Starting simulation: clEnqueueNDRangeKernel (-4)

[stefan.schmelz@…]

The following bug report has been submitted:
Platform:        Darwin-17.7.0-x86_64-i386-64bit
ChimeraX Version: 0.92 (2020-03-08)
Description
Happend during the crystallographic model building tutorial when starting the simulation "play" button within Isolde.

Log:
Startup Messages  
---  
warning | 'clip' is a prefix of an existing command 'clipper'  
  
UCSF ChimeraX version: 0.92 (2020-03-08)  
© 2016-2020 Regents of the University of California. All rights reserved.  
How to cite UCSF ChimeraX  

> toolshed show ISOLDE

> set selectionWidth 4

Done loading forcefield  

> open 1a0m structureFactors true

Summary of feedback from opening 1a0m fetched from pdb  
---  
warning | No free flags detected in this dataset! Automatically generated a
new random set with 459 free from 9264 observed reflections. You should save
your data to a new MTZ file and use this for any future rebuilding/refinement.  
notes | Resolution: 1.0944213038232693  
  
1a0m title:  
1.1 angstrom crystal structure of A-conotoxin [TYR15]-epi [more info...]  
  
Chain information for 1a0m  
---  
Chain | Description  
1.1/A 1.1/B | α-conotoxin [TYR15]-epi  
  
Non-standard residues in 1a0m #1.1  
---  
NH2 — amino group  
  
  

> save /Users/ssc/Desktop/1a0m.mtz #1

Writing data to: '/*/*/[FOBS, SIGFOBS]'  

> set bgColor white

> clipper isolate #1/A&protein

> isolde start

> clipper spotlight

> rama showFavored false

Expected a structures specifier or a keyword  

> rama showFavored true

> isolde restrain ligands #1

> addh

Summary of feedback from adding hydrogens to 1a0m #1.1  
---  
warning | Not adding hydrogens to /A ARG 7 CB because it is missing heavy-atom
bond partners  
notes | Termini for 1a0m (#1.1) chain A determined from SEQRES records  
Termini for 1a0m (#1.1) chain B determined from SEQRES records  
Chain-initial residues that are actual N termini: /A GLY 1, /B GLY 1  
Chain-initial residues that are not actual N termini:  
Chain-final residues that are actual C termini: /A NH2 17, /B NH2 17  
Chain-final residues that are not actual C termini:  
107 hydrogen bonds  
271 hydrogens added  
  

> hide HC

> select #1

547 atoms, 513 bonds, 26 models selected  
Traceback (most recent call last):  
File
"/Applications/ChimeraX-0.92.app/Contents/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-
packages/chimerax/core/triggerset.py", line 130, in invoke  
return self._func(self._name, data)  
File "/Users/ssc/Library/Application Support/ChimeraX/0.92/site-
packages/chimerax/isolde/delayed_reaction.py", line 65, in callback  
self.ff(*self.ff_args)  
File "/Users/ssc/Library/Application Support/ChimeraX/0.92/site-
packages/chimerax/isolde/openmm/openmm_interface.py", line 1641, in
_update_coordinates_and_repeat  
self.atoms.coords = th.coords  
File "/Users/ssc/Library/Application Support/ChimeraX/0.92/site-
packages/chimerax/isolde/openmm/openmm_interface.py", line 301, in coords  
f(self._c_pointer, n, pointer(coords))  
RuntimeError: Error invoking kernel computeBondedForces:
clEnqueueNDRangeKernel (-4)  
  
Error processing trigger "new frame":  
RuntimeError: Error invoking kernel computeBondedForces:
clEnqueueNDRangeKernel (-4)  
  
File "/Users/ssc/Library/Application Support/ChimeraX/0.92/site-
packages/chimerax/isolde/openmm/openmm_interface.py", line 301, in coords  
f(self._c_pointer, n, pointer(coords))  
  
See log for complete Python traceback.  
  




OpenGL version: 4.1 ATI-1.68.25
OpenGL renderer: AMD Radeon HD 6750M OpenGL Engine
OpenGL vendor: ATI Technologies Inc.

[Tristan Croll]

I'm afraid the GPU you're using isn't up to the task, and is running out 
of memory (OpenCL error -4 is CL_MEM_OBJECT _ALLOCATION_FAILURE - see 
https://streamhpc.com/blog/2013-04-28/opencl-error-codes/). ISOLDE 
really requires a fairly recent, at least moderately high-end GPU to run 
its simulations on. The 6750M is a 9-year-old GPU with only 1GB of 
on-board memory - quite a bit of which will already be taken up by the 
scene visualisation. ISOLDE typically needs between a few hundred MB and 
2GB of free memory on the GPU, depending on simulation size.

On 2020-09-09 17:00, ChimeraX wrote:

[Tristan Croll]

That being said... I see you're using a Mac. You should be able to force ISOLDE to use the on-board Intel GPU (slow, but gets to share the system memory so is fairly bulletproof) using the isolde set gpuDeviceIndex command. Try isolde set gpuDeviceIndex 0 then run a simulation. If that doesn't work, try isolde set gpuDeviceIndex 1.