mixbench-cuda hangs in the middle of test

[Sfinx]

Stock Ubuntu 20.04, CUDA 11.2 :

root@gpu:~/mixbench/build-cuda# ./mixbench-cuda-alt 
mixbench/alternating (v0.03-15-g514c757)
------------------------ Device specifications ------------------------
Device:              GeForce RTX 2080 Ti
CUDA driver version: 11.20
GPU clock rate:      1545 MHz
Memory clock rate:   3500 MHz
Memory bus width:    352 bits
WarpSize:            32
L2 cache size:       5632 KB
Total global mem:    11016 MB
ECC enabled:         No
Compute Capability:  7.5
Total SPs:           4352 (68 MPs x 64 SPs/MP)
Compute throughput:  13447.68 GFlops (theoretical single precision FMAs)
Memory bandwidth:    616.00 GB/sec
-----------------------------------------------------------------------
Total GPU memory 11551440896, free 11312300032
Buffer size:          64MB
Trade-off type:       compute with global memory (block strided)
----------------------------------------------------------------------------- CSV data -----------------------------------------------------------------------------
Experiment ID, Single Precision ops,,,,              Double precision ops,,,,              Half precision ops,,,,                Integer operations,,, 
Compute iters, Flops/byte, ex.time,  GFLOPS, GB/sec, Flops/byte, ex.time,  GFLOPS, GB/sec, Flops/byte, ex.time,  GFLOPS, GB/sec, Iops/byte, ex.time,   GIOPS, GB/sec
            0,      0.000,   12.36,    0.00, 694.90,      0.000,   30.69,    0.00, 559.76,      0.000,   12.78,    0.00, 672.24,     0.000,   12.59,    0.00, 682.44
            1,      0.129,   12.09,   88.80, 688.21,      0.065,   30.54,   35.15, 544.88,      0.258,   12.20,  176.05, 682.20,     0.129,   12.00,   89.49, 693.57
            2,      0.267,   11.68,  183.83, 689.36,      0.133,   29.53,   72.73, 545.51,      0.533,   11.62,  369.64, 693.08,     0.267,   11.68,  183.81, 689.29

mixbench-opencl runs normally

[Sfinx]

After some time the following error appears:

Cuda error in file '/root/mixbench/mixbench-cuda/mix_kernels_cuda.cu' in line 153 : device not ready.

[ekondis]

Could you set explicitly CC7.5 at CMakeLists.txt line 10 and try building and running again? Does this issue still occurs?

[Sfinx]

No luck - tried using GCC-7.5 and setting "-gencode arch=compute_75,code=[sm_75,compute_75]" too as GTX 2080 is Turing arch.

[levitation]

I had a similar issue with mixbench-cuda hanging. It was running at 100% GPU load, but not progressing.

I commented out the following lines in mix_kernels_cuda.cu, then the code ran ok.

//runbench<5>(cd, size, doHalfs);
//runbench<6>(cd, size, doHalfs);
//runbench<7>(cd, size, doHalfs);

//runbench<192>(cd, size, doHalfs);