Trying to use w/ Ladybug+Honeybee

[Joe Del Rocco]

Microsoft Windows 10 Enterprise - 10.0.14393 Build 14393

nVidia TITAN Xp

Rhino 6

Ladybug 0.0.66

Honeybee 0.0.63

Radiance 5.1.0

I installed Accelerad. Tested demos to make sure they work. They do.

I stomped the original Radiance files w/ Accerlerad versions (for now) and copied over the lib contents.

Trying to run an Image Based Simulation in Ladybug+Honeybee.

Get the attached errors output from Honeybee Run Daylight Simulation.

I'm not an expert with Radiance or Ladybug+Honeybee. But I have been able to use them successfully with basic simulations and parameters. I just wanted to try Accelerad to take advantage of my GPU. It could be that I just don't have it using the correct parameters.

Any other files/info I can provide, just ask!

[Mostapha Sadeghipour]

 I think we had a similar discussion posted a couple of months ago. It's most likely because how Honeybee parses the log from rpict.

We need some help from Nathaniel to change how Accelerad logs rpict progress. Once that's done we can go ahead and fix this on our side. Nathaniel can you confirm if the changes has been made?

--Mostapha

[Nathaniel Jones]

I agree with Mostapha's assessment. There are no Accelerad errors in the log, so the issue is probably with the way Honeybee parses the output.

Joe, you can work around this for now by providing the -w parameter to turn off warnings. I would also recommend setting the number of CPUs to 1. Multiple Accelerad instances have to take turns using the GPU, so they mostly just tie up the processors while they wait.

Mostapha, the fix is already in the source code in GitHub, but it's not in the latest compiled executable. Starting with version 0.7, all non-empty output lines will be prepended with the program name and a colon, for instance "accelerad_rpict: <message>" or if the user has changed the name as Joe did, then just "rpict: <message>". Technically, if the user changed the name to foobar.exe, then the each line would begin "foobar: <message>". The only exception is if the string <message> happens to contain a newline character, in which case only the first line of the string gets prepended with the program name. I think the code is generally good at not using multi-line warnings, though.

Nathaniel

[Joe Del Rocco]

So, -w did get me further. It did run and it did improve speed as expected. Great!

But now Accelerad crashes my machine hard with my "production" parameters.

Would you have any idea of which parameter might be the problem? Is there a log I can look at to see what might be causing the hard crash?

These "draft" parameters worked:

These "production" parameters crashed hard:

[Joe Del Rocco]

I'll try the TDR regedits and let you know if that's it.

I didn't try any of that at first because the docs said it was mainly for "older" graphics cards, and I have a nVidia Titan Xp.

[Joe Del Rocco]

Also not sure if any of this is useful, but this is the output of CUDA deviceQuery (9.0):

Device 0: "TITAN Xp"
  CUDA Driver Version / Runtime Version          9.0 / 9.0
  CUDA Capability Major/Minor version number:    6.1
  Total amount of global memory:                 12288 MBytes (12884901888 bytes)
  (30) Multiprocessors, (128) CUDA Cores/MP:     3840 CUDA Cores
  GPU Max Clock rate:                            1582 MHz (1.58 GHz)
  Memory Clock rate:                             5705 Mhz
  Memory Bus Width:                              384-bit
  L2 Cache Size:                                 3145728 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers  1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(32768, 32768), 2048 layers
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  2048
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and kernel execution:          Yes with 2 copy engine(s)
  Run time limit on kernels:                     No
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support:                        Disabled
  CUDA Device Driver Mode (TCC or WDDM):         WDDM (Windows Display Driver Model)
  Device supports Unified Addressing (UVA):      Yes
  Supports Cooperative Kernel Launch:            No
  Supports MultiDevice Co-op Kernel Launch:      No
  Device PCI Domain ID / Bus ID / location ID:   0 / 1 / 0
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 9.0, CUDA Runtime Version = 9.0, NumDevs = 1
Result = PASS

[Joe Del Rocco]

So I'm following this: https://nljones.github.io/Accelerad/documentation.html#drivercrash

Since I only have 1 CUDA device, I don't think I can do anything with CUDA_VISIBLE_DEVICES.

These are the extra Radiance params I passed. Not sure if that's passed correctly or not:

But I still get a hard crash on my card :/

If anyone has any suggestions on which of these registry keys I should add (just for temporary use for now), please feel free to share:

https://docs.microsoft.com/en-us/windows-hardware/drivers/display/tdr-registry-keys

I know this project was more of a proof of concept and not production, so I'm not complaining here. Was just trying to get some use out of my GPU. But it may not be worth it without a Tesla card specifically :/

[Nathaniel Jones]

From your description, it does sound like a TDR error. The Honeybee output log should contain all errors reported, but of course TDR errors are generated by Windows, not by Accelerad, so all the log will show is an unknown error number.

If you feel comfortable editing your registry, the necessary keys are TdrLevel and TdrDelay, described on Microsoft's site. On older Windows versions, these keys come predefined, but on new versions of Windows 10, you need to create them yourself.

Nathaniel