atom-ai / docs /artifacts /batched_training /bench_gpu_quickpod_rtx5090.json
thefinalboss's picture
ATOM (atom-ai): code snapshot (100fae0 + batched-train), README with measured numbers, curated checkpoints
9f1076a verified
Raw History Blame Contribute Delete
1.77 kB
[
{
"B": 64,
"legacy": false,
"device": "cuda",
"threads": 4,
"K": 128,
"d_model": 32,
"ticks": 256,
"repeat": 3,
"seconds_all": [
2.8985033044591546,
2.942989300005138,
2.934748121537268
],
"ticks_per_s": 88.32144493544686,
"bytes_per_s": 5652.572475868599,
"bytes_per_s_median": 5582.761900335691,
"cpu_seconds_all": [
2.897691068000001,
2.9425071880000004,
2.935314998000001
],
"bytes_per_cpu_s": 5654.1569185663075,
"peak_rss_mb": 1899.515625,
"cuda_peak_mb": 917.92578125
},
{
"B": 256,
"legacy": false,
"device": "cuda",
"threads": 4,
"K": 128,
"d_model": 32,
"ticks": 256,
"repeat": 3,
"seconds_all": [
2.93217777274549,
2.9207353172823787,
2.935780931264162
],
"ticks_per_s": 87.64916097847484,
"bytes_per_s": 22438.18521048956,
"bytes_per_s_median": 22350.62301104499,
"cpu_seconds_all": [
2.9317100380000003,
2.9199013679999997,
2.934995358
],
"bytes_per_cpu_s": 22444.593751770866,
"peak_rss_mb": 1902.234375,
"cuda_peak_mb": 3603.93798828125
},
{
"B": 1024,
"legacy": false,
"device": "cuda",
"threads": 4,
"K": 128,
"d_model": 32,
"ticks": 256,
"repeat": 3,
"seconds_all": [
3.0996312815696,
3.1405691895633936,
3.052405214868486
],
"ticks_per_s": 83.86828811358517,
"bytes_per_s": 85881.12702831121,
"bytes_per_s_median": 84572.63983581131,
"cpu_seconds_all": [
3.098436908,
3.139777196999999,
3.0514482990000005
],
"bytes_per_cpu_s": 85908.05883419621,
"peak_rss_mb": 1912.19921875,
"cuda_peak_mb": 14382.72119140625
}
]