bench: fresh self-play parallel grid on Space
Browse filesOdd result: 8w x 2t best at 2513 moves/s, throughput FALLS with more workers
(unlike the earlier 32x2 = 2258 peak with games=128 too). Likely noisy
neighbour / shared-node variance on the Space; 8-32 workers all in the
2000-2500 range
runs/bench_parallel_2026_11_17.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"cpu_count": 192,
|
| 3 |
+
"games": 128,
|
| 4 |
+
"runs": [
|
| 5 |
+
{"workers": 8, "threads": 2, "sec": 4.02, "moves_per_s": 2513.4, "games": 128},
|
| 6 |
+
{"workers": 16, "threads": 2, "sec": 4.99, "moves_per_s": 2035.7, "games": 128},
|
| 7 |
+
{"workers": 32, "threads": 2, "sec": 8.26, "moves_per_s": 1229.8, "games": 128},
|
| 8 |
+
{"workers": 64, "threads": 2, "sec": 16.21, "moves_per_s": 624.7, "games": 128},
|
| 9 |
+
{"workers": 96, "threads": 2, "sec": 29.31, "moves_per_s": 349.4, "games": 128}
|
| 10 |
+
],
|
| 11 |
+
"best": {"workers": 8, "threads": 2, "sec": 4.02, "moves_per_s": 2513.4, "games": 128}
|
| 12 |
+
}
|