having some issues in getting Optix renderer compiled on a target
> > GPU when using VMD OPTIX RTX versions, like 1.9.4a31 or the latest
> > 1.9.4a38 (ERROR: Invalid value (OptiXRenderer.C:1214).
> >
> > First of all, I have 2 GPUs installed on my machine. This is the log
> > when I run nvidia-smi:
> >
> > | NVIDIA-SMI 430.26 Driver Version: 430.26 CUDA Version:
> > 10.2 |
> > | 0 Quadro P6000 Off | 00000000:3B:00.0 On |
> > Off |
> > | 26% 57C P5 30W / 250W | 1846MiB / 24446MiB | 21%
> > Default |
> > +-------------------------------+----------------------+----------------------+
> > | 1 TITAN RTX Off | 00000000:AF:00.0 Off |
> > N/A |
> > | 41% 42C P8 15W / 280W | 345MiB / 24220MiB | 0%
> > Default |
> > +-------------------------------+----------------------+----------------------+
> >
> > The GPUs are detected, but for some reason vmd1.9.4a38 fails in
> > compiling Optix renderer on the TITAN RTX.
> > This is the log when I try to run vmd 1.9.4a38:
> >
> > Info) VMD for LINUXAMD64, version 1.9.4a38 (October 20, 2019)
> > Info) [3]https://urldefense.com/v3/__http://www.ks.uiuc.edu/Research/vmd/__;!!Mih3wA!Wc_2cph4f8JXiqRM6sFgFd7wqedSGncDH3YaNIdxJk96V7uocI1wGR6SeGBFTfg6$
> > Info) Email questions and bug reports to [4]vmd_at_ks.uiuc.edu
> > Info) Please include this reference in published work using VMD:
> > Info) Humphrey, W., Dalke, A. and Schulten, K., `VMD - Visual
> > Info) Molecular Dynamics', J. Molec. Graphics 1996, 14.1, 33-38.
> > Info) -------------------------------------------------------------
> > Info) Multithreading available, 72 CPUs detected.
> > Info) CPU features: SSE2 AVX AVX2 FMA AVX512F AVX512CD
> > Info) Free system memory: 955GB (94%)
> > Info) Creating CUDA device pool and initializing hardware...
> > Info) Detected 2 available CUDA accelerators:
> > Info) [0] TITAN RTX 72 SM_7.5 1.8 GHz, 24GB RAM SP32 AE3 ZC
> > Info) [1] Quadro P6000 30 SM_6.1 1.6 GHz, 24GB RAM SP32 KT AE2
> > ZC
> > Warning) Detected X11 'Composite' extension: if incorrect display
> > occurs
> > Warning) try disabling this X server option. Most OpenGL drivers
> > Warning) disable stereoscopic display when 'Composite' is enabled.
> > Info) OpenGL renderer: Quadro P6000/PCIe/SSE2
> > Info) Features: STENCIL MSAA(4) MDE CVA MTX NPOT PP PS GLSL(OVFGS)
> > Info) Full GLSL rendering mode is available.
> > Info) Textures: 2-D (32768x32768), 3-D (16384x16384x16384),
> > Multitexture (4)
> > Info) Using OptiX device mask '0x1'
> > Info) Detected 1 available TachyonL/OptiX ray tracing accelerator
> > Info) Compiling 1 OptiX shaders on 1 target GPU...
> > ERROR) OptiXRenderer) ERROR: Invalid value (OptiXRenderer.C:1214
> > ERROR) OptiXRenderer) Failed to create OptiX rendering context
> > ERROR) OptiXRenderer) ERROR: Invalid value (OptiXRenderer.C:1214
> > ERROR) OptiXRenderer) Failed to create OptiX rendering context
> >
> > I have set the environment variable to use only the first GPU
> > recognized by VMD (e.g., [0] TITAN RTX): export
> > VMDOPTIXDEVICEMASK=0x1
> > It also fails if I try to use the QuadroP6000 (export
> > VMDOPTIXDEVICEMASK=0x2).
> >
> > Instead, when I run vmd 1.9.3 or 1.9.4a27 everything works:
> >
> > Info) VMD for LINUXAMD64, version 1.9.4a27 (February 8, 2019)
> > Info) [5]https://urldefense.com/v3/__http://www.ks.uiuc.edu/Research/vmd/__;!!Mih3wA!Wc_2cph4f8JXiqRM6sFgFd7wqedSGncDH3YaNIdxJk96V7uocI1wGR6SeGBFTfg6$
> > Info) Email questions and bug reports to [6]vmd_at_ks.uiuc.edu
> > Info) Please include this reference in published work using VMD:
> > Info) Humphrey, W., Dalke, A. and Schulten, K., `VMD - Visual
> > Info) Molecular Dynamics', J. Molec. Graphics 1996, 14.1, 33-38.
