Add long benchmark at optimal dispatch size.

diff --git a/benchmarks.md b/benchmarks.md
index 8dd42d0911fe6a6566a91a3e4bc1b78e5c30bc3b..7347fde0bb4e6321335f31d8c8bf69ce2fb5dd18 100644 (file)
--- a/benchmarks.md
+++ b/benchmarks.md
@@ -237,6 +237,19 @@ NanoPow (WebGPU) 3070 (dispatch 0x1000 workgroup 8,8)
        "geometric": 192.06862657670237
 }
 
+NanoPow (WebGPU) 3070 (dispatch 0x800 workgroup 8,8)
+{
+       "count": 4096,
+       "total": 1061022.09999986,
+       "rate": 4.8102132476979635,
+       "min": 5.5,
+       "max": 2067.7000000029802,
+       "arithmetic": 259.0385986327783,
+       "truncated": 207.8909912109557,
+       "harmonic": 75.27207858671284,
+       "geometric": 156.19893668278434
+}
+
 NanoPow (WebGPU) 3070 (dispatch 0x800 workgroup 8,8)
 {
        "count": 512,