GPU issues and limitations
Known issues or limitations that you might encounter in specific system environments, or configurations.
There are a small number of limitations to be aware of with the graphics processing unit (GPU).
- Reversed arrays
- The sort operation on a reversed array, for example
[5,4,3,2,1], is slower on a GPU than the CPU. - Segmentation violation
- Segmentation violations are seen on some occasions. To avoid these errors, set a native stack size of 3 MB by specifying the option -Xmso3m.
- Initial sort operation slow
- Tests indicate that the first sort operation performed on the GPU can be slower than subsequent sort operations.
- Intermittent crash in CUDA4J module
- Under a heavy workload, an intermittent crash is observed in the CUDA4J
module, which results in a segmentation error with
vmState=0x00000000. The root cause of this issue is under investigation. - Running CUDA4J and sort operations simultaneously
- In a multi-threaded environment, running CUDA4J and GPU sort operations simultaneously can cause an application to hang. This issue is under investigation.