These variables are most useful when you run several RealityScan instances on one machine. Each instance can be limited to part of the available CPU, RAM, GPUs, or VRAM, so the instances don't use up all of the machine's resources.
Quick Reference
OMP_NUM_THREADS | CPU threads | One per logical CPU |
CR_RAM_LIMIT | System RAM | No limit |
CR_VRAM_LIMIT | GPU memory (all GPUs) | No limit |
CR_ENABLE_CUDA | CUDA support on/off | 1 (enabled) |
CR_ENABLE_HIP | HIP (AMD) support on/off | 1 (enabled) |
CR_CUDA_DEVICES | Which CUDA GPUs are used | All CUDA devices |
CR_HIP_DEVICES | Which HIP GPUs are used | All HIP devices |
CPU
OMP_NUM_THREADS
Limits the number of threads that parallel regions inside RealityScan can use, which reduces CPU load.
Default: one thread per logical CPU.
When to use: when several instances run on the same machine, give each one a share of the cores. For example, on a 32-thread machine running two instances, set OMP_NUM_THREADS=16 for each.
Memory
CR_RAM_LIMIT
Limits how much system RAM RealityScan treats as available when sizing its internal buffers and caches.
Format: an integer, optionally followed by a binary-unit suffix K, M, G, T, P, or E (powers of 1024) and an optional trailing B. Suffixes are case-insensitive. Examples: 16G, 8192M, 64GB.
The value is capped at the physical RAM installed, so a higher value has no effect.
Default: no limit. RealityScan uses all physical RAM.
When to use: when the machine is shared with other memory-intensive processes, including other RealityScan instances.
CR_VRAM_LIMIT
Limits the total GPU memory RealityScan can allocate across all the GPUs it uses.
Format: same as CR_RAM_LIMIT: an integer with an optional K, M, G, T, P, or E suffix and an optional trailing B, case-insensitive.
Default: no limit.
When to use: when two or more processes share the same GPU, so that no single process claims all the VRAM and starves the others.
GPU Selection
CR_ENABLE_CUDA
Turns CUDA support on or off.
Values: 1 (enabled) or 0 (disabled). Default: 1.
When set to 0, RealityScan reports zero CUDA devices whatever CUDA hardware is installed, and CR_CUDA_DEVICES is ignored.
When to use: to keep all CUDA-capable GPUs free for another process.
CR_ENABLE_HIP
Turns HIP (AMD GPU) support on or off.
Values: 1 (enabled) or 0 (disabled). Default: 1.
When set to 0, RealityScan reports zero HIP devices whatever HIP hardware is installed, and CR_HIP_DEVICES is ignored.
When to use: to keep all HIP-capable GPUs free for another process.
CR_CUDA_DEVICES
Restricts which CUDA device indices RealityScan can use.
Format: a comma-separated list of single indices and inclusive ranges, for example 0, 0,2, 1-2, or 4-7. Indices follow CUDA's own device numbering, the same numbering that nvidia-smi reports.
Default: all CUDA devices are available.
Has no effect when CR_ENABLE_CUDA=0, which overrides it.
When to use: on multi-GPU machines, to assign different cards to different RealityScan instances.
CR_HIP_DEVICES
The HIP counterpart of CR_CUDA_DEVICES. Restricts which HIP device indices RealityScan can use.
Format: same as CR_CUDA_DEVICES: comma-separated indices and inclusive ranges.
Default: all HIP devices are available.
Has no effect when CR_ENABLE_HIP=0, which overrides it.