> > Info) -------------------------------------------------------------
> > Info) Multithreading available, 72 CPUs detected.
> > Info) CPU features: SSE2 AVX AVX2 FMA AVX512F AVX512CD
> > Info) Free system memory: 956GB (94%)
> > Info) Creating CUDA device pool and initializing hardware...
> > Info) Detected 2 available CUDA accelerators:
> > Info) [0] TITAN RTX 72 SM_7.5 1.77 GHz, 24GB RAM, AE3, ZCP
> > Info) [1] Quadro P6000 30 SM_6.1 1.64 GHz, 24GB RAM, KTO, AE2,
> > ZCP
> > Warning) Detected X11 'Composite' extension: if incorrect display
> > occurs
> > Warning) try disabling this X server option. Most OpenGL drivers
> > Warning) disable stereoscopic display when 'Composite' is enabled.
> > Info) OpenGL renderer: Quadro P6000/PCIe/SSE2
> > Info) Features: STENCIL MSAA(4) MDE CVA MTX NPOT PP PS GLSL(OVFGS)
> > Info) Full GLSL rendering mode is available.
> > Info) Textures: 2-D (32768x32768), 3-D (16384x16384x16384),
> > Multitexture (4)
> > Info) Using OptiX device mask '0x1'
> > Info) Detected 1 available TachyonL/OptiX ray tracing accelerator
> > Info) Compiling 1 OptiX shaders on 1 target GPU...
> > Info) Using OptiX device mask '0x1'
> > Info) Using OptiX device mask '0x1'
> >
> > Could it be a problem of compatibility of the TITAN RTX, or of the
> > driver version?
> > Thank you for your help,
> >
> > Best regards,
> > Lorenzo
> >
> > --
> > NIH Center for Macromolecular Modeling and Bioinformatics
> > Beckman Institute for Advanced Science and Technology
> > University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
> > [7]https://urldefense.com/v3/__http://www.ks.uiuc.edu/*johns/__;fg!!Mih3wA!Wc_2cph4f8JXiqRM6sFgFd7wqedSGncDH3YaNIdxJk96V7uocI1wGR6SeIwfIk11$ Phone:
> > 217-244-3349
> > [8]https://urldefense.com/v3/__http://www.ks.uiuc.edu/Research/vmd/__;!!Mih3wA!Wc_2cph4f8JXiqRM6sFgFd7wqedSGncDH3YaNIdxJk96V7uocI1wGR6SeGBFTfg6$
> >
> > References
> >
> > Visible links
> > 1. https://amarolab.ucsd.edu/
> > 2. mailto:johns_at_ks.uiuc.edu
> > 3. https://urldefense.com/v3/__http://www.ks.uiuc.edu/Research/vmd/__;!!Mih3wA!Wc_2cph4f8JXiqRM6sFgFd7wqedSGncDH3YaNIdxJk96V7uocI1wGR6SeGBFTfg6$
> > 4. mailto:vmd_at_ks.uiuc.edu
> > 5. https://urldefense.com/v3/__http://www.ks.uiuc.edu/Research/vmd/__;!!Mih3wA!Wc_2cph4f8JXiqRM6sFgFd7wqedSGncDH3YaNIdxJk96V7uocI1wGR6SeGBFTfg6$
> > 6. mailto:vmd_at_ks.uiuc.edu
> > 7. https://urldefense.com/v3/__http://www.ks.uiuc.edu/*johns/__;fg!!Mih3wA!Wc_2cph4f8JXiqRM6sFgFd7wqedSGncDH3YaNIdxJk96V7uocI1wGR6SeIwfIk11$
> > 8. https://urldefense.com/v3/__http://www.ks.uiuc.edu/Research/vmd/__;!!Mih3wA!Wc_2cph4f8JXiqRM6sFgFd7wqedSGncDH3YaNIdxJk96V7uocI1wGR6SeGBFTfg6$
>
> --
> NIH Center for Macromolecular Modeling and Bioinformatics
> Beckman Institute for Advanced Science and Technology
> University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
> http://www.ks.uiuc.edu/~johns/ Phone: 217-244-3349
> http://www.ks.uiuc.edu/Research/vmd/ 

-- 
NIH Center for Macromolecular Modeling and Bioinformatics
Beckman Institute for Advanced Science and Technology
University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
http://www.ks.uiuc.edu/~johns/           Phone: 217-244-3349
http://www.ks.uiuc.edu/Research/vmd/