Table 1.
Method | Time per iteration (min : s) | Memory per node |
---|---|---|
250 000 variables | ||
glmnet | 2 : 20 | 46 GB |
gpu-lasso (1 CPU) | 0 : 54 | 415 MB |
gpu-lasso (1 GPU) | 0 : 1.85 | 415 MB |
gpu-lasso (2 GPU) | 0 : 0.93 | 208 MB |
1 million variables | ||
glmnet | NA | NA |
gpu-lasso (1 CPU) | 3 : 47 | 1.7 GB |
gpu-lasso (1 GPU) | 0 : 7.7 | 1.7 GB |
gpu-lasso (2 GPU) | 0 : 3.8 | 863 MB |
A total of 7000 subjects in all analyses.